Policy Assessment for the Reconsideration of the National Ambient Air Quality Standards for Particulate Matter


* O \

KWJ

*1 PRO^^

Policy Assessment for the Reconsideration of
the National Ambient Air Quality Standards for
Particulate Matter

-------

-------
EPA-452/R-22-004
May 2022

Policy Assessment for the Reconsideration of the National Ambient Air Quality Standards for

Particulate Matter

U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Health and Environmental Impacts Division
Research Triangle Park, NC

-------
DISCLAIMER

This Policy Assessment has been prepared by staff in the U.S. Environmental Protection
Agency's (EPA) Office of Air Quality Planning and Standards. Any findings and conclusions are
those of the authors and do not necessarily reflect the views of the EPA. Questions or comments
related to this document should be addressed to Dr. Lars Perlmutt, U.S. Environmental
Protection Agency, Office of Air Quality Planning and Standards, C539-06, Research Triangle
Park, North Carolina 27711 (email: perlmutt.lars@epa.gov).

-------
TABLE OF CONTENTS

1	INTRODUCTION	1-1

1.1	Purpose 	1-2

1.2	Legislative Requirements	1-3

1.3	History of Reviews of the PM NAAQS	1-6

1.3.1	Reviews Completed in 1971 and 1987 	1-8

1.3.2	Review Completed in 1997	1-8

1.3.3	Review Completed in 2006	1-10

1.3.4	Review Completed in 2012	1-12

1.3.5	Review Completed in 2020	1-12

1.4	Reconsideration of the 2020 PM NAAQS Final Action	1-14

1.4.1	Decision to Initiate a Reconsideration	1-14

1.4.2	Process for Reconsideration of the 2020 PM NAAQS Decision	1-15

1.4.3	Ongoing Litigation	1-18

References		1-19

2	PM AIR QUALITY	2-1

2.1	Distribution of Particle Size in Ambient Air	2-1

2.1.1 Sources of PM Emissions	2-3

2.2	Ambient PM Monitoring Methods and Networks	2-14

2.2.1	Total Suspended Particulates (TSP) Sampling	2-15

2.2.2	PMio Monitoring	2-16

2.2.3	PM2.5 Monitoring	2-17

2.2.4	PM10-2.5 Monitoring	2-22

2.2.5	Additional PM Measurements and Metrics	2-23

2.3	Ambient Air Concentrations	2-25

2.3.1	Trends in Emissions of PM and Precursor Gases	2-25

2.3.2	Trends in Monitored Ambient Concentrations	2-26

2.3.3	Predicted Ambient PM2.5 Based on Hybrid Modeling Approaches	2-43

2.4	Background PM	2-53

2.4.1	Natural Sources	2-55

2.4.2	International Transport	2-57

2.4.3	Estimating Background PM with Recent Data	2-59

1

-------
References		2-62

3	RECONSIDERATION OF THE PRIMARY STANDARDS FOR PM2.5	3-1

3.1	Background on the Current Standards	3-2

3.1.1 Considerations Regarding the Adequacy of the Existing Standards in the 2020
Review	3-4

3.2	General Approach and Key Issues in this Reconsideration of the 2020 Final
Decision 	3-12

3.3	Health Effects Evidence	3-15

3.3.1	Nature of Effects	3-18

3.3.2	Public Health Implications and At-Risk Populations	3-52

3.3.3	PM2.5 Concentrations in Key Studies Reporting Health Effects	3-59

3.3.4	Uncertainties in the Health Effects Evidence	3-133

3.4	Risk Information	3-136

3.4.1	Risk Assessment Overview	3-137

3.4.2	Results of the Risk Assessment	3-146

3.4.3	Conclusions of the Risk Assessment	3-167

3.5	CAS AC Advice and Public Comments	3-168

3.6	Key Considerations Regarding the Adequacy of the primary PM2.5 Standards. 3-171

3.6.1	Evidence-based Considerations	3-172

3.6.2	Risk-based Considerations	3-188

3.6.3	Conclusions on the Adequacy of the Current Primary PM2.5 Standards	3-198

3.7	Areas for Future Research and Data Collection	3-223

References		3-227

4	RECONSIDERATION OF THE PRIMARY STANDARD FOR PM10	4-1

4.1	Background on the Current Standard	4-2

4.1.1 Considerations Regarding the Adequacy of the Existing Standards in the 2020
Review	4-4

4.2	General Approach and Key Issues in this Reconsideration of the 2020 Final
Decision 	4-5

4.3	Health Effects Evidence	4-8

4.3.1 Nature of Effects	4-9

4.4	CAS AC Advice and Public Comments	4-16

4.5	Conclusions on the Adequacy of the Current Primary PM10 Standard	4-17

4.6	Areas for Future Research and Data Collection	4-19

11

-------
References		4-21

5 RECONSIDERATION OF THE SECONDARY STANDARDS FOR PM	5-1

5.1	Background on the Current Standards	5-2

5.1.1	Non-Visibility Effects	5-2

5.1.2	Visibility Effects	5-6

5.2	General Approach and Key Issues in this Reconsideration of the 2020 Final
Decision 	5-11

5.3	Welfare Effects and Quantitative Information	5-14

5.3.1	Visibility Effects	5-16

5.3.2	Non-Visibility Effects	5-36

5.4	CAS AC Advice and Public Comments	5-48

5.5	Conclusions Regarding the Adequacy of the Secondary PM Standards	5-49

5.6	Areas for Future Research and Data Collection	5-53

References		5-56

iii

-------
Appendix A.
Appendix B.
Appendix C.
Appendix D.

LIST OF APPENDICES

Supplemental Information on PM Air Quality Analyses

Supplemental Study Information: Selection Criteria, Study Methods and Details
Supplemental Information Related to the Human Health Risk Assessment
Quantitative Analyses for Visibility Impairment

iv

-------
LIST OF TABLES

Table 1-1. Summary of NAAQS promulgated for particulate matter 1971-2012 1-7

Table 2-1. PM Monitoring Networks 2-15

Table 2-2. Percent Changes in PM and PM precursor emissions in the NEI for the time periods

1990-2017 and 2002-2017 2-28

Table 2-3. Nationwide averages of ratios of maximum annual PM2.5 design values to average

composite monitor PM2.5 concentrations across CBSAs 2-47

Table 2-4. Mean 2011 PM2.5 concentration by region for predictions in Figure 2-29 2-54

Table 2-5. Average Annual PM2.5 Concentration (|ig/m3) by Year 2-60

Table 2-6. Three-Year Average of the Average Annual PM2.5 Concentrations (|ig/m3) 2-61

Table 2-7. Average Annual PM2.5 Concentrations and Ratios to Regulatory Design Values. ..2-62

Table 3-1. Key causality determinations for PM2.5 and UFP exposures 3-20

Table 3-2. National demographic information, 2019 3-56

Table 3-3. Prevalence of cardiovascular and respiratory diseases among adults by age, race, and

ethnicity in the U.S. in 2018

3-58

Table 3-4.

Summary of information from PM2.5 controlled human exposure studies

3-61

Table 3-5.

PM2.5 Concentrations Metrics from Monitor and Modeled Data

3-73

Table 3-6.

Key U.S. Epidemiologic Studies: Monitor-Based Exposure

3-92

Table 3-7.

Key Canadian Epidemiologic Studies: Monitor-Based Exposure

3-96

Table 3-8.

Key U.S. Epidemiologic Studies: Model-Based Exposure

3-99

Table 3-9.

Key Canadian Epidemiologic Studies: Model-Based Exposure

3-104

Table 3-10. Epidemiologic studies examining the health impacts associated with ambient PM2.5

concentrations when studies are conducted with restricted air quality exposures

3-121

Table 3-11. Summary of information from studies that use alternative methods for confounder

control 3-125

Table 3-12. Epidemiologic studies examining the health impacts of long-term reductions in

ambient PM2.5 concentrations 3-130

Table 3-13. Epidemiologic studies used to estimate PM2.5-associated risk 3-138

Table 3-14. Estimates of PM2.5-associated mortality for air quality adjusted to just meet the

current or alternative standards (47 urban study areas) 3-149

Table 3-15. Estimated reduction in PM2.5-associated mortality for alternative annual and 24-hour
standards (47 urban study areas) 3-150

-------
Table 3-16. Estimates of PIVfo.s-associated mortality for the current and potential alternative

annual standards in the 30 study areas where the annual standard is controlling	

	3-152

Table 3-17. Estimated delta and percent reduction in PIvfc.s-associated mortality for the current
and potential alternative annual standards in the 30 study areas where the annual
standard is controlling	3-153

Table 3-18. Estimates of PIVfo.s-associated mortality for the current 24-hour standard, and an

alternative, in the 11 study areas where the 24-hour standard is controlling	3-156

Table 3-19. Average national percent PM2.5 reduction in demographic populations aged 65 and
over residing in the full set of 47 study areas and subset of 30 study areas
controlled by the annual standard	3-161

Table 3-20. Average national percent PM2.5 risk reduction in demographic populations aged 65
and over residing in the full set of 47 study areas and subset of 30 study areas
controlled by the annual standard	3-161

Table 4-1. Key Causality Determinations for PM10-2.5 Exposures	4-9

Table 5-1. Key causality determinations for PM-related welfare effects	5-15

vi

-------
LIST OF FIGURES

Figure 2-1. Comparisons of PM2.5 and PM10 diameters to human hair and beach sand 2-2

Figure 2-2. Percent contribution of PM2.5 national emissions by source sectors 2-5

Figure 2-3. 2017 NEI PM2.5 Emissions Density Map, tons per square mile 2-6

Figure 2-4. Percent contribution of PM10 emissions by national source sectors 2-7

Figure 2-5. 2017 NEI PM10 Emissions Density Map, tons per square mile 2-7

Figure 2-6. Percent contribution to organic carbon (top panel) and elemental carbon (bottom

panel) national emissions by source sectors 2-9

Figure 2-7. 2017 NEI Elemental Carbon Emissions Density Map, tons per square mile 2-10

Figure 2-8. Percent contribution to sulfur dioxide (panel A), oxides of nitrogen (panel B),

ammonia (panel C), and volatile organic compounds (panel D) national emissions
by source sectors 2-11

Figure 2-9. SO2 Emissions Density Map, tons per square mile 2-12

Figure 2-10. NOx Emissions Density Map, tons per square mile 2-12

Figure 2-11. NFb Emissions Density Map, tons per square mile 2-12

Figure 2-12. Anthropogenic (including wildfires) VOC Emissions Density Map, tons per square
mile. 2-13

Figure 2-13. PM Monitoring stations reporting to EPA's AQS database by PM size fraction,

1970-2020 2-15

Figure 2-14. National emission trends of PM2.5, PM10, and precursor gases from 1990 to 2017

2-27

Figure 2-15. Annual average and 98th percentile of 24-hour PM2.5 concentrations (in |ag/m3) from
2017-2019 (top) and linear trends and their associated significance (based on p-
values) in PM2.5 concentrations from 2000-2019 (bottom) 2-29

Figure 2-16. Seasonally-weighted annual average PM2.5 concentrations in the U.S. from 2000 to
2019 (406 sites) 2-30

Figure 2-17. Pearson's correlation coefficient between annual average and 98th percentile of 24-
hour PM2.5 concentrations from 2000-2019 2-31

Figure 2-18. Scatterplot of CBSA maximum annual versus CBSA maximum daily design values
(2017-2019) with the solid black line representing the ratio of daily and annual
NAAQS values 2-32

Figure 2-19. Frequency distribution of 2017-2019 2-hour averages for sites meeting both or

violating either PM2.5 NAAQS for October to March (blue) and April to September
(red) 2-35

Figure 2-20. Annual average PM2.5 sulfate, nitrate, organic carbon, and elemental carbon

concentrations (in |ig/m3) from 2017-2019 2-36

vii

-------
Figure 2-21. Annual average and 2nd highest PMio concentrations (in |ag/m3) from 2017-2019
(top) and linear trends and their associated significance in PMio concentrations
from 2000-2019 (bottom) 2-38

Figure 2-22. National trends in Annual 2nd Highest 24-Hour PMio concentrations from 2000 to

2019 (262 sites) 2-39

Figure 2-23. Annual average PM2.5/PM10 ratio for 2017-2019 2-40

Figure 2-24. PM2.5/PM10 ratio on days ranking as the second highest yearly PMio concentration
and among the top four highest yearly PM2.5 concentrations for 2017-2019 2-40

Figure 2-25. Annual average and 98th percentile PM10-2.5 concentrations (|ag/m3) from 2017-2019
(top) and linear trends and their associated significance in PM10-2.5 concentrations
from 2000-2019 (bottom) 2-42

Figure 2-26. Average hourly particle number concentrations from three locations in the State of
New York for 2014 to 2015 (green is Steuben County, orange is Buffalo, red is
New York City) 2-43

Figure 2-27. Time series of annual average mass and number concentrations (left) and scatterplot
of mass vs. number concentration (right) between 2000-2019 in Bondville, IL. 2-44

Figure 2-28. Comparison of CBSA average annual design values and CBSA maximum annual

design values for 2017-2019 2-46

Figure 2-29. Comparison of CBSA average annual design values and CBSA maximum daily

design values for 2017-2019 2-48

Figure 2-30. R2 for ten-fold cross-validation of daily PM2.5 predictions in 2015 from three

methods for individual sites as a function of observed concentration 2-52

Figure 2-31. Comparison of 2011 annual average PM2.5 concentrations from four methods ... 2-53

Figure 2-32. Comparison of 2011 annual average PM2.5 concentrations from four methods for

regions centered on the (a) California (b) New Jersey, and (c) Arizona 2-55

Figure 2-33. (a) Spatial distribution of the CV (i.e., standard deviation divided by mean) in

percentage units for the four models in Figure 2-29. (b) Boxplot distributions of CV
for grid cells binned by the average PM2.5 concentration for the four models 2-56

Figure 2-34. Distance from the center of the 12-km grid cells to the nearest PM2.5 monitoring site
for PM2.5 measurements from the AQS database and IMPROVE network 2-56

Figure 2-35. Location of PM2.5 predictions by range in annual average concentration for the four
prediction methods at their native resolution 2-57

Figure 2-36. Annual mean PM2.5 from the VD2019 method (van Donkelaar et al., 2019) for

2001, 2006, 2011, and 2016 2-58

Figure 2-37. Spatial distribution of the annual average PM2.5 concentrations for 2015 using the

DI2019 surface nationwide (panel A) and for CBSAs only (panel B) 2-59

Figure 2-38. Smoke and fire detections observed by the MODIS instrument onboard the Aqua

satellite on August 4th, 2017 accessed through NASA Worldview 2-67

-------
Figure 2-39. Fine PM mass time series during 2017 from the North Cascades IMPROVE site in
north central Washington state 2-67

Figure 2-40. Speciated annual average IMPROVE PM2.5 in |ig/m3 at select remote monitors

during 2004 and 2016 2-72

Figure 2-41. Site locations for the IMPROVE monitors in Figure 2-40 2-72

Figure 3-1. Overview of general approach for the reconsideration of the 2020 final decision on

the primary PM2.5 standards 3-14

Figure 3-2. Estimated concentration-response function and 95% confidence intervals between

PM2.5 and cardiovascular mortality in the Six Cities Study (1974-2009) 3-68

Figure 3-3. Estimated PM2.5 concentrations using the DI2019 hybrid approach and monitoring

locations and design values for the state of Georgia and the Atlanta-Sandy Springs-
Roswell, Georgia CBSA 3-73

Figure 3-4. Epidemiologic studies examining associations between long-term PM2.5 exposures

and mortality 3-82

Figure 3-5. Epidemiologic studies examining associations between long-term PM2.5 exposures

and morbidity 3-84

Figure 3-6. Epidemiologic studies examining associations between short-term PM2.5 exposures

and mortality 3-86

Figure 3-7. Epidemiologic studies examining associations between short-term PM2.5 exposures
and morbidity 3-88

Figure 3-8. Monitor-based PM2.5 concentrations in key U.S. epidemiologic studies 3-109

Figure 3-9. Monitor-based PM2.5 concentrations in key Canadian epidemiologic studies 3-110

Figure 3-10. Hybrid model-predicted PM2.5 concentrations in key U.S. epidemiologic studies

3-111

Figure 3-11. Hybrid model-predicted PM2.5 concentrations in key Canadian epidemiologic

studies 3-112

Figure 3-12. Hybrid model-predicted PM2.5 concentrations in key U.S. epidemiologic studies,

subset by spatial scale 3-114

Figure 3-13. Hybrid model-predicted PM2.5 concentrations in key U.S. epidemiologic studies,
subset by method used to average grid cells in study-reported long-term mean
PM2.5 concentrations 3-115

Figure 3-14. Hybrid model-predicted PM2.5 concentrations in key U.S. epidemiologic studies,
subset by spatial scale and method used to average grid cells in study-reported
long-term mean PM2.5 concentrations 3-116

Figure 3-15. Illustration of approach to adjusting air quality to simulate just meeting annual

standards with levels of 11.0, 9.0, and 8.0 |ig/m3 3-139

Figure 3-16. Map of 47 urban study areas included in risk modeling 3-142

-------
Figure 3-17. Available epidemiologic associations between long-term PM2.5 exposure and

mortality outcomes in demographic populations 3-144

Figure 3-18. Distribution of total risk estimates (PIvfc.s-attibutable mortality) for the current and
alternative annual standards for the subset of 30 urban study areas where the annual
standard is controlling (blue and green bars represent the Pri-PM2.5 and Sec-PM2.5
estimates, respectively) 3-154

Figure 3-19. Distribution of the difference in risk estimates between the current annual standard
(level of 12.0 |ig/m3) and alternative annual standards with levels of 11.0, 10.0, 9.0,
and 8.0 |ig/m3 for the subset of 30 urban study areas where the annual standard is
controlling 3-155

Figure 3-20. Average PM2.5 exposure concentration and PM2.5-attributable risk estimates by

demographic population when just meeting current or alternative PM2.5 standards...

3-159

Figure 3-21. Average change in PM2.5 exposure concentration and PM2.5-attributable mortality
risk estimates by demographic population when moving from the current to
alternative PM2.5 standards 3-160

Figure 3-22. PM2.5 exposure concentrations and PM2.5-attributable mortality risk estimates by

demographic population when just meeting current or alternative PM2.5 standards...

3-163

Figure 3-23. Change in PM2.5 exposure concentrations and PM2.5-attributable mortality risk

estimates by demographic population when moving from the current to alternative
PM2.5 standards 3-164

Figure 5-1. Overview of general approach for the reconsideration of the 2020 final decision on

the secondary PM standards 5-13

Figure 5-2. Relationship of viewer acceptability ratings to light extinction 5-24

-------
1 INTRODUCTION

This document, Policy Assessment for the Reconsideration of the National Ambient Air
Quality Standards for Particulate Matter (hereafter referred to as the PA), presents the policy
assessment for the U.S. Environmental Protection Agency's (EPA's) reconsideration of the
review of the national ambient air quality standards (NAAQS) for particulate matter (PM)
completed in 2020.1 The overall plan for the 2020 review was presented in the Integrated Review
Plan for the National Ambient Air Quality Standards for Particulate Matter (IRP; U.S. EPA,
2016). The IRP also identified key policy-relevant issues to be addressed in the 2020 review and
discussed the key documents that generally inform NAAQS reviews, including an Integrated
Science Assessment (ISA) and a Policy Assessment (PA). The key considerations presented in
this PA are intended to provide updates to the policy information to support the reconsideration
of the 2020 PM NAAQS final action, which retained the primary and secondary PM2.5 and PM10
standards without revision (85 FR 82684, December 18, 2020). In reconsidering the 2020 final
action, the EPA will consider the scientific and technical analyses on which the December 2020
PM NAAQS final action was based, as well as the newly available scientific information
evaluated in the Supplement to the 2019 Integrated Science Assessment for Particulate Matter
(hereafter referred to as the ISA Supplement; (U.S. EPA, 2022) and the policy implications of
the new scientific evidence and updated quantitative analyses presented in this PA. Much of the
information in this PA is drawn directly from information included in the 2019 ISA (U.S. EPA,
2019) and the 2020 PA (U.S. EPA, 2020).

This document is organized into five chapters. Chapter 1 presents introductory
information on the purpose of the PA, legislative requirements for reviews of the NAAQS, an
overview of the history of the PM NAAQS, including background information on prior reviews,
and a summary of the progress to date for the reconsideration of the 2020 final decision. Chapter
2 provides an overview of the available information on PM-related emissions, atmospheric
chemistry, monitoring and air quality. Chapter 3 focuses on policy-relevant aspects of the
currently available health effects evidence as presented in the 2019 ISA and ISA Supplement, as
well as updated exposure/risk information, and identifies and summarizes the key considerations
related to this reconsideration of the primary PM2.5 standards. Chapter 4 draws substantially from
the information presented in the 2020 PA on the policy-relevant aspects of the health effects
evidence presented in the 2019 ISA and identifies and summarizes the key considerations related

1 On June 10, 2021, the Agency announced its decision to reconsider the 2020 PM NAAQS final action. The press
release fortius announcement is available at: https://www.epa.gov/newsreleases/epa-reexamine-health-standards-
harmful-soot-previous-administration-left-unchanged

1-1

-------
to this reconsideration of the primary standard PMio. Chapter 5 focuses on policy-relevant
aspects of the currently available welfare effects evidence as presented in the 2019 ISA and ISA
Supplement, as well as updated quantitative analyses for visibility effects, and identifies and
summarizes the key considerations related to this reconsideration of the secondary PM
standards.2 More detail about the process for this reconsideration is described in section 1.4.2
below, and the approach for considering the available information for this reconsideration is
presented within Chapters 3, 4, and 5 of this PA.

1.1 PURPOSE

The PA evaluates the potential policy implications of the available scientific evidence, as
assessed in the ISA, and the potential implications of the available air quality, exposure or risk
analyses. The role of the PA is to help "bridge the gap" between the Agency's scientific
assessments and quantitative technical analyses, and the judgments required of the Administrator
in determining whether it is appropriate to retain or revise the NAAQS.

In evaluating the question of adequacy of the current standards, and whether it may be
appropriate to consider alternative standards, the PA focuses on information that is most
pertinent to evaluating the standards and their basic elements: indicator, averaging time, form,
and level.3 These elements, which together serve to define each standard, must be considered
collectively in evaluating the health and welfare protection the standards afford.

The development of the PA is also intended to facilitate advice to the Agency and
recommendations to the Administrator from an independent scientific review committee, the
Clean Air Scientific Advisory Committee (CASAC), as provided for in the Clean Air Act
(CAA). As discussed below in section 1.2, the CASAC is to advise on subjects including the
Agency's assessment of the relevant scientific information and on the adequacy of the current
standards, and to make recommendations as to any revisions of the standards that may be

2	The welfare effects considered in this reconsideration include visibility impairment, climate effects, and materials

effects (i.e., damage and soiling). Ecological effects associated with PM, and the adequacy of protection provided
by the secondary PM standards for them, are being addressed in the separate review of the secondary NAAQS for
oxides of nitrogen, oxides of sulfur and PM in recognition of the linkages between oxides of nitrogen, oxides of
sulfur, and PM with respect to atmospheric chemistry and deposition, and with respect to ecological effects.
Information on the current review of the secondary NAAQS for oxides of nitrogen, oxides of sulfur and PM can
be found at https://www.epa.gov/naaqs/nitrogen-clioxide-no2-and-sulfur-dioxide-so2-secondary-air-quality-
standards.

3	The indicator defines the chemical species or mixture to be measured in the ambient air for the purpose of

determining whether an area attains the standard. The averaging time defines the period over which air quality
measurements are to be averaged or otherwise analyzed. The form of a standard defines the air quality statistic
that is to be compared to the level of the standard in determining whether an area attains the standard. For
example, the form of the annual NAAQS for fine particulate matter is the average of annual mean concentrations
for three consecutive years, while the form of the 8-hour NAAQS for carbon monoxide is the second-highest 8-
hour average in a year. The level of the standard defines the air quality concentration used for that purpose.

1-2

-------
appropriate. The EPA generally makes available to the CASAC and the public one or more drafts
of the PA for CASAC review and public comment.

In this PA, we4 take into account the available scientific evidence, as assessed in the
Integrated Science Assessment for Particulate Matter (Final Report) (2019 ISA [U.S. EPA,
2019]) and in the ISA Supplement (U.S. EPA, 2022), as well as additional policy-relevant
analyses of air quality and risks. The evaluation and conclusions in this document have also been
informed by the advice received from the CASAC in its review of the draft PA, and also by
public comment received thus far in the reconsideration.

The PA is designed to assist the Administrator in considering the currently available
scientific and risk information and formulating judgments regarding the standards. Accordingly,
the PA will inform the Administrator's decision in this reconsideration. Beyond informing the
Administrator and facilitating the advice and recommendations of the CASAC, the PA is also
intended to be a useful reference to all parties interested in the review of the PM NAAQS. In
these roles, it is intended to serve as a source of policy-relevant information that informs the
Agency's review of the NAAQS for PM, and it is written to be understandable to a broad
audience.

1.2 LEGISLATIVE REQUIREMENTS

Two sections of the Clean Air Act (CAA) govern the establishment and revision of the
NAAQS. Section 108 (42 U.S.C. 7408) directs the Administrator to identify and list certain air
pollutants and then to issue air quality criteria for those pollutants. The Administrator is to list
those pollutants "emissions of which, in his judgment, cause or contribute to air pollution which
may reasonably be anticipated to endanger public health or welfare"; "the presence of which in
the ambient air results from numerous or diverse mobile or stationary sources"; and for which he
"plans to issue air quality criteria...." (42 U.S.C. § 7408(a)(1)). Air quality criteria are intended
to "accurately reflect the latest scientific knowledge useful in indicating the kind and extent of all
identifiable effects on public health or welfare which may be expected from the presence of [a]
pollutant in the ambient air...." 42 U.S.C. § 7408(a)(2).

Section 109 [42 U.S.C. 7409] directs the Administrator to propose and promulgate
"primary" and "secondary" NAAQS for pollutants for which air quality criteria are issued [42
U.S.C. § 7409(a)], Section 109(b)(1) defines primary standards as ones "the attainment and
maintenance of which in the judgment of the Administrator, based on such criteria and allowing

4 The terms "we," "our," and "staff' throughout this document refer to the staff in the EPA's Office of Air Quality
Planning and Standards (OAQPS).

1-3

-------
an adequate margin of safety, are requisite to protect the public health."5 Under section
109(b)(2), a secondary standard must "specify a level of air quality the attainment and
maintenance of which, in the judgment of the Administrator, based on such criteria, is requisite
to protect the public welfare from any known or anticipated adverse effects associated with the
presence of [the] pollutant in the ambient air."6

In setting primary and secondary standards that are "requisite" to protect public health
and welfare, respectively, as provided in section 109(b), the EPA's task is to establish standards
that are neither more nor less stringent than necessary. In so doing, the EPA may not consider the
costs of implementing the standards. See generally, Whitman v. American Trucking Associations,
531 U.S. 457, 465-472, 475-76 (2001). Likewise, "[attainability and technological feasibility are
not relevant considerations in the promulgation of national ambient air quality standards."
American Petroleum Institute v. Costle, 665 F.2d 1176, 1185 (D.C. Cir. 1981). At the same time,
courts have clarified the EPA may consider "relative proximity to peak background ...
concentrations" as a factor in deciding how to revise the NAAQS in the context of considering
standard levels within the range of reasonable values supported by the air quality criteria and
judgments of the Administrator. American Trucking Associations, Inc. v. EPA, 283 F.3d 355, 379
(D.C. Cir. 2002).

The requirement that primary standards provide an adequate margin of safety was
intended to address uncertainties associated with inconclusive scientific and technical
information available at the time of standard setting. It was also intended to provide a reasonable
degree of protection against hazards that research has not yet identified. See Lead Industries
Association v. EPA, 647 F.2d 1130, 1154 (D.C. Cir 1980), cert, denied, 449 U.S. 1042 (1980);
American Petroleum Institute v. Costle, 665 F.2d at 1186 (D.C. Cir. 1981), cert, denied, 455 U.S.
1034 (1982); Coalition of Battery Recyclers Ass'n v. EPA, 604 F.3d 613, 617-18 (D.C. Cir.
2010); Mississippi v. EPA, 744 F.3d 1334, 1353 (D.C. Cir. 2013). Both kinds of uncertainties are
components of the risk associated with pollution at levels below those at which human health
effects can be said to occur with reasonable scientific certainty. Thus, in selecting primary
standards that include an adequate margin of safety, the Administrator is seeking not only to
prevent pollution levels that have been demonstrated to be harmful but also to prevent lower

5 The legislative history of section 109 indicates that a primary standard is to be set at "the maximum permissible

ambient air level. .. which will protect the health of any [sensitive] group of the population," and that for this
purpose "reference should be made to a representative sample of persons comprising the sensitive group rather
than to a single person in such a group." S. Rep. No. 91-1196, 91st Cong., 2d Sess. 10 (1970).

6 Under CAA section 302(h) (42 U.S.C. § 7602(h)), effects on welfare include, but are not limited to, "effects on

soils, water, crops, vegetation, manmade materials, animals, wildlife, weather, visibility, and climate, damage to
and deterioration of property, and hazards to transportation, as well as effects on economic values and on personal
comfort and well-being."

1-4

-------
pollutant levels that may pose an unacceptable risk of harm, even if the risk is not precisely
identified as to nature or degree. The CAA does not require the Administrator to establish a
primary NAAQS at a zero-risk level or at background concentration levels, see Lead Industries
v. EPA, 647 F.2d at 1156 n.51, Mississippi v. EPA, 744 F.3d at 1351, but rather at a level that
reduces risk sufficiently so as to protect public health with an adequate margin of safety.

In addressing the requirement for an adequate margin of safety, the EPA considers such
factors as the nature and severity of the health effects involved, the size of the sensitive
population(s), and the kind and degree of uncertainties. The selection of any particular approach
to providing an adequate margin of safety is a policy choice left specifically to the
Administrator's judgment. S qq Lead Industries Association v. EPA, 647 F.2d at 1161-62;
Mississippi v. EPA, 744 F.3d at 1353.

Section 109(d)(1) of the Act requires a review be completed every five years and, if
appropriate, revision of existing air quality criteria to reflect advances in scientific knowledge on
the effects of the pollutant on public health and welfare. Under the same provision, the EPA is
also to review every five years and, if appropriate, revise the NAAQS, based on the revised air
quality criteria.7

Section 109(d)(2) addresses the appointment and advisory functions of an independent
scientific review committee. Section 109(d)(2)(A) requires the Administrator to appoint this
committee, which is to be composed of "seven members including at least one member of the
National Academy of Sciences, one physician, and one person representing State air pollution
control agencies." Section 109(d)(2)(B) provides that the independent scientific review
committee "shall complete a review of the criteria... and the national primary and secondary
ambient air quality standards... and shall recommend to the Administrator any new... standards
and revisions of existing criteria and standards as may be appropriate...." Since the early 1980s,
this independent review function has been performed by the Clean Air Scientific Advisory
Committee (CAS AC) of the EPA's Science Advisory Board. A number of other advisory
functions are also identified for the committee by section 109(d)(2)(C), which reads:

Such committee shall also (i) advise the Administrator of areas in which
additional knowledge is required to appraise the adequacy and basis of existing,
new, or revised national ambient air quality standards, (ii) describe the research
efforts necessary to provide the required information, (iii) advise the
Administrator on the relative contribution to air pollution concentrations of
natural as well as anthropogenic activity, and (iv) advise the Administrator of any
adverse public health, welfare, social, economic, or energy effects which may

7 This section of the Act requires the Administrator to complete these reviews and make any revisions that may be
appropriate "at five-year intervals."

1-5

-------
result from various strategies for attainment and maintenance of such national
ambient air quality standards.

As previously noted, the Supreme Court has held that section 109(b) "unambiguously bars cost
considerations from the NAAQS-setting process" (Whitman v. Am. Trucking Associations, 531
U.S. 457, 471 [2001]). Accordingly, while some of these issues regarding which Congress has
directed the CASAC to advise the Administrator are ones that are relevant to the standard setting
process, others are not. Issues that are not relevant to standard setting may be relevant to
implementation of the NAAQS once they are established.8

1.3 HISTORY OF REVIEWS OF THE PM NAAQS

This section summarizes the PM NAAQS that have been promulgated in past reviews
(Table 1-1). Each of these reviews is discussed briefly below.

8 Some aspects of CASAC advice may not be relevant to EPA's process of setting primary and secondary standards
that are requisite to protect public health and welfare. Indeed, were EPA to consider costs of implementation
when reviewing and revising the standards "it would be grounds for vacating the NAAQS." Whitman, 531 U. S. at
471 n.4. At the same time, the Clean Air Act directs CASAC to provide advice on "any adverse public health,
welfare, social, economic, or energy effects which may result from various strategies for attainment and
maintenance" of the NAAQS to the Administrator under section 109(d)(2)(C)(iv). In Whitman, the Court
clarified that most of that advice would be relevant to implementation but not standard setting, as it "enable [s] the
Administrator to assist the States in carrying out their statutory role as primary implementers of the NAAQS." Id.
at 470 (emphasis in original). However, the Court also noted that CASAC's "advice concerning certain aspects of
'adverse public health ... effects' from various attainment strategies is unquestionably pertinent" to the NAAQS
rulemaking record and relevant to the standard setting process. Id. at 470 n.2.

1-6

-------
Table 1-1. Summary of NAAQS promulgated for particulate matter 1971-2012.

Review
Completed

Indicator

Averaging
Time

Level

Form

1971

Total

Suspended

Particles

(TSP)

24-hour

260 |jg/m3
(primary)
150 pg/m3
(secondary)

Not to be exceeded more than once per year

Annual

75 |jg/m3
(primary)
60 |jg/m3
(secondary)

Annual geometric mean

1987

PM10

24-hour

150 |jg/m3

Not to be exceeded more than once per year on
average over a 3-year period

Annual

50 |jg/m3

Annual arithmetic mean, averaged over 3 years

1997

PM2.5

24-hour

65 |jg/m3

98th percentile, averaged over 3 years

Annual

15.0 |jg/m3

Annual arithmetic mean, averaged over 3 years3

PM10

24-hour

150 |jg/m3

99th percentile, averaged over 3 yearsb

Annual

50 |jg/m3

Annual arithmetic mean, averaged over 3 years

2006

PM2.5

24-hour

35 |jg/m3

98th percentile, averaged over 3 years

Annual

15.0 |jg/m3

Annual arithmetic mean, averaged over 3 years0

PM10

24-hourd

150 |jg/m3

Not to be exceed more than once per year on average
over a 3-year period

2012

PM2.5

24-hour

35 |jg/m3

98th percentile, averaged over 3 years

Annual

12.0 pg/m3
(primary)
15.0 pg/m3
(secondary)

Annual mean, averaged over 3 yearse

PM10

24-hour

150 pg/m3

Not to be exceeded more than once per year on
average over 3 years

Note: When not specified, primary and secondary standards are identical.

a The level of the 1997 annual PM2.5 standard was to be compared to measurements made at the community-
oriented monitoring site recording the highest concentration or, if specific constraints were met, measurements
from multiple community-oriented monitoring sites could be averaged (i.e., "spatial averaging") (62 FR 38652,

July 18, 1997).

b When the 1997 standards were vacated (see below), the form of the 1987 standards remained in place (i.e., not
to be exceeded more than once per year on average over a 3-year period).

c The EPA tightened the constraints on the spatial averaging criteria by further limiting the conditions under which
some areas may average measurements from multiple community-oriented monitors to determine compliance (71
FR 61144, October 17, 2006).

d The EPA revoked the annual PM10 NAAQS in 2006 (71 FR 61144, October 17, 2006).
e In the 2012 decision, the EPA eliminated the option for spatial averaging (78 FR 3086, January 15, 2013).

1-7

-------
1.3.1 Reviews Completed in 1971 and 1987

The EPA first established NAAQS for PM in 1971 (36 FR 8186, April 30, 1971), based
on the original Air Quality Criteria Document (AQCD) (DHEW, 1969).9 The federal reference
method (FRM) specified for determining attainment of the original standards was the high-
volume sampler, which collects PM up to a nominal size of 25 to 45 micrometers (|im) (referred
to as total suspended particulates or TSP). The primary standards were set at 260 |ig/m3, 24-hour
average, not to be exceeded more than once per year, and 75 |ig/m3, annual geometric mean. The
secondary standards were set at 150 |ig/m3, 24-hour average, not to be exceeded more than once
per year, and 60 |ig/m3, annual geometric mean.

In October 1979 (44 FR 56730, October 2, 1979), the EPA announced the first periodic
review of the air quality criteria and NAAQS for PM. Revised primary and secondary standards
were promulgated in 1987 (52 FR 24634, July 1, 1987). In the 1987 decision, the EPA changed
the indicator for particles from TSP to PMio, in order to focus on the subset of inhalable particles
small enough to penetrate to the thoracic region of the respiratory tract (including the
tracheobronchial and alveolar regions), referred to as thoracic particles.10 The level of the 24-
hour standards (primary and secondary) was set at 150 |ig/m3, and the form was one expected
exceedance per year, on average over three years. The level of the annual standards (primary and
secondary) was set at 50 |ig/m3, and the form was annual arithmetic mean, averaged over three
years.

1.3.2 Review Completed in 1997

In April 1994, the EPA announced its plans for the second periodic review of the air
quality criteria and NAAQS for PM, and in 1997 the EPA promulgated revisions to the NAAQS
(62 FR 38652, July 18, 1997). In the 1997 decision, the EPA determined that the fine and coarse
fractions of PMio should be considered separately. This determination was based on evidence
that serious health effects were associated with short- and long-term exposures to fine particles in
areas that met the existing PMio standards. The EPA added new standards, using PM2.5 as the
indicator for fine particles (with PM2.5 referring to particles with a nominal mean aerodynamic
diameter less than or equal to 2.5 |im). The new primary standards were as follows: (1) an annual
standard with a level of 15.0 |ig/m3, based on the 3-year average of annual arithmetic mean

9 Prior to the review initiated in 2007 (see below), the AQCD provided the scientific foundation (i.e., the air quality

criteria) for the NAAQS. Beginning in that review, the ISA has replaced the AQCD.

10 PMio refers to particles with a nominal mean aerodynamic diameter less than or equal to 10 |im. More
specifically, 10 |im is the aerodynamic diameter for which the efficiency of particle collection is 50 percent.

1-8

-------
PM2.5 concentrations from single or multiple community-oriented monitors;11 and (2) a 24-hour
standard with a level of 65 |ig/m3, based on the 3-year average of the 98th percentile of 24-hour
PM2.5 concentrations at each monitor within an area. Also, the EPA established a new reference
method for the measurement of PM2.5 in the ambient air and adopted rules for determining
attainment of the new standards. To continue to address the health effects of the coarse fraction
of PM10 (referred to as thoracic coarse particles or PM10-2.5; generally including particles with a
nominal mean aerodynamic diameter greater than 2.5 |im and less than or equal to 10 |im), the
EPA retained the annual primary PM10 standard and revised the form of the 24-hour primary
PM10 standard to be based on the 99th percentile of 24-hour PM10 concentrations at each monitor
in an area. The EPA revised the secondary standards by setting them equal in all respects to the
newly established primary standards.

Following promulgation of the 1997 PM NAAQS, petitions for review were filed by
several parties, addressing a broad range of issues. In May 1999, the U.S. Court of Appeals for
the District of Columbia Circuit (D.C. Circuit) upheld the EPA's decision to establish fine
particle standards, holding that "the growing empirical evidence demonstrating a relationship
between fine particle pollution and adverse health effects amply justifies establishment of new
fine particle standards." American Trucking Associations v. EPA, 175 F. 3d at 1027, 1055-56
(D.C. Cir. 1999). The D.C. Circuit also found "ample support" for the EPA's decision to regulate
coarse particle pollution, but vacated the 1997 PM10 standards, concluding that the EPA had not
provided a reasonable explanation justifying use of PM10 as an indicator for coarse particles.
American Trucking Associations v. EPA, 175 F. 3d at 1054-55. Pursuant to the D.C. Circuit's
decision, the EPA removed the vacated 1997 PM10 standards, and the pre-existing 1987 PM10
standards remained in place (65 FR 80776, December 22, 2000). The D.C. Circuit also upheld
the EPA's determination not to establish more stringent secondary standards for fine particles to
address effects on visibility. American Trucking Associations v. EPA, 175 F. 3d at 1027.

The D.C. Circuit also addressed more general issues related to the NAAQS, including
issues related to the consideration of costs in setting NAAQS and the EPA's approach to
establishing the levels of NAAQS. Regarding the cost issue, the court reaffirmed prior rulings
holding that in setting NAAQS the EPA is "not permitted to consider the cost of implementing
those standards." American Trucking Associations v. EPA, 175 F. 3d at 1040-41. Regarding the
levels of NAAQS, the court held that the EPA's approach to establishing the level of the

11 The 1997 annual PM2.5 standard was to be compared with measurements made at the community-oriented
monitoring site recording the highest concentration or, if specific constraints were met, measurements from
multiple community-oriented monitoring sites could be averaged (i.e., "spatial averaging"). In the 2012 review,
the EPA replaced the term "community-oriented" monitor with the term "area-wide" monitor. Area-wide
monitors are those sited at the neighborhood scale or larger, as well as those monitors sited at micro- or middle-
scales that are representative of many such locations in the same CBSA (78 FR 3236, January 15, 2013).

1-9

-------
standards in 1997 (i.e., both for PM and for the ozone NAAQS promulgated on the same day)
effected "an unconstitutional delegation of legislative authority." American Trucking
Associations v. EPA, 175 F. 3d at 1034-40. Although the court stated that "the factors EPA uses
in determining the degree of public health concern associated with different levels of ozone and
PM are reasonable," it remanded the rule to the EPA, stating that when the EPA considers these
factors for potential non-threshold pollutants "what EPA lacks is any determinate criterion for
drawing lines" to determine where the standards should be set.

The D.C. Circuit's holding on the cost and constitutional issues were appealed to the
United States Supreme Court. In February 2001, the Supreme Court issued a unanimous decision
upholding the EPA's position on both the cost and constitutional issues. Whitman v. American
Trucking Associations, 531 U.S. 457, 464, 475-76. On the constitutional issue, the Court held
that the statutory requirement that NAAQS be "requisite" to protect public health with an
adequate margin of safety sufficiently guided the EPA's discretion, affirming the EPA's
approach of setting standards that are neither more nor less stringent than necessary.

The Supreme Court remanded the case to the Court of Appeals for resolution of any
remaining issues that had not been addressed in that court's earlier rulings. Id. at 475-76. In a
March 2002 decision, the Court of Appeals rejected all remaining challenges to the standards,
holding that the EPA's PM2.5 standards were reasonably supported by the administrative record
and were not "arbitrary and capricious" American Trucking Associations v. EPA, 283 F. 3d 355,
369-72 (D.C. Cir. 2002).

1.3.3 Review Completed in 2006

In October 1997, the EPA published its plans for the third periodic review of the air
quality criteria and NAAQS for PM (62 FR 55201, October 23, 1997). After the CASAC and
public review of several drafts, the EPA's NCEA finalized the AQCD in October 2004 (U.S.
EPA, 2004a, U.S. EPA, 2004b). The EPA's OAQPS finalized a Risk Assessment and Staff Paper
in December 2005 (Abt Associates, 2005, U.S. EPA, 2005).12 On December 20, 2005, the EPA
announced its proposed decision to revise the NAAQS for PM and solicited public comment on a
broad range of options (71 FR 2620, January 17, 2006). On September 21, 2006, the EPA
announced its final decisions to revise the primary and secondary NAAQS for PM to provide
increased protection of public health and welfare, respectively (71 FR 61144, October 17, 2006).
With regard to the primary and secondary standards for fine particles, the EPA revised the level

12 Prior to the review initiated in 2007, the Staff Paper presented the EPA staffs considerations and conclusions
regarding the adequacy of existing NAAQS and, when appropriate, the potential alternative standards that could
be supported by the evidence and information. More recent reviews present this information in the Policy
Assessment.

1-10

-------
of the 24-hour PM2.5 standards to 35 |ig/m3, retained the level of the annual PM2.5 standards at
15.0 |ig/m3, and revised the form of the annual PM2.5 standards by narrowing the constraints on
the optional use of spatial averaging. With regard to the primary and secondary standards for
PM10, the EPA retained the 24-hour standards, with levels at 150 |ig/m3, and revoked the annual
standards.13 The Administrator judged that the available evidence generally did not suggest a link
between long-term exposure to existing ambient levels of coarse particles and health or welfare
effects. In addition, a new reference method was added for the measurement of PM10-2.5 in the
ambient air in order to provide a basis for approving federal equivalent methods (FEMs) and to
promote the gathering of scientific data to support future reviews of the PM NAAQS.

Several parties filed petitions for review following promulgation of the revised PM
NAAQS in 2006. These petitions addressed the following issues: (1) selecting the level of the
primary annual PM2.5 standard; (2) retaining PM10 as the indicator of a standard for thoracic
coarse particles, retaining the level and form of the 24-hour PM10 standard, and revoking the
PM10 annual standard; and (3) setting the secondary PM2.5 standards identical to the primary
standards. On February 24, 2009, the U.S. Court of Appeals for the District of Columbia Circuit
issued its opinion in the case American Farm Bureau Federation v. EPA, 559F.3d512 (D.C.
Cir. 2009). The court remanded the primary annual PM2.5 NAAQS to the EPA because the
Agency failed to adequately explain why the standards provided the requisite protection from
both short- and long-term exposures to fine particles, including protection for at-risk populations.
American Farm Bureau Federation v. EPA, 559 F. 3d 512, 520-27 (D.C. Cir. 2009). With regard
to the standards for PM10, the court upheld the EPA's decisions to retain the 24-hour PM10
standard to provide protection from thoracic coarse particle exposures and to revoke the annual
PM10 standard. American Farm Bureau Federation, 559 F. 2d at 533-38. With regard to the
secondary PM2.5 standards, the court remanded the standards to the EPA because the Agency
failed to adequately explain why setting the secondary PM standards identical to the primary
standards provided the required protection for public welfare, including protection from visibility
impairment. American Farm Bureau Federation, 559 F. 2d at 528-32. The EPA responded to the
court's remands as part of the next review of the PM NAAQS, which was initiated in 2007
(discussed below).

13 In the 2006 proposal, the EPA proposed to revise the 24-hour PM10 standard in part by establishing a new PMi0-2.5
indicator for thoracic coarse particles (i.e., particles generally between 2.5 and 10 |im in diameter). The EPA
proposed to include any ambient mix of PMi 0-2.5 that was dominated by resuspended dust from high density
traffic on paved roads and by PM from industrial sources and construction sources. The EPA proposed to exclude
any ambient mix of PMi 0-2.5 that was dominated by rural windblown dust and soils and by PM generated from
agricultural and mining sources. In the final decision, the existing PM10 standard was retained, in part due to an
"inability.. .to effectively and precisely identify which ambient mixes are included in the [PMi0-2.5] indicator and
which are not" (71 FR 61197, October 17, 2006).

1-11

-------
1.3.4 Review Completed in 2012

In June 2007, the EPA initiated the fourth periodic review of the air quality criteria and
the PM NAAQS by issuing a call for information in the Federal Register (72 FR 35462, June 28,
2007). Based on the NAAQS review process, as revised in 2008 and again in 2009,14 the EPA
held science/policy issue workshops on the primary and secondary PM NAAQS (72 FR 34003,
June 20, 2007; 72 FR 34005, June 20, 2007), and prepared and released the planning and
assessment documents that comprise the review process (i.e., IRP (U.S. EPA, 2008), ISA (U.S.
EPA, 2009a), REA planning documents for health and welfare (U.S. EPA, 2009b, U.S. EPA,
2009c), a quantitative health risk assessment (U.S. EPA, 2010a) and an urban-focused visibility
assessment (U.S. EPA, 2010b), and PA (U.S. EPA, 2011)). In June 2012, the EPA announced its
proposed decision to revise the NAAQS for PM (77 FR 38890, June 29, 2012).

In December 2012, the EPA announced its final decisions to revise the primary NAAQS
for PM to provide increased protection of public health (78 FR 3086, January 15, 2013). With
regard to primary standards for PM2.5, the EPA revised the level of the annual PM2.5 standard15 to
12.0 |ig/m3 and retained the 24-hour PM2.5 standard, with its level of 35 |ig/m3. For the primary
PM10 standard, the EPA retained the 24-hour standard to continue to provide protection against
effects associated with short-term exposure to thoracic coarse particles (i.e., PM10-2.5). With
regard to the secondary PM standards, the EPA generally retained the 24-hour and annual PM2.5
standards16 and the 24-hour PM10 standard to address visibility and non-visibility welfare effects.

As with previous reviews, petitioners challenged the EPA's final rule. Petitioners argued
that the EPA acted unreasonably in revising the level and form of the annual standard and in
amending the monitoring network provisions. On judicial review, the revised standards and
monitoring requirements were upheld in all respects. NAMv EPA, 750 F.3d 921 (D.C. Cir.

2014).

1.3.5 Review Completed in 2020

In December 2014, the EPA announced the initiation of the periodic review of the air
quality criteria for PM and of the PM2.5 and PM10 NAAQS and issued a call for information in
the Federal Register (79 FR 71764, December 3, 2014). On February 9 to 11, 2015, the EPA's
NCEA and OAQPS held a public workshop to inform the planning for the review of the PM
NAAQS (announced in 79 FR 71764, December 3, 2014). Workshop participants, including a

14 The history of the NAAQS review process, including revisions to the process, is discussed at

https://www.epa.gov/naaqs/historical-information-naaqs-review-process.

15 The EPA also eliminated the option for spatial averaging.

16 Consistent with the primary standard, the EPA eliminated the option for spatial averaging with the annual
standard.

1-12

-------
wide range of external experts as well as EPA staff representing a variety of areas of expertise
(e.g., epidemiology, human and animal toxicology, risk/exposure analysis, atmospheric science,
visibility impairment, climate effects), were asked to highlight significant new and emerging PM
research, and to make recommendations to the Agency regarding the design and scope of the
review. This workshop provided for a public discussion of the key science and policy-relevant
issues around which the EPA structured the review of the PM NAAQS and of the most
meaningful new scientific information that would be available in the review to inform our
understanding of these issues.

The input received at the workshop guided the EPA staff in developing a draft IRP,
which was reviewed by the CASAC Particulate Matter Panel and the seven-member chartered
CASAC, and was discussed on public teleconferences held in May 2016 (81 FR 13362, March
14, 2016) and August 2016 (81 FR 39043, June 15, 2016). Advice from the seven-member
chartered CASAC, as developed with support from the CASAC Particulate Matter Panel, and the
public were considered in developing the final IRP for the review (U.S. EPA, 2016). The final
IRP discussed the approaches to be taken in developing key scientific, technical, and policy
documents in the review and the key policy-relevant issues that frame the EPA's consideration of
whether the primary and/or secondary NAAQS for PM should be retained or revised.

In May 2018, the Administrator issued a memorandum describing a "back-to-basics"
process for reviewing the NAAQS (Pruitt, 2018). This memo announced the Agency's intention
to conduct the review of the PM NAAQS in such a manner as to ensure that any necessary
revisions were finalized by December 2020. Following this memo, on October 10, 2018 the
Administrator additionally announced that the role of reviewing the key science assessments
developed as part of the ongoing review of the PM NAAQS (i.e., drafts of the ISA and PA)
would be performed only by the seven-member chartered CASAC (i.e., without the support of
the CASAC Particulate Matter Panel that reviewed the draft IRP).17

The EPA released the draft ISA in October 2018 (83 FR 53471, October 23, 2018). The
draft ISA was reviewed by the chartered CASAC at a public meeting held in Arlington, VA in
December 2018 (83 FR 55529, November 6, 2018) and was discussed on a public teleconference
in March 2019 (84 FR 8523, March 8, 2019). The CASAC provided its advice on the draft ISA
in a letter to the EPA Administrator dated April 11, 2019 (Cox, 2019a). The EPA took steps to
address these comments in the final ISA, which was released in December 2019 (U.S. EPA,
2019).

The EPA released the draft PA in September 2019 (84 FR 47944, September 11, 2019).
The draft PA was reviewed by the chartered CASAC and discussed in October 2019 at a public

17 Announcement available at: https://www.regulations.gov/document/EPA-HQ-OAR-2015-0072-0223

1-13

-------
meeting held in Cary, NC. Public comments were received via a separate public teleconference
(84 FR 51555, September 30, 2019). A public meeting to discuss the chartered CAS AC letter
and response to charge questions on the draft PA was held in Cary, NC in December 2019 (84
FR 58713, November 1, 2019), and the CAS AC provided its advice on the draft PA, including its
advice on the current primary and secondary PM standards, in a letter to the EPA Administrator
dated December 16, 2019 (Cox, 2019b). With regard to the primary standards, the CAS AC
recommended retaining the current 24-hour PM2.5 and PM10 standards but did not reach
consensus on the adequacy of the current annual PM2.5 standard. With regard to the secondary
standards, the CASAC recommended retaining the current standards. In response to the
CASAC's comments, the 2020 final PA incorporated a number of changes (U.S. EPA, 2020), as
described in detail in section I.C.5 of the 2020 proposal (85 FR 24100, April 30, 2020).

On April 14, 2020, the EPA proposed to retain all of the primary and secondary PM
standards, without revision. These proposed decisions were published in the Federal Register on
April 30, 2020 (85 FR 24094, April 30, 2020). The EPA's final decision on the PM NAAQS was
published in the Federal Register on December 18, 2020 (85 FR 82684, December 18, 2020). In
the 2020 final decision, the EPA retained the primary and secondary PM2.5 and PM10 standards,
without revision. The EPA received three petitions for judicial review (described in more detail
in section 1.4.3 below), as well as three petitions for reconsideration of the 2020 final action.

1.4 RECONSIDERATION OF THE 2020 PM NAAQS FINAL ACTION

On January 20, 2021, President Biden issued an "Executive Order on Protecting Public
Health and the Environment and Restoring Science to Tackle the Climate Crisis," (Executive
Order 13990; 86 FR 7037, January 25, 2021)18 which directed review of certain agency actions.
An accompanying fact sheet provides a non-exclusive list of agency actions that agency heads
will review in accordance with that order, including the 2020 Particulate Matter NAAQS
Decision.19

1.4.1 Decision to Initiate a Reconsideration

On June 10, 2021, the Agency announced its decision to reconsider the 2020 PM
NAAQS final action.20 The EPA is reconsidering the December 2020 decision because the

18	See https://www.whitehouse.gOv/briefing-room/presidential-actions/2021/01/20/executive-order-protecting-
public-health-and-environment-and-restoring-science-to-tackle-climate-crisis/

19	See https://www.whitehouse.gOv/briefing-room/statements-releases/2021/01/20/fact-sheet-list-of-agency-actions-
for-review/

20	The press release fortius announcement is available at: https://www.epa.gov/newsreleases/epa-reexamine-health-
standards-harmful-soot-previous-administration-left-unchanged

1-14

-------
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. We note
that the 2020 PA concluded that the scientific evidence and information supported revising the
level of the primary annual PM2.5 standard to below the current level of 12.0 |ig/m3 while
retaining the primary 24-hour PM2.5 standard (U.S. EPA, 2020). The EPA also notes that the
2020 PA concluded that the available scientific evidence and information supported retaining the
primary PM10 standard and secondary PM standards without revision (U.S. EPA, 2020).

1.4.2 Process for Reconsideration of the 2020 PM NAAQS Decision

In its announcement of the reconsideration of the PM NAAQS, the Agency explained
that, in support of the reconsideration, it would develop a supplement to the 2019 ISA and a
revised PA. The EPA also explained that the draft ISA Supplement and draft PA would be
reviewed at a public meeting by the CASAC, and the public would have opportunities to
comment on these documents during the CASAC review process, as well as to provide input
during the rulemaking through the public comment process and public hearings on the proposed
rulemaking.

On March 31, 2021, the Administrator announced his decision to reestablish the
membership of the CASAC to "ensure the agency received the best possible scientific insight to
support our work to protect human health and the environment."21 Consistent with this
memorandum, a call for nominations of candidates to the EPA's chartered CASAC was
published in the Federal Register (86 FR 17146, April 1, 2021). On June 17, 2021, the
Administrator announced his selection of the seven members to serve on the chartered CASAC.22
23 Additionally, a call for nominations of candidates to a PM-specific panel was published in the
Federal Register (86 FR 33703, June 25, 2021). The members of the PM CASAC panel were
announced on August 30, 2021.24

The draft ISA Supplement was released in September 2021 (86 FR 54186, September 30,
2021). The CASAC PM panel met at a virtual public meeting in November 2021 to review the
draft ISA Supplement (86 FR 52673, September 22, 2021). A virtual public meeting was then

21	The press release fortius announcement is available at: https://www.epa.gov/newsreleases/administrator-regan-
directs-epa-reset-critical-science-focused-federal-advisory

22	The press release fortius announcement is available at: https://www.epa.gov/newsreleases/epa-announces-
selections-charter-members-clean-air-scientific-advisory-committee

23	The list of members of the chartered CASAC and their biosketches are available at:

https://yosemite.epa. gov/sab/sabpeople.nsf/WebExternalCommitteeRosters?OpenView&committee=CASAC&sec
ondname=Clean%20Air%20Scientific%20Advisory%20Committee%20

24	The list of members of the PM CASAC panel and their biosketches are available at:

https://casac. epa.gov/ords/sab/f?p=105:14:997922956404 7::: 14:P 14COMMITTEEON:2021%20CASAC%20P
M%>20Panel

1-15

-------
held in February 2022, and during this meeting the chartered CASAC considered the CASAC
PM panel's draft letter to the Administrator on the draft ISA Supplement (87 FR 958, January 7,
2022). The chartered CASAC provided its advice on the draft ISA Supplement in a letter to the
EPA Administrator dated March 18, 2022 (Sheppard, 2022a). The EPA took steps to address
these comments in the final ISA Supplement, which was released in May 2022 (U.S. EPA,

2022).

The evidence presented within the 2019 ISA, along with the targeted identification and
evaluation of new scientific information in the ISA Supplement, provides the scientific basis for
the reconsideration of the 2020 PM NAAQS final decision. The ISA Supplement focuses on a
thorough evaluation of some studies that became available after the literature cutoff date of the
2019 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 ISA. In
selecting the health effects to evaluate within the ISA Supplement, the EPA focused on health
effects for which the evidence supported a "causal relationship" because those were the health
effects that were most useful in informing conclusions in the 2020 PA (U.S. EPA, 2022, section
1.2.1).25 Consistent with the rationale for the focus on certain health effects, in selecting the non-
ecological welfare effects to evaluate within the ISA supplement, the EPA focused on the non-
ecological welfare effects for which the evidence supported a "causal relationship" and for which
quantitative analyses could be supported by the evidence because those were the welfare effects
that were most useful in informing conclusions in the 2020 PA.26 Specifically, for non-ecological
welfare effects, the focus within the ISA Supplement is on visibility effects. The ISA
Supplement also considers recent health effects evidence that addresses key scientific topics

25	As described in section 1.2.1 of the ISA Supplement: "In considering the public health protection provided by the
current primary PM2 5 standards, and the protection that could be provided by alternatives, [the U.S. EPA, within
the 2020 PM PA] emphasized health outcomes for which the ISA determined that the evidence supports either a

'causal' or a 'likely to be causal' relationship with PM2 5 exposures" (U.S. EPA, 2020). Although the 2020 PA
initially focused on this broader set of evidence, the basis of the discussion on potential alternative standards
primarily focused on health effect categories where the 2019 PM ISA concluded a 'causal relationship'' (i.e.,
short- and long-term PM2 5 exposure and cardiovascular effects and mortality) as reflected in Figures 3-7 and 3-8
of the 2020 PA (U.S. EPA, 2020)."

26	As described in section 1.2.1 of the ISA Supplement: "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 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 withPM
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, 2020)."

1-16

-------
where the literature has evolved since the 2020 review was completed, specifically since the
literature cutoff date for the 2019 ISA.27

Building on the rationale presented in section 1.2.1, the ISA Supplement considered peer-
reviewed studies published from approximately January 2018 through March 2021 that meet the
following criteria:

•	Health effects:

-	U.S. and Canadian epidemiologic studies for health effect categories where the
2019 PM ISA concluded a "causal relationship" (i.e., short- and long-term PM2.5
exposure and cardiovascular effects and mortality).

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)

•	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

•	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.

-	U.S. and Canadian-based epidemiologic studies that examined the relationship
between PM2.5 exposures and severe acute respiratory syndrome coronavirus 2
(SARS-CoV-2) infection and coronavirus disease 2019 (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 socioeconomic status (SES)

Given the narrow scope of the ISA Supplement, it is important to recognize that the
evaluation does not encompass the full multidisciplinary evaluation presented within the 2019
ISA that would result in weight-of-evidence conclusions on causality (i.e., causality
determinations). The ISA Supplement critically evaluates and provides key study specific
information for those recent studies deemed to be of greatest significance for informing
preliminary conclusions on the PM NAAQS in the context of the body of evidence and scientific
conclusions presented in the 2019 ISA. In its review of the draft ISA Supplement, the CAS AC

27 These key scientific topics include experimental studies conducted at near-ambient concentrations, epidemiologic
studies that employed alternative methods for confounder control 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, studies that examine disparities in PM2 5 exposure and the risk of health effects
(U.S. EPA, 2022, section 1.2.1).

1-17

-------
noted that they found "the Draft ISA Supplement to be a well-written, comprehensive evaluation of
the new scientific information published since the 2019 PM ISA" (Sheppard, 2022a, p. 2 of letter).
Furthermore, the CASAC stated that "the final Integrated Science Assessment (ISA) Supplement...
deserve[s] the Administrator's full consideration and [is] adequate for rulemaking" (Sheppard,
2022a, p. 2 of letter). However, recognizing the limited scope of the draft ISA Supplement, the
CASAC stated that "[although this limitation is appropriate for the targeted purpose of the Draft ISA
Supplement.. .this limiting of scope applies only to this document and is not intended to establish a
precedent for future IS As" (Sheppard, 2022a, p. 2 of letter).

The draft PA was released in October 2021 (86 FR 56263, October 8, 2021). The
CASAC PM panel met at a virtual public meeting in December 2021 to review the draft PA (86
FR 52673, September 22, 2021). A virtual public meeting was then held in February 2022 and
March 2022, and during this meeting the chartered CASAC considered the CASAC PM panel's
draft letter to the Administrator on the draft PA (87 FR 958, January 7, 2022). The chartered
CASAC provided its advice on the draft PA in a letter to the EPA Administrator dated March 18,
2022 (Sheppard, 2022b). The EPA took steps to address these comments finalizing this PA. This
PA considers the scientific evidence presented in the 2019 ISA and ISA Supplement, and
considers the quantitative and technical information presented in the 2020 PA, along with
updated and newly available analyses since the completion of the 2020 review. For those health
and welfare effects for which the ISA Supplement evaluated recently available evidence and for
which updated quantitative analyses were supported (i.e., PM2.5-related health effects and
visibility effects), the PA includes consideration of this newly available scientific and technical
information in reaching preliminary conclusions. For those health and welfare effects for which
newly available scientific and technical information were not evaluated (i.e., PMio-2 5-related
health effects and non-visibility effects), the conclusions presented in this PA rely heavily on the
information that supported the conclusions in the 2020 PA.

1.4.3 Ongoing Litigation

Following publication of the 2020 final action, several parties filed petitions for review of
the EPA's final decision in the D.C. Circuit and the Court consolidated the cases. In order to
consider whether reconsideration of the 2020 final action was warranted, the EPA moved for two
90-day abeyances in these consolidated cases, which the Court granted. After the EPA
announced that is reconsidering the 2020 final decision, the EPA filed a motion with the Court to
hold the consolidated cases in abeyance until March 1, 2023, which the court granted on October
1, 2021.

1-18

-------
REFERENCES

Abt Associates, Inc. (2005). Particulate matter health risk assessment for selected urban areas:
Draft report. EPA Contract No. 68-D-03-002. U.S. Environmental Protection Agency.
Research Triangle Park, NC. Available at:

http://www3.epa.gov/ttn/naaqs/standards/pm/data/PMrisk2005122Q.pdf.

Cox, LA. (2019a). Letter from Louis Anthony Cox, Jr., Chair, Clean Air Scientific Advisory
Committee, to Administrator Andrew R. Wheeler. Re: CASAC Review of the EPA's
Integrated Science Assessment for Particulate Matter (External Review Draft - October
2018). April 11, 2019. EPA-CASAC-19-002. Office of the Administrator, Science
Advisory Board U.S. EPA HQ, Washington DC. Available at:

https://vosemite.epa.gov/sab/sabproduct.nsf/LookupWebReportsLastMonthCASAC/932
D1DF8C2A9043F852581000048170D?QpenDocument&TableRow=2.3#2.

Cox, LA. (2019b). Letter from Louis Anthony Cox, Jr., Chair, Clean Air Scientific Advisory
Committee, to Administrator Andrew R. Wheeler. Re: CASAC Review of the EPA's
Policy Assessment for the Review of the National Ambient Air Quality Standards for
Particulate Matter (External Review Draft - September 2019). December 16, 2019. EPA-
CASAC-20-001. Office of the Administrator, Science Advisory Board U.S. EPA HQ,
Washington DC. Available at:

https://vosemite.epa.gov/sab/sabproduct.nsf/264cbl227d55e02c852574020Q7446a4/E2F
6C7173 72016128525 84D20069DFB l/$File/EP A-C AS AC-20-001 .pdf.

DHEW (1969). Air Quality Criteria for Particulate Matter. National Air Pollution Control
Administration. Washington, D.C. U.S. Department of Health. January 1969.

Pruitt, E. (2018). Memorandum from E. Scott Pruitt, Administrator, U.S. EPA to Assistant

Administrators. Back-to-Basics Process for Reviewing National Ambient Air Quality
Standards. May 9, 2018. Office of the Administrator U.S. EPA HQ, Washington DC.
Available at: https://www.epa.gov/criteria-air-pollutants/back-basics-process-reviewing-
national-ambient-air-qualitv-standards.

Sheppard, EA. (2022a). Letter from Elizabeth A. (Lianne) Sheppard, Chair, Clean Air Scientific
Advisory Committee, to Administrator Michael S. Regan. Re: CASAC Review of the
EPA's Supplement to the 2019 Integrated Science Assessment for Particulate Matter
(External Review Draft - October 2021). March 18, 2022. EPA-CASAC-22-001. Office
of the Administrator, Science Advisory Board U.S. EPA HQ, Washington DC. Available
at:

https://casac.epa. gov/ords/sab/f?p=l 13:18:7147163581076: ::RP.18:P18_ID:2606#report.

Sheppard, EA. (2022b). Letter from Elizabeth A. (Lianne) Sheppard, Chair, Clean Air Scientific
Advisory Committee, to Administrator Michael S. Regan. Re: CASAC Review of the
EPA's Policy Assessment for the Review of the National Ambient Air Quality Standards
for Particulate Matter (External Review Draft - October 2021). March 18, 2022. EPA-
CAS AC-22-002. Office of the Administrator, Science Advisory Board U.S. EPA HQ,
Washington DC. Available at:

1-19

-------
https://casac.epa.gov/ords/sab/f?p=105:18:10792850355838:::RP.18:P18 ID:2607#report

U.S. EPA (2004a). Air Quality Criteria for Particulate Matter. (Vol I of II). Office of Research
and Development. Research Triangle Park, NC. U.S. EPA. EPA-600/P-99-002aF.

October 2004. Available at:

https://nepis.epa. gov/Exe/ZvPURL.cgi?Dockev=P 100LFIQ.txt.

U.S. EPA (2004b). Air Quality Criteria for Particulate Matter. (Vol II of II). Office of Research
and Development. Research Triangle Park, NC. U.S. EPA. EPA-600/P-99-002bF.

October 2004. Available at:

https://nepis.epa. gov/Exe/ZvPURL.cgi?Dockev=P100LG7Q.txt.

U.S. EPA (2005). Review of the National Ambient Air Quality Standards for Particulate Matter:
Policy Assessment of Scientific and Technical Information, OAQPS Staff Paper. Office
of Air Quality Planning and Standards. Research Triangle Park, NC. U.S. EPA. EPA-
452/R-05-005a. December 2005. Available at:
https://nepis.epa. gov/Exe/ZvPURL.cgi?Dockev=P 1009MZM.txt.

U.S. EPA (2008). Integrated Review Plan for the National Ambient Air Quality Standards for
Particulate Matter. Office of Research and Development, National Center for
Environmental Assessment; Office of Air Quality Planning and Standards, Health and
Environmental Impacts Division. Research Triangle Park, NC. U.S. EPA. EPA 452/R-08-
004. March 2008. Available at:

https://nepis.epa. gov/Exe/ZvPURL.cgi?Dockev=P1001FB9.txt.

U.S. EPA (2009a). Integrated Science Assessment for Particulate Matter (Final Report). Office
of Research and Development, National Center for Environmental Assessment. Research
Triangle Park, NC. U.S. EPA. EPA-600/R-08-139F. December 2009. Available at:
http s: //cfpub. epa. gov/ncea/ri sk/recordi spl av. cfm? dei d=216546.

U.S. EPA (2009b). Particulate Matter National Ambient Air Quality Standards: Scope and

Methods Plan for Health Risk and Exposure Assessment. Office of Air Quality Planning
and Standards, Health and Environmental Impacts Division. Research Triangle Park, NC.
U.S. EPA. EPA-452/P-09-002. February 2009. Available at:
https://nepis.epa. gov/Exe/ZvPURL.cgi?Dockev=P100FLWP.txt.

U.S. EPA (2009c). Particulate Matter National Ambient Air Quality Standards: Scope and

Methods Plan for Urban Visibility Impact Assessment. Office of Air Quality Planning
and Standards, Health and Environmental Impacts Division. Research Triangle Park, NC.
U.S. EPA. EPA-452/P-09-001. February 2009. Available at:
https://nepis.epa. gov/Exe/ZvPURL.cgi?Dockev=P100FLUX.txt.

U.S. EPA (2010a). Quantitative Health Risk Assessment for Particulate Matter (Final Report).
Office of Air Quality Planning and Standards, Health and Environmental Impacts
Division. Research Triangle Park, NC. U.S. EPA. EPA-452/R-10-005. June 2010.
Available at: https://nepis.epa. gov/Exe/ZvPURL.cgi?Dockev=P1007RFC.txt.

1-20

-------
U.S. EPA (2010b). Particulate Matter Urban-Focused Visibility Assessment (Final Document).
Office of Air Quality Planning and Standards, Health and Environmental Impacts
Division. Research Triangle Park, NC. U.S. EPA. EPA-452/R-10-004. July 2010.
Available at: https://nepis.epa.gov/Exe/ZvPURL.cgi?Dockev=P100FQ5D.txt.

U.S. EPA (2011). Policy Assessment for the Review of the Particulate Matter National Ambient
Air Quality Standards. Office of Air Quality Planning and Standards, Health and
Environmental Impacts Division. Research Triangle Park, NC. U.S. EPA. EPA-452/R-
11-003. April 2011. Available at:

https://nepis.epa. gov/Exe/ZvPURL.cgi?Dockev=P 100AUMY.txt.

U.S. EPA (2016). Integrated review plan for the national ambient air quality standards for

particulate matter. Office of Air Quality Planning and Standards. Research Triangle Park,
NC. U.S. EPA. EPA-452/R-16-005. December 2016. Available at:
https://www3.epa.gov/ttn/naaqs/standards/pm/data/201612-final-integrated-review-
plan.pdf.

U.S. EPA (2019). Integrated Science Assessment (ISA) for Particulate Matter (Final Report).
U.S. Environmental Protection Agency, Office of Research and Development, National
Center for Environmental Assessment. Washington, DC. U.S. EPA. EPA/600/R-19/188.
December 2019. Available at: https://www.epa.gov/naaqs/particulate-matter-pm-
standards-integrated-science-assessments-current-review.

U.S. EPA (2020). Policy Assessment for the Review of the National Ambient Air Quality

https://www.epa.gov/svstem/files/documents/2021-10/final-policv-assessment-for-the-
review-of-the-pm-naaqs-01-2020.pdf.

U.S. EPA (2022). Supplement to the 2019 Integrated Science Assessment for Particulate Matter
(Final Report). U.S. Environmental Protection Agency, Office of Research and
Development, Center for Public Health and Environmental Assessment. Research
Triangle Park, NC. U.S. EPA. EPA/600/R 22/028. May 2022. Available at:
https://www.epa.gov/naaqs/particulate-matter-pm-standards-integrated-science-
assessments-current-review.

1-21

-------
2 PM AIR QUALITY

This chapter provides an overview of recent ambient air quality with respect to PM. It
summarizes information on the distribution of particle size in ambient air, including discussions
about size fractions and components (section 2.1), ambient monitoring of PM in the U.S. (section
2.2), ambient concentrations of PM in the U.S. (section 2.3), and background PM (section 2.4).

2.1 DISTRIBUTION OF PARTICLE SIZE IN AMBIENT AIR

In ambient air, PM is a mixture of substances suspended as small liquid and/or solid
particles. Particle size is an important consideration for PM, as distinct health and welfare effects
have been linked with exposures to particles of different sizes. Particles in the atmosphere range
in size from less than 0.01 to more than 10 micrometers (|im) in diameter (U.S. EPA, 2019a,
section 2.2). When describing PM, subscripts are used to denote the aerodynamic diameter1 of
the particle size range in micrometers (jam) of 50% cut points of sampling devices. The EPA
defines PM2.5, also referred to as fine particles, as particles with aerodynamic diameters
generally less than or equal to 2.5 [^m. The size range for PM10-2.5, also called coarse or thoracic
coarse particles, includes those particles with aerodynamic diameters generally greater than 2.5
[j,m and less than or equal to 10 [j,m. PM10, which is comprised of both fine and coarse fractions,
includes those particles with aerodynamic diameters generally less than or equal to 10 [j,m.

Figure 2-1 provides perspective on these particle size fractions. In addition, ultrafine particles
(UFP) are often defined as particles with a diameter of less than 0.1 [j,m based on physical size,
thermal diffusivity or electrical mobility (U.S. EPA, 2019a, section 2.2).

1 Aerodynamic diameter is the size of a sphere of unit density (i.e., 1 g/cm3) that has the same terminal settling
velocity as the particle of interest (U.S. EPA, 2018; U.S. EPA, 2019a, section 4.1.1).

2-1

-------
90 (im (microns) in diameter
FINE BEACH SAND

WPM10

Dust, pollen, mold, etc.
<10|ilT) (microns) in diameter

C PM2.5

Combustion particles, organic

compounds, metals, etc.
<2.5 )J.m (microns) in diameter

Figure 2-1. Comparisons of PM2.5 and PM10 diameters to human hair and beach sand.

(Adapted from: https: n inr. epcLgov/pm-polhition/particulate-matter-pm-basics)

Atmospheric distributions of particle size generally exhibit distinct modes that roughly
align with the PM size fractions defined above. The nucleation mode is made up of freshly
generated particles, formed either during combusti on or by atmospheric reactions of precursor
gases. The nucleation mode is especially prominent near sources like heavy traffic, industrial
emissions, biomass burning, or cooking (Vu et al., 2015). While nucleation mode particles are
only a minor contributor to overall ambient PM mass and surface area, they are the main
contributors to ambient particle number (U.S. EPA, 2019a, section 2.2). By number, most
nucleation mode particles fall into the LFP size range, though some fraction of the nucleation
mode number distribution can extend above 0.1 |j,m in diameter. Nucleation mode particles can
grow rapidly through coagulation or uptake of gases by particle surfaces, giving rise to the
accumulation mode. The accumulation mode is typically the predominant contributor to PM2.5
mass and surface area, though only a minor contributor to particle number (U.S. EPA, 2019a,
section 2.2). PM2.5 sampling methods measure most of the accumulation mode mass, although a
small fraction of particles that make up the accumulation mode are greater than 2.5 (am in
diameter. Coarse mode particles are formed by mechanical generation, and through processes
like dust resuspension and sea spray formation (Whitby et al,, 1972). Most coarse mode mass is
captured by PM10-2.5 sampling, but small fractions of coarse mode mass can be smaller than 2.5
urn or greater than 10 (j,m in diameter (U.S. EPA, 2019a, section 2.2).

Most particles are found in the lower troposphere, where they can have residence times
ranging from a few hours to weeks. Particles are removed from the atmosphere by wet

2-2

-------
deposition, such as when they are carried by rain or snow, or by dry deposition, by gravitational
settling or surface collision. Atmospheric lifetimes are generally longest for PM2.5, which often
remains in the atmosphere for days to weeks (U.S. EPA, 2019a, Table 2-1) before being removed
by wet or dry deposition. In contrast, atmospheric lifetimes for UFP and PM10-2.5 are shorter.
Within hours, UFP can undergo coagulation and condensation that lead to formation of larger
particles in the accumulation mode, or can be removed from the atmosphere by evaporation,
deposition, or reactions with other atmospheric components. PM10-2.5 are also generally removed
from the atmosphere within hours, through wet or dry deposition (U.S. EPA, 2019a, Table 2-1).

2.1.1 Sources of PM Emissions

PM is composed of both primary (directly emitted particles) and secondary chemical
components. Primary PM is derived from direct particle emissions from specific PM sources
while secondary PM originates from gas-phase chemical compounds present in the atmosphere
that have participated in new particle formation or condensed onto existing particles (U.S. EPA,
2019a, section 2.3). Primary particles, and gas-phase compounds contributing to secondary
formation PM, are emitted from both anthropogenic and natural sources.

Anthropogenic sources of PM include both stationary and mobile sources. Stationary
sources include fuel combustion for electricity production and other purposes, industrial
processes, agricultural activities, and road and building construction and demolition. Mobile
sources of PM include diesel- and gasoline-powered highway vehicles and other engine-driven
sources (e.g., ships, aircraft, and construction and agricultural equipment). Both stationary and
mobile sources directly emit primary PM to ambient air, along with secondary PM precursors
(e.g., SO2) that contribute to the secondary formation of PM in the atmosphere (U.S. EPA,
2019a, section 2.3, Table 2-2).

Natural sources of PM include dust from the wind erosion of natural surfaces, sea salt,
wild fires, primary biological aerosol particles (PBAP) such as bacteria and pollen, oxidation of
biogenic hydrocarbons such as isoprene and terpenes to produce secondary organic aerosol
(SOA), and geogenic sources such as sulfate formed from volcanic production of SO2 (U.S.
EPA, 2009, section 3.3, Table 3-2). Contributions of natural emission sources to PM2.5
concentrations can be interconnected with anthropogenic emissions through atmospheric
chemistry, such as the modulation of biogenic SOA production by anthropogenic NOx and SO2
emissions (Budisulistiorini et al., 2015, Carlton et al., 2010, Carlton et al., 2018).

Generally, the sources of PM for different size fractions vary. While PM2.5 in ambient air
is largely emitted directly by sources such as those described above or through secondary PM
formation in the atmosphere, PM10-2.5 is almost entirely from primary sources (i.e., directly
emitted) and is produced by surface abrasion or by suspension of sea spray or biological

2-3

-------
materials such as microorganisms, pollen, and plant and insect debris (U.S. EPA, 2019a, section
2.3.2.1).

In sections 2.1.1.1 and 2.1.1.2 below, we describe the most recently available information
on sources contributing to PM2.5 and PM10 emissions into ambient air, respectively, based on the
2017 National Emissions Inventory (NEI).2 In section 2.1.1.3, we describe information on
sources contributing to emissions of PM components and precursor gases, with a focus on the
2017 NEI. Section 2.3.1 discusses emission trends and identifies the sectors that have
experienced the most change in direct PM and precursor emissions from 1990 to 2017. It should
be noted that major decreases have been observed in NOx and SO2 emissions over this time, with
continued reductions observed from the 2014 NEI to the 2017 NEI. For a more detailed review
of the changes in PM and PM precursor emissions from the 2014 NEI to the 2017 NEI, please
refer to the 2017 NEI Technical Support Document (U.S. EPA, 2021).

2.1.1.1 Sources Contributing to Primary PM2.5 Emissions

The National Emissions Inventory (NEI) is a comprehensive and detailed estimate of air
emissions of criteria pollutants, criteria pollutant precursors, and hazardous air pollutants from a
comprehensive set of air emissions sources, including point sources (e.g., electric generating
units, boilers, etc.), nonpoint (or area) sources (e.g., oil & gas, residential wood combustion, and
many other dispersed sources), mobiles sources, and events (large fires). There are over 3,000
sources for which the NEI is developed. The NEI is released every three years based primarily
upon data provided by state, local, and tribal air agencies for sources in their jurisdictions and
supplemented by data developed by the EPA. The NEI is built using the Emissions Inventory
System (EIS) first to collect the data from state, local, and tribal air agencies and then to blend
that data with other data sources.

Based on the 2017 NEI, approximately 5.7 million tons/year of PM2.5 were estimated to
be directly emitted to the atmosphere from a number of source sectors in the U.S. This total
excludes sources that are not a part of the NEI (e.g., windblown dust, geogenic sources). As
shown in Figure 2-2, nearly half of the total primary PM2.5 emissions nationally are contributed
by the dust and fire sectors together. Dust includes agricultural, construction, and road dust. Of
these, agricultural dust and road dust in sum make the greatest contributions to PM2.5 emissions
nationally. Fires include wildfires, prescribed fires, and agricultural fires, with wildfires and
prescribed fires accounting for most of the fire-related primary PM2.5 emissions nationally (U.S.

2 These sections do not provide a comprehensive list of all sources, nor do they provide estimates of emission rates
or emission factors for all source categories. Individual subsectors of source types were aggregated up to a sector
level as used in Figure 2-2 and Figure 2-4. More information about the sectors and subsectors can be found as a
part of the 2017 NEI (U.S. EPA, 2021).

2-4

-------
EPA, 2019a, section 2.3.1.1). Other lesser-contributing anthropogenic sources ofPlVh.s
emissions nationally include stationary fuel combustion and agriculture sources.

Mobile Sources
5%

Miscellaneous
6%

Agriculture

14%

Industrial
Processes
5%

Stationary Fuel
Combustion
11% |

Figure 2-2. Percent contribution of PM2.5 national emissions by source sectors. (Source:
2017 NEI)

The relative contributions of specific sources to annual emissions of primary PM2.5 can
vary from location to location, with a notable difference in contributions of sources of PM2.5
emissions in urban areas compared to national emissions. For example, the 2019 ISA illustrates
this variation of primary PM2.5 emissions with data from five urban counties in the U.S. (U.S.
EPA, 2019a, Figure 2-3).3 Across the majority of these urban areas, the largest PIVh.s-emitting
sectors are mobile sources and fuel combustion. This is in contrast to fires, which account for the
largest fraction of primary emissions nationally but make much smaller contributions in many
urban counties (U.S. EPA, 2019a, section 2.3.1.2, Figure 2-3). While primary PM2.5 from mobile
sources are a dominant contributor in some urban areas, accounting for an estimated 13 to 30%
of the total primary PM2.5 emissions, mobile sources contribute only about 5% to total primary
PM2.5 emissions nationally as shown in Figure 2-2.

Another way to examine the emissions data shown in Figure 2-2 is by county. Figure 2-3
presents county-based total PM2.5 emissions divided by the area of the county to normalize for
differences in county size. This "emissions density" map highlights regions of the country with

3 The five counties included in the 2019 ISA analysis include Queens County, NY, Philadelphia County, PA, Los
Angeles County, CA, Sacramento County, CA, and Maricopa County (Phoenix), AZ (U.S. EPA, 2019a, section
2.3.1.2).

2-5

-------
the highest total PM2.5 emissions by county accounting for county size. While Figure 2-3 shows
total PM2.5 emissions, different sectors will contribute at different levels across the country.

Tons Per Sq IVIi

0.8316- 1.2793

0 - 0.8063

Figure 2-3. 2017 NEI PM2.5 Emissions Density Map, tons per square mile

2.1.1.2 Sources Contributing to Primary PM10 Emissions

Although the NEI does not estimate emissions of PM10-2.5 (coarse PM) specifically,
estimates of PM10 emissions can provide insight into sources of coarse particles. Thus, the
discussion below focuses on PM10 emissions. The relative contributions of key sources to
national PM10 emissions, based on the 2017 NEI, are shown in Figure 2-4. Total PM10 emissions
are estimated to be about 17 million tons. National emissions of PM10 are dominated by dust and
agriculture, contributing a combined 70% of the total emissions. Current NEI estimates of dust
emissions across the U.S. are based on limited emissions profile and activity information. For a
number of reasons, quantification of dust emissions is highly uncertain. Much like wildfires, dust
emissions are common but intermittent emissions sources. Additionally, the suspension and
resuspension of dust is difficult to quantify. Moreover, some dust particles in the PM10-2.5 size
range are also transported internationally and are considered as a part of the background
component of PM as opposed to a primary emission of coarse PM (U.S. EPA, 2019a, section
2.3.3).

As with PM2.5, the relative contributions of sources to total PM10 emissions varies from
location to location (e.g., depending on local climate, geography, degree of urbanization, etc.).
However, unlike PM2.5, the sectors included in Figure 2-4 are expected to be among the most
important contributors to coarse PM emissions at both the national and more regional levels,
particularly given the sources of the particles in these source categories (e.g., mineral dust,
primary biological aerosols (including pollen), sea spray). As noted previously, the NEI does not
include sources such as pollen, sea spray, windblown dust, or geogenic sources, though those
sources also likely contribute to PM10 emissions. Figure 2-4 shows the national contributions to

2-6

-------
PMio emissions from particular source sectors and Figure 2-5 exhibits the corresponding
emissions density map for PMio.

Figure 2-4. Percent contribution of PMio emissions by national source sectors. (Source:
2017 NEI)

10.0785 - 528 8188

6.5071 - 9.7905

4.6178-6.3843

2.8936-4.5438

Figure 2-5. 2017 NEI PMio Emissions Density Map, tons per square mile

2.1.1.3 Sources Contributing to Emissions of PM Components and Precursor Gases
Understanding the components of PM is particularly important for providing insight into
which sources contribute to PM mass, as well as to better understand the health and welfare
effects of particles. Major components of PM2.5 mass include sulfate (SO42"), nitrate (NO3"),

Tons Per Sq Mi

0-2.8108

2-7

-------
elemental or black carbon (EC or BC), organic carbon (OC), and crustal materials. Some of these
PM components are emitted directly to the air (e.g., EC/BC) while others are formed secondarily
through reactions by gaseous precursors (e.g., sulfate, nitrate). The following sections
specifically discuss the sources that contribute to the specific PM2.5 components, including
particulate carbon (section 2.1.1.3.1) and precursor gases (section 2.1.1.3.2).

2.1.1.3.1 Sources Contributing to Emissions of Particulate Carbon

Of the directly emitted components of PM2.5, emissions of elemental (or black) carbon
and organic carbon often make up the largest percentage of directly emitted PM2.5 mass. Figure
2-6 illustrates the sources that contribute to national emissions of elemental and organic carbon
based on the 2017 NEI. The top panel of Figure 2-6 shows that fires account for most (i.e., 63%)
of the 1.8 million tons of particulate OC emissions estimated in the 2017 NEI, while the bottom
panel of Figure 2-6 shows that fires and mobile sources (mostly diesel sources) contribute 71%
of the estimated ~ 284,000 tons of particulate EC in the 2017 NEI. It should be noted that the
fraction of EC to PM2.5 was lower in the 2017 NEI compared to the 2014 NEI, owing to a
significantly lower contribution of EC from fires in the 2017 NEI compared to previous NEIs.
This change in the EC fraction resulted from an in-house research program to investigate the
PM2.5 chemical composition of the emissions from fires burning different fuels and in different
combustion phases. It should be noted that the OC contributions on a percentage basis increased
in accordance with the EC decreases. While these results have not yet been directly published,
this information has been acknowledged and used in other EPA analyses (Kelly et al., 2019b,
Figure 13).

2-8

-------
Mobile Sources
4%

I

Miscellaneous



Agriculture

Dust

9%



5%

^ 3%

Industrial
Processes
2%

Stationary Fuel
Combustion
14%

Elemental Carbon

Miscellaneous
9%

Mobile Sources
42%

Industrial

Processes [
2%

Agriculture
4%

Stationary Fuel
Combustion
13%

Figure 2-6. Percent contribution to organic carbon (top panel) and elemental carbon
(bottom panel) national emissions by source sectors. (Source: 2017 NEI)

Figure 2-7 shows the emissions density map for elemental carbon. This map illustrates
that the EC emissions signals are strong in the Southeast U.S, the central region of the U.S. (i.e.,
Kansas and Oklahoma), and parts of the West and Northwest U.S., where fires make substantial
contributions to PM2.5. In addition, areas where diesel off-road and on-road sources are a large
part of the emissions mix also stand out (urban and highway corridors). The OC density map (not
shown) shows the highest emissions density in locations with substantial biomass burning
activity, consistent with most of the OC emissions coming from fires (Figure 2-6).

2-9

-------
¦F? -I *14 . j ;

Tons Per Sq Mi

0.1678-18.9127

0.0984 - 0.1609

0.0662 - 0.0965

0.0401 - 0.0650

«»

0 - 0.0391

Figure 2-7. 2017 NEI Elemental Carbon Emissions Density Map, tons per square mile.

2.1.1.3.2 Sources Contributing to Emissions of Precursor Gases

As discussed further in the 2019 ISA (U.S. EPA, 2019a, section 2.3.2.1), secondary PM
is formed in the atmosphere by photochemical oxidation reactions of both inorganic and organic
gas-phase precursors. Precursor gases include SO2, NOx, ammonia, and volatile organic
compound (VOC) gases of anthropogenic or natural origin (U.S. EPA, 2019a, section 2.3.2.1).
Anthropogenic SO2 and NOx are the predominant precursor gases in the formation of secondary
PM2.5 sulfate and nitrate, and ammonia is the gas-phase precursor for PM2.5 ammonium. PM2.5
ammonium formation is enhanced by particle acidity resulting from sulfuric acid and nitric acid
condensation onto particles. Atmospheric oxidation of VOCs, both anthropogenic and biogenic,
is an important source of organic aerosols, particularly in summer. The semi-volatile and non-
volatile products of VOC oxidation reactions can condense onto existing particles or can form
new particles (U.S. EPA, 2009, section 3.3.2; U.S. EPA, 2019a, section 2.3.2).

Emissions of each of the precursor gases noted above are estimated in the NEI and have
unique source signatures at the national level. Figure 2-8 illustrates the source contributions at
the national level for these PM2.5 precursor gases. As shown in Panel A in Figure 2-8, stationary
fuel combustion sources contribute nearly 70% of the estimated total of 2.8 million tons of
national SO2 national emissions. Within this source category, nearly all of the SO2 emitted to the
atmosphere comes from electricity generating units, or EGUs. Anthropogenic NOx emissions,
shown in panel B, are emitted by a range of combustion sources, including mobile sources (59%)
and stationary fuel combustion sources (25%). In the 2017 NEI, there is an estimated total of
10.3 million tons of NOx emitted. Of the total estimated 4.3 million tons of anthropogenic
ammonia (NH3) emissions shown in panel C of Figure 2-8, NH3 emissions are dominated by the
agriculture source categories. In these categories, NH3 is predominantly emitted by livestock
waste from animal husbandry operations (56%) and fertilizer application (25%). In urban areas,
on-road mobile sources may also contribute significantly to NH3 emissions (U.S. EPA, 2019a,

2-10

-------
Figure 2-3; Sun et al., 2014; U.S. EPA, 2020). Of the estimated 17.2 million tons of VOC
emissions from anthropogenic sources, fires (39%) and "miscellaneous" (22%)4 are the highest
contributors, followed by mobile sources (17%) and industrial processes (18%), as shown in
Figure 2-8 panel D. It should be noted that as these traditional combustion sources of VOCs are
reduced by regulations and controls, new non-combustion sources, such as volatile chemical
products (solvents) are emerging as key contributors to anthropogenic VOC totals in some parts
of the country, and particularly in urban corridors. In addition, biogenic sources (not shown in
Figure 2-8) are significant contributors to both VOC and NOx emissions.

Figure 2-8. Percent contribution to sulfur dioxide (panel A), oxides of nitrogen (panel B),
ammonia (panel C), and volatile organic compounds (panel D) national emissions by
source sectors. (Source: 2017 NEI). All graphics only show anthropogenic contributions.

Figure 2-9 to Figure 2-12 below show the emissions density maps corresponding to each
of the PM2.5 precursors included in Figure 2-8.

4 The "miscellaneous" category includes such tilings as solvents, commercial cooking and waste disposal.

2-11

-------
* OoenStreetV

Tons Per Sq Mi

0.6500 - 284.7683

0.1740 - 0.5923

0.0802 -0 1661

0.0370 - 0.0774

0 - 0.0348

Figure 2-9. SO2 Emissions Density Map, tons per square mile.

r L T "*1 llSKT

-	-f

Tons Per Sq Mi

7.2452 - 830.8265

3.8459 - 6.9474

2.4582 - 3.7804

1.5045-2.4218

0-1.4700

Figure 2-10. NOx Emissions Density Map, tons per square mile.

j,

% *1

•> -

«j5

V|A»

%

l£j

. ""."iktiirlv1 j? . .. «

V " «*i

¦•>«••• SS«»",|" v.-'-' - »/*if . '	V

'.kifa " "• * "r -&

M. •*?: *Y -

t

'v.

Tons Per Sq Mi

2.0750-138.U33
1.2897-2.0271

0.7747-1.2547

0.4020-0.7569

0-0.3784

Figure 2-11. NH3 Emissions Density Map, tons per square mile.

2-12

-------
Tons Per Sq Mi

29.8035 - 707.2287
18.5502 - 29.1950
10.4074 - 18.1336

5.8970 - 10.1220

0 - 5.S357

Figure 2-12. Anthropogenic (including wildfires) VOC Emissions Density Map, tons per
square mile.

2.1.1.3.3 Uncertainty in Emission Estimates

Accuracy in an emissions inventory reflects the extent to which the inventory represents
the actual emissions that occurred. Anthropogenic emissions of air pollutants result from a
variety of sources such as power plants, industrial sources, motor vehicles and agriculture. The
emissions from any individual source typically vary in both time and space. It is not practically
possible to monitor each of the emission sources individually and, therefore, emission
inventories necessarily contain assumptions, and must rely too on interpolation and extrapolation
from a limited set of sample data.

The NEI process is based on a "bottom up" approach to developing emission estimates.
This means that a combination of activity and an appropriate emissions factor is used to estimate
emissions for all processes, including accounting for controls as possible. For the thousands of
sources that make up the NEI, there is uncertainty in one or all of these factors. For some
sources, such as EGUs, direct emission measurements enable the emission factors to be more
certain than for sources without such direct measurements. For example, emission factors for
residential wood combustion are taken from information available in the literature, regardless of
its pedigree and direct applicability to the source in question. Many of these issues related to the
analysis of uncertainty in the NEI are discussed by Day et al. (2019).

Some of the uncertainty in air quality model outputs is likely due to uncertainty in
emission estimates. The EPA uses information from air quality models and feedback from
modelers and other stakeholders to help identify which sectors to prioritize for emissions data
methods improvements.

* f*'

«*'•

2-13

-------
2.2 AMBIENT PM MONITORING METHODS AND NETWORKS

To promote uniform enforcement of the air quality standards set forth under the CAA and
to achieve the degree of public health and welfare protection intended for the NAAQS, the EPA
established PM Federal Reference Methods (FRMs)5 for both PMio and PM2.5 (40 CFR
Appendix J and L to Part 50) and performance requirements for approval of Federal Equivalent
Methods (FEMs) (40 CFR Part 53). Amended following the 2006 and 2012 PM NAAQS
reviews, the current PM monitoring network relies on FRMs and automated continuous FEMs, in
part to support changes necessary for implementation of the revised PM standards. The
requirements for measuring ambient air quality and reporting ambient air quality data and related
information are the basis for 40 CFR Appendices A through E to Part 58.

The EPA and its partners at state, local, and tribal monitoring agencies manage and
operate the nation's ambient air monitoring networks. The EPA provides minimum monitoring
requirements for criteria pollutants and related monitoring (e.g., the Chemical Speciation
Network (CSN)), including identification of an FRM for criteria pollutants and guidance
documents to support implementation and operation of the networks. Monitoring agencies carry
out and perform ambient air monitoring in accordance with the EPA's requirements and
guidance as well as often meeting their own state monitoring needs that may go beyond the
minimum federal requirements. Data from the ambient air monitoring networks are available
from two national databases: 1) the Air Quality System (AQS) database, which is the EPA's
long-term repository of ambient air monitoring data and 2) the AirNow database, which provides
near real-time data used in public reporting and forecasting of the Air Quality Index (AQI).6

The EPA and monitoring agencies manage and operate robust national networks for both
PM10 and PM2.5, as these are the two measurement programs directly supporting the PM
NAAQS. PM10 measurements are based on gravimetric mass, while PM2.5 measurements include
gravimetric mass and chemical speciation. A smaller network of stations is operating and
reporting data for PM10-2.5 gravimetric mass and a few monitors are operated to support special
projects, including pilot studies, for continuous speciation and particle count data. Monitoring
networks and additional monitoring efforts for each of the various PM size fractions and for PM

5	FRMs provide the methodological basis for comparison to the NAAQS and also serve as the "gold standard" for

the comparison of other methods being reviewed for potential approval as equivalent methods. The EPA keeps a
complete list of designated reference and equivalent methods available on its Ambient Monitoring Technology
Information Center (AMTIC) website (https://www.epa.gov/amtic/air-monitoring-methods-criteria-pollutants').

6	The AQI is an index for reporting daily air quality and translates air quality data into numbers and colors to help

people understand how clean or polluted the air is, and what associated health effects might be a concern,
especially for ozone and particle pollution.

2-14

-------
composition are discussed below.7 Section 2.2.1 provides information on monitoring for total
suspended particulates (TSP), section 2.2.2 provides information on monitoring for PMio, section
2.2.3 provides information on monitoring PM2.5, section 2.2.4 provides information on
monitoring for PM10-2.5, and section 2.2.5 provides information on additional PM metrics. Figure
2-13 below illustrates the changes in PM monitoring stations reporting to the EPA's AQS
database by size fraction since 1970. Table 2-1 provides details on PM networks described in this
section. Station and site counts are based on data submitted to the EPA for calendar year 2020,
unless otherwise noted.

PM Monitoring Stations Reporting to EPA's AQS
database by Size Fraction, 1970 - 2020

	TSP — PMlOTotaO-lOumSTP	PM25-Local Conditions — PM10-2.5

Figure 2-13. PM Monitoring stations reporting to EPA's AQS database by PM size
fraction, 1970-2020.

Table 2-1. PM Monitoring Networks.

Network

Measurements

Operating
Stations

Collection
Frequency

Major Siting Types 1

Monitoring
Objectives

PM mass networks

PM10

Filter-based FRMs;
continuous FEMs

680

Typically 1:6 for
FRMs; hourly for
continuous FEM

Highest
concentration,
Population oriented,
Source impact

NAAQS andAQI

7 More information on ambient monitoring networks can be found at https://www.epa. gov/amtic

2-15

-------
PM2.5

Filter-based FRMs;
continuous FEMs

950

Typically 1:3 for
FRM; hourly for
continuous FEM

Highest
concentration,
Population oriented

NAAQS andAQI



Continuous pre-FEM

290

Hourly

PM10-2.5

Filter-based FRMs;
continuous FEMs;
IMPROVE

287

Typically 1:3 for
FRM and
IMPROVE; hourly
for continuous
FEM

Population oriented
in CBSAs; Welfare-
related impacts
and/or

General/background
& transport in rural
areas

Trends; Support
future reviews of
the NAAQS

TSP

TSP sampler

Lead (Pb) only
-104; Lead
and mass- 25

Typically 1:6

Highest
concentration,
Source impact

NAAQS for lead

Multi-pollutant networks (PM measurements at these stations are accounted for above)

Near-Road
Network

CO, NO2, and PM2.5

54 (note: there
are additional
N-R sites as
NO2 only)

Hourly for gases
and automated
PM2.5 FEMs; 1:3
for PM2.5 FRMs

Source impact

NAAQS andAQI

National Core
(NCore)

Gases - CO, SO2,
NO/NOy, and 03.
Particles - PM2.5,
(continuous mass,
filter mass, speciation
[CSN or IMPROVE])
PM10-2.5, (mass).

Basic meteorology
(WS, WD, Temp., RH).

78

(63 urban and
suburban; 15
rural)

Gases and

meteorology:

hourly.

Particles: Filter-
based -1:3 day,
Continuous -
hourly

Population oriented
in CBSAs;

General/background
& transport in rural
areas

Maximize
multipollutant
information for:
input to health
and atmospheric
studies; NAAQS
revisions;
validate AQ
models; assess
emission
reduction
programs

Chemical Speciation Networks and Measurements (some of these are collocated with NCore above)

Chemical
Speciation
Network (CSN)

Elements, ions, carbon

139

1:3 for Speciation
Trends Network
(STN) and NCore;
1:6 for

supplemental sites

Population oriented

Develop control

strategies,

Assess

progress/ trends

IMPROVE

Elements, ions,
carbon, PM2.5 mass,
and PM10 mass

148

(110 in base
network; 38
protocol sites)

1:3

Welfare-related
impacts

Base network -
assess visibility
impairment for
Regional Haze
program;
protocol sites
have the same
objectives as
CSN

Aethalometers

and other

commercially

available

continuous

carbon

Indicators of black
carbon and wood
smoke

72

Hourly

Population oriented
and source impact

Understanding
PM source
contributions

1 Site Types as described in Table D-1 of Appendix D to Part 58.

2-16

-------
2.2.1 Total Suspended Particulates (TSP) Sampling

The EPA first established NAAQS for PM in 1971, based on the original air quality
criteria document (DHEW, 1969). The reference method specified for determining attainment of
the original standards was the high-volume sampler, which collects PM up to a nominal size of
25 to 45 [j,m (referred to as total suspended particles or TSP). TSP was replaced by PMio as the
indicator for the PM NAAQS in the 1987 final rule (52 FR 24854, July 1, 1987). TSP sampling
remains in operation at a limited number of locations primarily to provide aerosol collection for
TSP lead (Pb) analysis as well as for instances where a state may continue to have state standards
for TSP. The size of the TSP network peaked in the mid-1970s when over 4,300 TSP samplers
were in operation. As of 2020, there were 104 TSP samplers still in operation as part of the Pb
monitoring program; of these, 25 also report TSP mass.

2.2.2 PMio Monitoring

To support the 1987 PMio NAAQS, the EPA and its state and local partners implemented
the first size-selective PM monitoring network in 1990 with the establishment of a PMio network
consisting of mainly high-volume samplers. The network design criteria emphasize monitoring at
middle8 and neighborhood9 scales to effectively characterize the emissions from both mobile and

8 For PMio, middle-scale is defined as follows: Much of the short-term public exposure to PMio is on this scale and

on the neighborhood scale. People moving through downtown areas or living near major roadways or stationary
sources, may encounter particulate pollution that would be adequately characterized by measurements of this
spatial scale. Middle scale PMio measurements can be appropriate for the evaluation of possible short-term
exposure public health effects. In many situations, monitoring sites that are representative of micro-scale or
middle-scale impacts are not unique and are representative of many similar situations. This can occur along traffic
corridors or other locations in a residential district. In this case, one location is representative of a neighborhood
of small-scale sites and is appropriate for evaluation of long-term or chronic effects. This scale also includes the
characteristic concentrations for other areas with dimensions of a few hundred meters such as the parking lot and
feeder streets associated with shopping centers, stadia, and office buildings. In the case of PMio, unpaved or
seldomly swept parking lots associated with these sources could be an important source in addition to the
vehicular emissions themselves.

9 For PMio, neighborhood scale is defined as follows: Measurements in this category represent conditions

throughout some reasonably homogeneous urban sub-region with dimensions of a few kilometers and of
generally more regular shape than the middle scale. Homogeneity refers to the particulate matter concentrations,
as well as the land use and land surface characteristics. In some cases, a location carefully chosen to provide
neighborhood scale data would represent not only the immediate neighborhood but also neighborhoods of the
same type in other parts of the city. Neighborhood scale PMio sites provide information about trends and
compliance with standards because they often represent conditions in areas where people commonly live and
work for extended periods. Neighborhood scale data could provide valuable information for developing, testing,
and revising models that describe the larger-scale concentration patterns, especially those models relying on
spatially smoothed emission fields for inputs. The neighborhood scale measurements could also be used for
neighborhood comparisons within or between cities.

2-17

-------
stationary sources, although not ruling out microscale10 monitoring in some instances (40 CFR
Part 58 Appendix D, 4.6 (b)). The PMio monitoring network peaked in size in 1995 with 1,665
stations reporting data.

In 2020, there were 680 PMio stations in operation to support comparison of the PMio
data to the NAAQS, trends, and reporting and forecasting of the AQI. Though the PMio network
is relatively stable, monitoring agencies may continue divesting of some of the PMio monitoring
stations where concentration levels are low relative to the NAAQS.

While the PMio network is national in scope, there are areas of the west, such as
California and Arizona, with substantially higher PMio station density than the rest of the
country. In the PMio mass network, 385 of the stations operate automated continuous mass
monitors approved as FEMs and 295 operate FRMs. About 30 of the PMio stations have
collocation with both continuous FEMs and FRMs. More than half of the PMio stations with
FRMs operate on a sample frequency of one in every sixth day, with about 55 stations operating
every third day and another 55 stations operating every day.

2.2.3 PM2.5 Monitoring

To support the 1997 PM2.5 NAAQS, the first PM standard with PM2.5 as an indicator, the
EPA and states implemented a PM2.5 network consisting of ambient air monitoring sites with
mass and/or chemical speciation measurements. Network operation began in 1999 with nearly
1,000 monitoring stations operating FRMs to measure fine particle mass. The PM2.5 monitoring
program remains one of the major ambient air monitoring programs operated across the country.

For most urban locations, PM2.5 monitors are sited at the neighborhood scale,11 where
PM2.5 concentrations are reasonably homogeneous throughout an entire urban sub-region. In each

i° for PMio, microscale is defined as follows: This scale would typify areas such as downtown street canyons, traffic
corridors, and fence line stationary source monitoring locations where the general public could be exposed to
maximum PMio concentrations. Microscale particulate matter sites should be located near inhabited buildings or
locations where the general public can be expected to be exposed to the concentration measured. Emissions from
stationary sources such as primary and secondary smelters, power plants, and other large industrial processes
may, under certain plume conditions, likewise result in high ground level concentrations at the microscale. In the
latter case, the microscale would represent an area impacted by the plume with dimensions extending up to
approximately 100 meters. Data collected at microscale sites provide information for evaluating and developing
hot spot control measures.

11 For PM2 5, neighborhood scale is defined as follows: Measurements in this category would represent conditions
throughout some reasonably homogeneous urban sub-region with dimensions of a few kilometers and of
generally more regular shape than the middle scale. Homogeneity refers to the particulate matter concentrations,
as well as the land use and land surface characteristics. Much of the PM2 5 exposures are expected to be associated
with this scale of measurement. In some cases, a location carefully chosen to provide neighborhood scale data
would represent the immediate neighborhood as well as neighborhoods of the same type in other parts of the city.
PM2 5 sites of this kind provide good information about trends and compliance with standards because they often
represent conditions in areas where people commonly live and work for periods comparable to those specified in
the NAAQS. In general, most PM2 5 monitoring in urban areas should have this scale.

2-18

-------
CBS A with a monitoring requirement, at least one PM2.5 monitoring station representing area-
wide air quality is to be sited in an area of expected maximum concentration. Only sites that
represent relatively unique microscale, localized hot-spot, or unique middle scale impact sites are
eligible for comparison to the 24-hour PM2.5 NAAQS.

There are three main components of the current PM2.5 monitoring program: FRMs, PM2.5
continuous mass monitors, and CSN samplers. The FRMs are primarily used for comparison to
the NAAQS, but also serve other important purposes such as developing trends and evaluating
the performance of PM2.5 continuous mass monitors. PM2.5 continuous mass monitors are
automated methods primarily used to support forecasting and reporting of the AQI, but are also
used for comparison to the NAAQS where approved as FEMs. The CSN and related Interagency
Monitoring of Protected Visual Environments (IMPROVE) network are used to provide
chemical composition of the aerosol which serve a variety of objectives. This section provides an
overview of each of these components of the PM2.5 monitoring program and of recent changes to
PM2.5 monitoring requirements.

2.2.3.1 Federal Reference Method and Continuous Monitors

As noted above, the PM2.5 monitoring network began operation in 1999 with nearly 1,000
monitoring stations operating FRMs. The PM2.5 FRM network peaked in operation in 2001 with
over 1,150 monitoring stations. In the PM2.5 network for 2020 there were 527 FRM filter-based
samplers that provide 24-hour PM2.5 mass concentration data. Of these operating FRMs, 68 are
providing daily PM2.5 data, 340 every third day, and 119 every sixth day.

As of 2020, there are 950 continuous PM2.5 mass monitors that provide hourly data on a
near real-time basis reporting across the country. A total of 660 of the PM2.5 continuous monitors
are FEMs and therefore used both for comparison with the NAAQS and to report the AQI.
Another 290 monitors not approved as FEMs are operated primarily to report the AQI. These
legacy PM2.5 continuous monitors were largely purchased prior to the availability of PM2.5
continuous FEMs.

The first method approved as a continuous PM2.5 FEM was the Met One BAM 1020. This
method, approved in 2008, accounts for just over a third of the operating PM2.5 continuous FEMs
in the country. The EPA has approved a total of 11 PM2.5 continuous methods as FEMs. Other
methods approved as continuous PM2.5 FEMs include beta attenuation from multiple instrument
manufacturers; optical methods such as the GRIMM and Teledyne T640; and methods
employing the Tapered Element Oscillating Microbalance (TEOM) with a Filter Dynamic
Measurement System (FDMS) manufactured by Thermo Fisher Scientific.

The quality of the data generated by PM2.5 FRMs and automated FEMs were analyzed for
years 2018-2020. Data quality terms for measurement uncertainty regularly assessed in the PM2.5

2-19

-------
monitoring program include precision and bias. Precision is calculated by comparing data from
collocated methods of the same make and model operated by the same monitoring organization.
Bias is calculated by comparing data from routinely operated FRMs or automated FEMs by the
monitoring organization and comparing that to data from reference method audit samplers
temporarily collocated and operated independently from the staff in the monitoring organization.
Goals for measurement uncertainty are defined in Appendix A to 40 CFR Part 58. They state
"Measurement Uncertainty for Automated and Manual PM2.5 Methods. The goal for acceptable
measurement uncertainty is defined for precision as an upper 90 percent confidence limit for the
coefficient of variation (CV) of 10 percent and ±10 percent for total bias." The most recent three-
year average estimate of national aggregate PM2.5 FRM precision is 7.6% and bias is -7.5%.

Automated PM2.5 FEMs include a wide variety of approved methods which can have
different measurement principles. Data aggregated across all automated FEMs for years 2018-
2020 result in a collocated precision of 12.8%. Bias can be calculated from the reference method
audit program and by comparing continuous FEMs to collocated FRMs run by the monitoring
agency. The 2018-2020 reference method audit program had a bias of -1.7% with a sample size
of 573 audits across all continuous FEMs. Continuous FEMs compared to collocated monitoring
agency FRMs were biased higher by 11.5% with a large sample size of 85,539 collocated pairs
for 2018-2020 (all cases where both the FRM and continuous FEM are at or above 3.0 (J,g/m3).
When evaluating automated FEMs as individual methods, only two of the seven methods with
available collocated precision data met the measurement uncertainty goal and six of the eleven
methods met the bias goal. However, for collocated precision data and when considering a
requirement for approval of candidate FEMs: "Statistical analyses based on the DQO model
show that the precision of a candidate method is not, statistically, very important to annual
concentration averages used for NAAQS attainment decisions, but would be important for a
daily standard" (71 FR 2620, January 17, 2006) In summary, PM2.5 automated FEMs tend to
have higher collocated precision than FRMs and tend to have a positive bias relative to state and
local operated FRMs.

2.2.3.2 Chemical Speciation and IMPROVE Networks

Due to the complex nature of fine particles, the EPA and states implemented the CSN to
better understand the components of fine particle mass at selected locations across the country.
The CSN was first piloted at 13 sites in 2000, and after the pilot phase, the program continued
with deployment of the Speciation Trends Network (STN) later that year. The CSN ultimately
grew to 54 trends sites and peaked in operation in 2005 with 252 stations: the 54 trends stations
and nearly 200 supplemental stations. The original CSN program had multiple sampler
configurations including the Thermo Andersen RAAS, Met One SASS/SuperSASS, and URG

2-20

-------
MASS. During the 2000s, the EPA and states worked to align the network to one common
sampler for elements and ions, which was the Met One SASS/SuperSASS. In 2005, the CASAC
provided recommendations to the EPA for making changes to the CSN. These changes were
intended to improve data comparability with the rural IMPROVE carbon concentration data. To
accomplish this, the EPA replaced the existing carbon channel sampling and analysis methods
with a new modified IMPROVE version III module C sampler, the URG 3000N. Implementation
of the new carbon sampler and analysis was broken into three phases starting in May 2007
through October 2009.

In the 2020 PM2.5 CSN, long-term measurements are made at about 75 largely urban
locations comprised of either the STN or the National Core (NCore) network.12 NCore is a
multipollutant network measuring particles, gases, and basic meteorology that has been in formal
operation since January 1, 2011. Particle measurements made at NCore include PM2.5 filter-based
mass, which is largely the FRM, except in some rural locations that utilize the IMPROVE
program PM2.5 mass filter-based measurement; PM2.5 speciation using either the CSN program or
IMPROVE program; and PM10-2.5 mass utilizing an FRM, FEM or IMPROVE for some of the
rural locations. As of 2020, the NCore network includes a total of 78 stations of which 63 are in
urban or suburban stations designed to provide representative population exposure and another
15 rural stations designed to provide background and transport information. The NCore network
is deployed in all 50 States, DC, and Puerto Rico with at least one station in each state and two or
more stations in larger population states (California, Florida, Illinois, Michigan, New York,

North Carolina, Ohio, Pennsylvania, and Texas).

Both the STN and NCore networks are intended to remain in operation indefinitely. The
CSN measurements at NCore and STN stations operate every third day. Six of these stations
have collocated sets of CSN samplers where the collocated samplers operate every sixth day to
provide precision calculations of each chemical species measured. Another approximately 70
CSN stations, known as supplemental sites, are intended to be potentially less permanent
locations used to support State Implementation Plan (SIP) development and other monitoring
objectives.13 Supplemental CSN stations typically operate every sixth day. In January 2015, 38

12 In most cases where a city has an STN station, it is located at the same site as the NCore station. In a few cases, a
city may have an STN station located at a different location than the NCore station.

13 See https://www.epa.gov/amtic/chemical-speciation-netwoik-csn for more information on the PM2 5 speciation
monitoring program.

2-21

-------
supplemental CSN stations that are largely located in the eastern half of the country stopped
operations to ensure a sustainable CSN network moving forward.14

Specific components of fine particles are also measured through the IMPROVE
monitoring program,15 which supports regional haze characterization and tracks changes in
visibility in Class I areas16 as well as many other rural and some urban areas. As of 2018, the
IMPROVE network includes 110 monitoring locations that are part of the base network
supporting regional haze and another 38 locations operated as IMPROVE protocol sites where a
monitoring agency has requested participation in the program. These IMPROVE protocol sites
operate the same way as the IMPROVE program, but they may serve several monitoring
objectives (i.e., the same objectives as the CSN) and are not explicitly tied to the Regional Haze
Program. Samplers at IMPROVE stations operate every third day. In January 2016, eight
IMPROVE protocol stations stopped operating to ensure a sustainable IMPROVE program
moving forward. Details on the process and outcomes of the CSN supplemental and IMPROVE
protocol assessments used to identify sites that would no longer be funded are available on a
website.17 Together, the CSN and IMPROVE data provide chemical species information for fine
particles that are critical for use in health and epidemiologic studies to help inform reviews of the
primary PM NAAQS. CSN and IMPROVE data can also be used to better understand visibility
through calculation of light extinction using the IMPROVE algorithm18 to support reviews of the
secondary PM NAAQS.

14 Based on assessments of the CSN network and IMPROVE protocol sites, monitoring resources were redistributed
to focus on new or high priorities. More information on the CSN and IMPROVE protocol assessments is
available at https://www.epa.gov/amtic/csn-and-improve-protocol-network-assessment.

15 Recognizing the importance of visual air quality. Congress included legislation in the 1977 Clean Air Act to
prevent future and remedy existing visibility impairment in Class I areas. To aid the implementation of this
legislation the IMPROVE program was initiated in 1985 and substantially expanded in 2000-2003. This program
implemented an extensive long-term monitoring program to establish the current visibility conditions, track
changes in visibility and determine causal mechanism for the visibility impairment in the National Parks and
Wilderness Areas. For more information see https://vista.cira.colostate.edu/Improve/.

16 See Regional Haze rule text at 50 CFR Part 51.308(d)(4) and (f)(6) (pasted below) lists SIP requirements, one of
which is a "Monitoring Strategy...". This part of the rule doesn't necessarily require IMPROVE, rather it simply
assures states that IMPROVE will meet this requirement. Specifically, this text reads: "(6) Monitoring strategy
and other implementation plan requirements. The State must submit with the implementation plan a monitoring
strategy for measuring, characterizing, and reporting of regional haze visibility impairment that is representative
of all mandatory Class I Federal areas within the State. Compliance with this requirement may be met through
participation in the Interagency Monitoring of Protected Visual Environments network."

17 See the CSN and IMPROVE Protocol Network Assessment Website at: https://www.epa.gov/amtic/csn-and-
improve-protocol-network-assessment

18 The IMPROVE algorithm is an equation to estimate light extinction based on the measured concentration of
several PM components and is used to track visibility progress in the Regional Haze Rule. More information
about the IMPROVE algorithm is at available at: http://vista.cira.colostate.edu/Improve/the-improve-algorithm.

2-22

-------
The quality of the data generated by the PM2.5 speciation networks (CSN and IMPROVE)
is assessed regularly, using a variety of metrics. Overall network precision, including
uncertainties associated with both field operations and laboratory analyses, is assessed using the
subset of sites with collocated samplers. Fractional uncertainty is one metric that both speciation
networks regularly calculates using collocated data pairs above the MDL and reflects the overall
percent uncertainty for the measurements. For CSN data collected between June 2016 and
December 2019, the fractional uncertainties range from 5.6% for sulfate to 36.4% for chlorine.19
For IMPROVE data collected in 2016 and 2017, the fractional uncertainties range from 2% for
sulfur and sulfate to 27% for phosphorus.20 In general, uncertainties are higher for species with
concentrations near the detection limit. Bias for the speciation networks can be assessed using
reports from interlaboratory comparisons.21

2.2.3.3 Recent Changes to PM2.5 Monitoring Requirements

Key changes made to the EPA's monitoring requirements as a result of the 2012 PM
NAAQS review included the addition of PM2.5 monitoring at near-road locations in core-based
statistical areas (CBSAs) over 1 million in population; the clarification of terms used in siting of
PM2.5 monitors and their applicability to the NAAQS; and the provision of flexibility on data
uses to monitoring agencies where their PM2.5 continuous monitors are not providing data that
meets the performance criteria used to approve the continuous method as an FEM. The addition
of PM2.5 monitoring at near-road locations was phased in from 2015 to 2017. On January 1,
2015, 22 CBSAs with a population of 2.5 million or more were required to have a PM2.5 FRM or
FEM operating at a near-road monitoring station. On January 1, 2017, 30 CBSAs with a
population between 1 million and 2.5 million were required to have a PM2.5 FRM or FEM
operating are a near-road monitoring station.

The terms clarified as a part of the 2012 rulemaking ensure consistency with all other
NAAQS and long-standing definitions used by the EPA (78 FR 3234, January 15, 2013). The
flexibility provided to monitoring agencies ensures that the incentives of utilizing PM2.5
continuous monitors (e.g., efficiencies in operation and availability of hourly data in near-real
time) are realized without having potentially poor performing data being used in situations where
the data is not applicable to the NAAQS (78 FR 3241, January 15, 2013).

19	See https://airqualitv.ucdavis.edu/sites/g/files/dgvnskl671/files/inline-
files/CSN AnnualReport 2016Data 03.06.2019 FINAL APPROVED.pdf

20	See http://vista.cira.colostate.edu/improve/wp-content/uploads/2019/11/IMPROVE OAReport 11.15.2019.pdf

21	See https://www.epa.gov/amtic/chemical-speciation-network-interlaboratorv-performance-evaluation-comparison-
results

2-23

-------
2.2.4 PMio-2.5 Monitoring

In the 2006 PM NAAQS review, the EPA promulgated a new FRM for the measurement
of PM10-2.5 mass in ambient air. Although the standard for coarse particles uses a PMio indicator,
a new FRM for PM10-2.5 mass was developed to provide a basis for approving FEMs and to
promote the gathering of scientific data to support future reviews of the PM NAAQS. The PM10-
2.5 FRM (or approved FEMs, where available) was implemented at required NCore stations by
January 1, 2011. In addition to NCore, there are other collocated PM10 and PM2.5 low-volume
FRMs operating across the country that are essentially providing the PM10-2.5 FRM measurement
by the difference method.

PM10-2.5 measurements are currently performed across the country at NCore stations,
IMPROVE monitoring stations, and at a few additional locations where state or local agencies
choose to operate a PM10-2.5 method. For urban NCore stations and other State and Local Air
Monitoring Stations (SLAMS) the method employed is either a PM10-2.5 FRM, which is
performed using a low-volume PM10 FRM collocated with a low volume PM2.5 FRM of the same
make and model, or FEMs for PM10-2.5, including filter-based dichotomous methods and
continuous methods of which several makes and models are approved. Filter-based PM10-2.5
measurements at NCore (i.e., the FRM or dichotomous filter-based FEM) operate every third
day, while continuous methods have data available every hour of every day. PM10-2.5 filter-based
methods at other SLAMS typically operate every third or sixth day. For IMPROVE, which is
largely a rural network, PM10-2.5 measurements are made with two sample channels; one each for
PM10 and PM2.5. All IMPROVE program samplers operate every third day. All together there
were 287 stations in 2020 where PM10-2.5 data were being reported to the AQS database.

There is no operating chemical speciation network for characterizing the specific
components of coarse particles. In 2015, Washington University at St. Louis, under contract to
the EPA, reported on a coarse particle speciation pilot study with several objectives aimed at
addressing this issue, such as evaluating a coarse particle species analyte list and evaluating
sampling and analytical methods (U.S. EPA, 2015). The coarse particle speciation pilot study
provides useful information for any organization wishing to pursue coarse particle speciation.

2.2.5 Additional PM Measurements and Metrics

There are additional PM measurements and metrics made at a much smaller number of
stations. These measurements may be associated with special projects or are complementary
measurements to other networks where the monitoring agency has prioritized having the
measurements. None of these measurements are required by regulation. They include PM
measurements such as particle counts, continuous carbon, and continuous sulfate.

2-24

-------
The EPA and state and local agencies have also been working together to pilot additional
PM methods at near-road monitoring stations that may be of interest to data users. These
methods include such techniques as particle counters, particle size distribution, and black carbon
by aethalometer. These methods and their rationale for use at near-road monitoring stations are
described in a Technical Assistance Document (TAD) on NO2 near-road monitoring (U.S. EPA,
2012, section 16).

Aethalometer measurements of the concentration of optically absorbing particles have
been submitted to AQS for many years. Data uses include characterizing black carbon and wood
smoke. Ambient air monitoring stations that may have aethalometers include some of the near-
road monitoring stations and National Air Toxics Trends Stations (NATTS). In 2020, data from
72 monitoring sites across the country were reported from aethalometers and other related
commercially available continuous carbon analyzers. While aethalometer and related continuous
carbon data are available at high time resolutions (e.g., 5-minute data), they are typically
reported to the AQS database in 1-hour periods.

Continuous elemental and organic carbon data were monitored at select locations
participating in a pilot of the Sunset EC/OC analyzer as well as a few additional sites that were
already operating before the EPA initiated the pilot study.22 The Sunset EC/OC analyzer
provides high-time-resolution carbon data, typically every hour, but in some remote locations the
instrument is programmed to run every two hours to ensure collection of enough aerosol. The
data from the Sunset EC/OC analyzer was compared to filter-based carbon methods from the
carbon channel of the CSN program. The Sunset EC/OC analyzer was operated at each of the
study sites for at least three years. Results from this pilot study are available in an EPA report
(U.S. EPA, 2019b). A key finding from the study suggests that when the Sunset instrument was
working well, OC and optical EC were comparable to CSN OC and EC; however, the time and
resources needed to keep a Sunset analyzer operational did not merit replacement of CSN OC
and EC measurements.

As of 2020, continuous sulfate is measured at two remaining monitoring sites, one each
in Maine and North Carolina. Several other stations have historical data but are no longer
monitoring continuous sulfate. Discontinued monitoring efforts for continuous sulfate is likely an
outcome of the significantly lower sulfate concentrations throughout the east where these
methods were operated. The continuous sulfate analyzer provides hourly data and these data can
be readily compared to 24-hour sulfate data which are collected from the ion channel in both the
CSN and IMPROVE programs.

22 The six sites that participated in the study were Washington, DC; Chicago, IL; St. Louis, MO; Houston, TX; Las
Vegas, NV; and Los Angeles, CA.

2-25

-------
In addition, over the last few years, the EPA has investigated the use of several PM
sensor technologies as one of several areas of research intended to address the next generation of
air measurements. The investigation into air sensors is envisioned to work towards near real-time
or continuous measurement options that are smaller, cheaper, and more portable than traditional
FRM or FEM methods. These sensor devices have the potential to be used in several applications
such as identifying hotspots, informing network design, providing personal exposure monitoring,
supporting risk assessments, and providing background concentration data for permitting. The
EPA has hosted workshops and published several documents and peer-reviewed articles on this
work.23

2.3 AMBIENT AIR CONCENTRATIONS

This section summarizes available information on recent ambient PM concentrations.
Section 2.3.1 presents trends in emissions of PM and precursor gases, while section 2.3.2
presents trends in monitored ambient concentrations of PM in the U.S. Section 2.3.3 discusses
approaches for predicting ambient PM2.5 by hybrid modeling approaches.

2.3.1 Trends in Emissions of PM and Precursor Gases

Direct emissions of PM have remained relatively unchanged in recent years, while
emissions of some precursor gases have declined substantially.24 As illustrated in Figure 2-14,25
from 1990 to 2017, SO2 emissions have undergone the largest declines while NH3 emissions
have undergone the smallest change. Declining SO2 emissions during this time period
areprimarily a result of reductions at stationary sources such as EGUs, with substantial
reductions also from mobile sources (U.S. EPA, 2019a, section 2.3.2.1; Abt Associates, 2014). In
more recent years (i.e., 2002 to 2017), emissions of SO2 and NOx have undergone the largest
declines, while direct PM2.5 and NH3 emissions have undergone the smallest changes, as shown
in Table 2-2. Regional trends in emissions can differ from the national trends illustrated in Figure
2-14 and Table 2-2. For example, Hand et al. (2012) studied reductions in EGU-related annual
SO2 emissions during the 2001-2010 period and found that while SO2 emissions decreased
throughout the U.S. by an average of 4.9% per year, the amount of change varied across the U.S.
with the largest percent reductions in the western U.S. at 20.1% per year.

23 For more information, see https://www.epa.gov/sciencematters/epas-next-generation-air-measuring-research and
https ://www. epa. gov/air-sensor-toolbox

24 More information on these trends, including details on methods and explanations on the noted changes over time
is available at https://www.epa.gov/air-emissions-inventories/air-pollutant-emissions-trends-data.

25 Emission trends in Figure 2-14 do not include wildfire emissions.

2-26

-------
It should be noted that the reductions shown in PM2.5 and PM10 emissions in Figure 2-14,
Table 2-2, and any subsequent discussions of emission trends are most likely due to changes in
the methods used by the EPA to estimate emissions for source sectors over time In all likelihood,
emissions from dust and fires have increased over this time, which has been noted earlier in this
document and mentioned broadly in the literature as well (Pu and Ginoux, 2017; Li et al., 2021;
Liu et al., 2014; Schoennagel et al., 2017). It should also be noted that these data (in Figure 2-14
and Table 2-2) do not include emissions from wildfires, and these emissions can fluctuate greatly
from year to year.

30000

O^C\lfO'tLO(OSCO(T)0^(Mr0^l^(OSCO(J)0'-CMrO-^LO(OS

8888888888888888888855555555

Year

_«_NH3 -*-NOx -*-PM25 -*-PM10 -*-S02 -*-VOCs

Figure 2-14. National emission trends of PM2.5, PM10, and precursor gases from 1990 to
2017.

2-27

-------
Table 2-2. Percent Changes in PM and PM precursor emissions in the NEI for the time
periods 1990-2017 and 2002-2017.

Pollutant

Percent Change
in Emissions:
1990 to 2017

Percent Change
in Emissions:
2002 to 2017

Major Sources that contribute to
changes over time

nh3

-3.1%

+5.6%

Agricultural Sources (Fertilizer and
Livestock Waste), Fires

N0X

-62%

-60%

EGUs, Mobile Sources

S02

-90%

-84%

EGUs, other Stationary Sources

VOCs

-45%

-26%

Solvents, Fires, Mobile Sources

PM2.5

-36%

-14%

Dust, Fires

PM10

-43%

-25%

Dust, Fires

2.3.2 Trends in Monitored Ambient Concentrations
2.3.2.1 National Characterization of PM2.5 Mass

At long-term monitoring sites in the U.S., annual PM2.5 concentrations from 2017 to 2019
averaged 8.0 [j,g/m3 (with the 10th and 90th percentiles at 5.9 and 10.0 (j,g/m3, respectively) and
the 98th percentiles of 24-hour concentrations averaged 21.3 [j,g/m3 (with the 10th and 90th
percentiles at 14.0 and 29.7 (J,g/m3, respectively). Figure 2-15 (top panels) shows that the highest
ambient PM2.5 concentrations occur in the west, particularly in California and the Pacific
northwest. Much of the eastern U.S. has lower ambient concentrations, with annual average
concentrations generally well below 12.0 |ig/m3 and 98th percentiles of 24-hour concentrations
generally at or below 30 [j,g/m3.

These concentrations are distinct from design values in part because they include days
with episodic events like wildfires and dust storms which can have very high PM2.5 and/or PM10
concentrations. The EPA's Exceptional Events Rule (81 FR 68216, October 3, 2016), most
recently updated in 2016, describes the process by which these events can be excluded from the
design values used for comparison to the NAAQS. For the remainder of Chapter 2, episodic
events are included in the calculations of PM concentrations. When design values are discussed
in Chapter 2, regionally-concurred exceptional events (as of June 2021) have been excluded from
the analysis.26

26 Regionally-concurred exceptional events are unusual or naturally-occurring events such as wildfires or high wind
dust events that have 1) resulted in PM2 5 concentrations above the level of the NAAQS, 2) been submitted by
tribal, state or local air agencies under the EPA's Exceptional Events Rule to their respective EPA Region, and 3)
received concurrence.

2-28

-------
2017-2019 Annual Average

2017-2019 98th Percentile

Possible
A Increase

~ Possible * (0.05
-------
Analysis of monthly data indicate distinct peaks in national ambient PM2.5 concentrations
during the summer and the winter (U.S. EPA, 2019a, Figure 2-22). Through 2008, the summer
peaks reflected the highest national average PM2.5 concentrations. These summer peaks in
ambient PM2.5 concentrations were largely a consequence of summertime peaks in SO2
emissions from power plants in the eastern U.S., and subsequent sulfate formation. However,
substantial reductions in SO2 emissions (see above and U.S. EPA, 2019a, sections 2.5.1.1.1 and
2.5.2.2.1) have changed this pattern. Starting in 2009, winter peaks in national average PM2.5
concentrations have been higher than those in the summer (U.S. EPA, 2019a, section 2.5.2.2.1).
This pattern is illustrated by data from 2013 to 2015, when average winter PM2.5 concentrations
were about 11 (J,g/m3, average summer concentrations were about 9 (j,g/m3, and average spring
and fall concentrations were about 7 [j,g/m3 (Chan et al., 2018).

The ambient PM2.5 concentrations in Figure 2-15 reflect the substantial reductions that
have occurred across much of the U.S. over recent years (Figure 2-15, bottom panels and Figure
2-16). From 2000 to 2019, national annual average PM2.5 concentrations have declined from 13.5
[j,g/m3 to 7.6 (j,g/m3, a 43% decrease (Figure 2-16).27 These declines have occurred at both urban
and rural monitoring sites, although urban PM2.5 concentrations remain consistently higher than
those in rural areas (Chan et al., 2018) due to the so-called "urban increment" of PM2.5 from
local sources in an urban area that is additive to the regional and natural background PM2.5
concentrations.

25-

3 20-

—1—1—1—1—1—1—1—1—1—1—1—1—1—1—1—1—1—1—r

22222222222222222222

00000000000000000000

00000000001 1 11 1 11 11 1

012345678901 23456789

Figure 2-16. Seasonally-weighted annual average PM2.5 concentrations in the U.S. from
2000 to 2019 (406 sites). (Note: The white line indicates the mean concentration while the
gray shading denotes the 10th and 90th percentile concentrations.)

27 See https://www.epa.gov/air-trends/particulate-matter-pm25-trends for up-to-date PM2 5 trends information.

2-30

-------
Analyses at individual monitoring sites indicate that declines in ambient PM2.5
concentrations have been most consistent across the eastern U.S. and in parts of coastal
California, where both annual average and 98th percentiles of 24-hour concentrations have
declined significantly (Figure 2-15, bottom panels). In contrast, trends in ambient PM2.5
concentrations have been less consistent over much of the western U.S., with no significant
changes since 2000 observed at some sites in the Pacific northwest, the northern Rockies and
plains, and the southwest, particularly for 98th percentiles of 24-hour concentrations (Figure 2-
15, bottom panels). Trends in annual average PM2.5 concentrations have been highly correlated
with trends in 98th percentiles of 24-hour concentrations at individual sites (Figure 2-17). Such
correlations are highest across the eastern U.S. and in coastal California, and are somewhat
lower, though still generally positive, at sites in the Central and Western U.S. (i.e., outside of
coastal California).

Figure 2-17. Pearson's correlation coefficient between annual average and 98th percentile
of 24-hour PM2.5 concentrations from 2000-2019.

2.3.2.2 Characterization of PM2.5 Mass at Finer Spatial and Temporal Scales
2.3.2.2.1 CBSA Maximum Annual Versus Daily Design Values

Analysis of recent air quality indicates that maximum annual and daily PM2.5 design
values within a CBSA are positively correlated with some noticeable regional variability (Figure
2-18). In the Southeast, Northeast, and Industrial Midwest regions, the annual design values are
high relative to the daily design values due in part to the infrequent impacts of episodic events

2-31

-------
like wildfire or dust storms. On the other hand, the Northwest region has very high daily design
values relative to the annual design values. This is due to episodically high PM2.5 concentrations
that affect the region, both from wintertime stagnation events and summer/fall wildfire smoke
events.28 The relatively small population and low emissions in the region result in much lower
PM2.5 concentrations during the other parts of the year not affected by these episodes.

E
01

ru
>
c

'in

1—1

x
ro

<
LO

Other (AK, HI)

• IndustMidwest

• Northeast

• Northwest

• SoCal

• Southeast

• Southwest

• UpperMidwest

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

CBSA Maximum 2017-2019 Annual Design Value (|jg nr3)

Figure 2-18. Scatterplot of CBSA maximum annual versus CBSA maximum daily design
values (2017-2019) with the solid black line representing the ratio of daily and annual
NAAQS values.

2.3.2.2.2 PM2.5 Near Major Roadways

Because of its longer atmospheric lifetime (U.S. EPA, 2019a, section 2.2), PM2.5 is
expected to exhibit less spatial variability on an urban scale than UFP or PM10-2.5 (U.S. EPA,

28 Due to the recent time period shown in Figure 2-18, it is likely that some of the annual and daily design values are
affected by potential exceptional events associated with wildfire smoke that have yet to be regionally-concurred
and removed from the design value calculations. The EPA defines exceptional events as unusual or natural-
occurring events that that affect air quality but are not reasonably controllable using techniques that tribal, state,
or local air agencies may implement. This is especially likely for the daily design values in the Northwe st region
which experienced frequent wildfire smoke events during the 2017-2019 period.

2-32

-------
2019a, section 2.5.1.2.1). Analyses in the 2009 ISA for PM indicated that correlations between
PM2.5 monitoring sites up to a distance of 100 km from each other were greater than 0.75 in most
urban areas. However, more substantial spatial variation has been reported for some urban areas,
due in part to proximity between monitors and emissions sources (U.S. EPA, 2019a, section
2.5.1.2.1). The recent deployment of PM2.5 monitors near major roads in large urban areas
provides some insight into the localized PM2.5 concentration increment due to vehicular
emissions.

As discussed above, in the 2012 review of the PM NAAQS, the EPA required monitoring
of PM2.5, along with NO2 and CO, near major roads in CBS As with populations greater than 1
million. PM2.5 monitoring was required to start for the largest CBS As at the beginning of 2015,
and several years of data are now available for analysis at these sites. DeWinter et al. (2018)
analyzed these data and found that the average near-road increment (difference between near-
road PM2.5 concentrations and the concentrations at other sites in the same CBSA) was 1.2 [j,g/m3
for 2014-2015. Gantt et al. (2021) found that this near-road increment from the 74 near-road sites
deployed in 2018 exhibited a diurnal cycle, with a peak of 1.8 [j,g/m3 during the morning rush
hour relative to a 1.0 [j,g/m3 daily average. This near-road increment likely is additive to the
urban increment of PM2.5 from local sources in the CBSA including mobile sources on the
numerous non-highway roads that are not monitored by the near-road network. For 2016-2018,
Gantt et al. (2021) also reported that 52% and 24% of the time the near-road sites reported the
highest annual and 24-hour PM2.5 design value in the CBSA, respectively. Of the CBS As with
the highest annual design values at near-road sites reported by Gantt et al. (2021), those design
values were, on average, 0.8 |ig/m3 higher than at the highest measuring non-near-road sites
(range is 0.1 to 2.1 |ig/m3 higher at near-road sites).

Although most near-road monitoring sites do not have sufficient data to evaluate long-
term trends in near-road PM2.5 concentrations, Gantt et al. (2021) analyzed data at one long-term
near-road-like site in Elizabeth, NJ,29 and found that the annual average PM2.5 increment has
generally decreased between 2001 and 2018 from about 2.0 [j,g/m3 to about 1.3 |ig/m3. The trend
in the near-road increment of elemental carbon at the Elizabeth, NJ site has shown a similar
reduction, with values of-1.0 [j,g/m3 in 2001 decreasing to -0.5 [j,g/m3 in 2018. These data are
consistent with the timing of EPA emission standards for motor vehicles.30 Although long-term
data are not available at other near-road sites, the national scope of the diesel vehicle controls

29 The Elizabeth Lab site in Elizabeth, NJ is situated approximately 30 meters from travel lanes of the Interchange
13 toll plaza of the New Jersey Turnpike and within 200 meters of travel lanes for Interstate 278 and the New
Jersey Turnpike.

30 See https://www.epa.gOv/diesel-fuel-standards/diesel-Iuel-standards-and-rulemakings#nonroad-diesel.

2-33

-------
suggests the near-road environment across the U.S. may have experienced similar decreasing
trends in near-road PM2.5 increments.

2.3.2.2.3 Sub-Daily Concentrations ofPMxs

Ambient PM2.5 concentrations can exhibit a diurnal cycle that varies due to impacts from
intermittent emission sources, meteorology, and atmospheric chemistry. The PM2.5 monitoring
network in the U.S. has an increasing number of continuous FEM monitors reporting hourly
PM2.5 mass concentrations that reflect this diurnal variation. The 2019 ISA describes a two-
peaked diurnal pattern in urban areas, with morning peaks attributed to rush-hour traffic and
afternoon peaks attributed to a combination of rush hour traffic, decreasing atmospheric dilution,
and nucleation (U.S. EPA, 2019a, section 2.5.2.3, Figure 2-32). Because a focus on annual
average and 24-hour average PM2.5 concentrations could mask sub-daily patterns, and because
some health studies examine PM exposure durations shorter than 24-hours, it is useful to
understand the broader distribution of sub-daily PM2.5 concentrations across the U.S. Figure 2-19
below presents the frequency distribution of 2-hour average PM2.5 mass concentrations from all
FEM PM2.5 monitors in the U.S. for 2017-2019.31 At sites meeting the current primary PM2.5
standards, these 2-hour concentrations generally remain below 10 ng/m3, and rarely exceed 30
Hg/m3. Two-hour concentrations are higher at sites violating the current standards, generally
remaining below 16 |ag/m3 and rarely exceeding 80 |ag/m3.

31 As discussed further in section 3.2, PM2 5 controlled human exposure studies often examine 2-hour exposures.
Thus, when evaluating those studies in the context of the current primary PM2 5 standards, it is useful to consider
the distribution of 2-hour PM2 5 concentrations. Similar analyses of 4-hour and 5-hour PM2 5 concentrations are
presented in Appendix A, Figure A-2 and Figure A-3, respectively.

2-34

-------
Sites meeting both NAAQS

Sites violating either NAAQS

Percentiles (pg m

180

Concentration (|ig m"3)

Concentration (jug m )

Figure 2-19. Frequency distribution of 2017-2019 2-hour averages for sites meeting both or
violating either PM2.5 NAAQS for October to March (blue) and April to September
(red).

The extreme upper end of the distribution of 2-hour PM2.5 concentrations is shifted higher
during the warmer months (red in Figure 2-19), generally corresponding to the period of peak
wildfire frequency (April to September) in the U.S. At sites meeting the current primary
standards, the highest 2-hour concentrations measured rarely occur outside of the period of peak
wildfire frequency. Most of the sites measuring these very high concentrations are in the
northwestern U.S. and California, where wildfires have been relatively common in recent years
(see Appendix A, Figure A-l). When the period of peak wildfire frequency is excluded from the
analysis (blue in Figure 2-19), the extreme upper end of the distribution is reduced.

2.3.2.3 Chemical Composition of PM2.5

Based on recent air quality data, the major chemical components of PM2.5 have distinct
spatial distributions. Sulfate concentrations tend to be highest in the eastern U.S., while in the
Ohio Valley, Salt Lake Valley, and California nitrate concentrations are highest and relatively
high concentrations of organic carbon are widespread across most of the Continental U.S., as
shown in Figure 2-20. Elemental carbon, crustal material, and sea-salt are found to have the
highest concentrations in the northeast U.S., southwest U.S., and coastal areas, respectively.

2-35

-------
2017-2019 Sulfate 2017-2019 Nitrate

Figure 2-20. Annual average PM2.5 sulfate, nitrate, organic carbon, and elemental carbon
concentrations (in jitg/m3) from 2017-2019.

An examination of PM2.5 composition trends can provide insight into the factors
contributing to overall reductions in ambient PM2.5 concentrations. The biggest change in PM2.5
composition that has occurred in recent years is the reduction in sulfate concentrations due to
reductions in SO2 emissions. Between 2000 and 2015, the nationwide annual average sulfate
concentration decreased by 17% at urban sites and 20% at rural sites. This change in sulfate
concentrations is most evident in the eastern U.S. and has resulted in organic matter or nitrate
now being the greatest contributor to PM2.5 mass in many locations (U.S. EPA, 2019a, Figure 2-
19). The overall reduction in sulfate concentrations has contributed substantially to the decrease
in national average PM2.5 concentrations as well as the decline in the fraction of PM10 mass
accounted for by PM2.5 (U.S. EPA, 2019a, section 2.5.1.1.6; section 2.3.1 above).

2.3.2.4 National Characterization of PiVlio Mass

At long-term monitoring sites in the U.S., the 2017-2019 average of 2nd highest 24-hour
PM10 concentration was 68 ug/'rn3 (with the 10th and 90th percentiles at 28 and 124 pg/m3,

2-36

-------
respectively) (Figure 2-21, top panels).32 The highest PMio concentrations tend to occur in the
western U.S. Seasonal analyses indicate that ambient PMio concentrations are generally higher in
the summer months than at other times of year, though the most extreme high concentration
events are more likely in the spring (U.S. EPA, 2019a, Table 2-5). This is due to fact that the
major PMio emission sources, dust and agriculture, are more active during the warmer and drier
periods of the year.

32 The form of the current 24-hour PMio standard is one-expected-exceedance, averaged over three years.

2-37

-------
2017-2019 Annual Average

2017-2019 2nd Highest

<0.05
-------
Recent ambient PMio concentrations reflect reductions that have occurred across much of the
U.S. (Figure 2-21, bottom panels). From 2000 to 2019, 2nd highest 24-hour PMio concentrations
have declined by about 46% (Figure 2-22).33 Analyses at individual monitoring sites indicate that
annual average PMio concentrations have declined at most sites across the U.S., with much of the
decrease in the eastern U.S. associated with reductions in PM2.5 concentrations. Annual second
highest 24-hour PMio concentrations have generally declined in the eastern U.S., while
concentrations in the much of the midwest and western U.S. have remained unchanged or
increased since 2000 (Figure 2-21, bottom panels).

00000000000000000000

00000000001 1111 111 11

0 1234567890123456789

Figure 2-22. National trends in Annual 2nd Highest 24-Hour PMio concentrations from
2000 to 2019 (262 sites). (Note: The white line indicates the mean concentration while the
gray shading denotes the 10th and 90th percentile concentrations.)

Compared to previous reviews, data available from the NCore monitoring network in the
current reconsideration allows a more comprehensive analysis of the relative contributions of
PM2.5 and PM10-2.5 to PMio mass. PM2.5 generally contributes more to annual average PMio mass
in the eastern U.S. than the western U.S. (Figure 2-23). At most sites in the eastern U.S., the
majority of PMio mass is comprised of PM2.5. As ambient PM2.5 concentrations have declined in
the eastern U.S. (section 2.3.2.2, above), the ratios of PM2.5 to PMio have also declined.

33 For more information, see https://www.epa.gOv/air-trends/particulate-matter-pml0-trends#pmnat

2-39

-------
2017-2019 PM2.5/PM10 Ratio (Annual Average)

Figure 2-23. Annual average PM2.5/PM10 ratio for 2017-2019.

For sites with days having concurrently very high PM2.5 and PM10 concentrations (Figure
2-24), the PM2.5/PM10 ratios are typically higher than the annual average ratios. This is
particularly true in the northwestern U.S. where the high PM10 concentrations can occur during
wildfires with high PM2.5.

2017-2019 PM2 5/PM10 Ratio (2nd Highest PM10)

Figure 2-24. PM2.5/PM10 ratio on days ranking as the second highest yearly PM10

concentration and among the top four highest yearly PM2.5 concentrations for 2017-
2019.

2-40

-------
2.3.2.5 National Characterization of PM10-2.5 Mass

Since the 2012 review, the availability of PM10-2.5 ambient concentration data has greatly
increased. As illustrated in Figure 2-2534 (top panels), annual average and 98th percentile PM10-2.5
concentrations exhibit less distinct differences between the eastern and western U.S. than for
either PM2.5 or PM10. Due to the short atmospheric lifetime of PM10-2.5 relative to PM2.5, many of
the high concentration sites in Figure 2-25 are isolated and likely near emission sources
associated with wind-blown and fugitive dust. The spatial distributions of annual average and
98th percentile concentations of PM10-2.5 are more similar than that of PM2.5, suggesting that the
same dust-related emission sources are affecting both long-term and episodic concentrations.
The highest concentrations of PM10-2.5 are in the southwest U.S. where widespread dry and
windy conditions contribute to wind-blown dust emissions. Additionally, compared to PM2.5 and
PM10, changes in PM10-2.5 concentrations have been small in magnitude and inconsistent in
direction (Figure 2-25, lower panels). The majority of PM10-2.5 sites in the U.S. do not have a
concentration trend from 2000-2019, reflecting the relatively consistent level of dust emissions
across the U.S. during the same time period.35

34 The sites shown in Figure 2-25 have a data completeness of either 75% or >182 valid days in each year.

35 PM from dust emissions in the NEI remain fairly consistent from year-to-year, except when there are severe
weather incursions or there is a dust event that transports or causes major local dust storms to occur (particularly
in the western U.S.). These dust events and weather incursions needed to effect dust emissions on a national level
are not common and only seldomly occur. In the emissions trends analysis presented in section 2.1.1 above, dust
is included in the NEI sector labeled "miscellaneous."

2-41

-------
2017-2019 98th Percentile

2000-2019 Annual Average Trend

~ No Trend
(p>0.15)

T Reduction Possible

(pso.05) ^ Increase

~ Possible (0.05
-------
2.3.2.6 Characterization of the Ultrafine Fraction of PM2.5 Mass

Compared to PM2.5 mass, there is relatively little data on U.S. particle number
concentrations, which are dominated by UFP. In the published literature, annual average particle
number concentrations reaching about 20,000 to 30,000 cm3 have been reported in U.S. cities
(U.S. EPA, 2019a). In addition, based on UFP measurements in two urban areas (New York
City, Buffalo) and at a background site (Steuben County) in New York, there is a pronounced
difference in particle number concentration between different types of locations (Figure 2-26;
U.S. EPA, 2019a, Figure 2-18). Urban particle number counts were several times higher than at
the background site, and the highest particle number counts in an urban area with multiple sites
(Buffalo) were observed at a near-road location. Hourly data indicate that particle numbers
remain fairly constant throughout the day at the background site, that they peak around 8:00 a.m.
in Buffalo and New York City (NYC), and that they remain high into the evening hours with
distinct rush hour and early afternoon peaks.

0123456789 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Hour of the Day

Figure 2-26. Average hourly particle number concentrations from three locations in the
State of New York for 2014 to 2015 (green is Steuben County, orange is Buffalo, red is
New York City). (Source: Figure 2-18 in U.S. EPA, 2019a).

Long-term trends in UFP are generally not available at U.S. monitoring sites. However,
data on number size distribution have been reported for an 8-year period from 2002 to 2009 in
Rochester, NY. Number concentrations averaged 4,730 cnT3 for 0.01 to 0.05 [j,m particles and
1,838 cnT3 for 0.05 to 0.1 [j,m particles (Wang et al., 2011). On average over the 8 years that
UFP data were collected in Rochester, total particle number concentrations declined from the
earlier period evaluated (i.e., 2001 to 2005) to the later period (2006 to 2009). This decline was

2-43

-------
most evident for particles between 0.01 and 0.1 |iin and was attributed to changes in local
sources resulting from the 2007 Heavy Duty Highway Rule (66 FR 5002, January 18, 2001), a
reduction in local industrial activity, and the closure of a nearby coal-fired power plant (Wang et
al., 2011; U.S. EPA, 2019a, section 2.5.2.1.4).

In addition, at a site in Illinois the annual average particle number concentration declined
between 2000 and 2019, closely matching the reductions in annual PM2.5 mass over that same
period (Figure 2-27, below). Particle number concentrations at this site are closer to those of the
background site in Figure 2-27 than the urban sites. A recent study found that particle number
concentrations in an urban area (Pittsburgh, PA) decreased between 2001-2002 and 2016-2017
along with decreases in PM2.5 associated with SO2 emission reductions (Saha et al., 2018).
However, the relationship between changes in ambient PM2.5 and UFPs cannot be
comprehensively characterized due to the high variability and limited monitoring of UFPs.

16 n
14

m 12

o
+->
ra

4-J

c
QJ
U

c
o
u

l/l
c/1

8 H
6
4

2 H

y = -0.3558x + 725.29
R2 = 0.806

• •'

y = -140.11x + 285875
R2 = 0.7476

7000
6000

- 5000

- 4000
3000

b 2000
1000
0

c
o
'+->
03

i—
•*—>
c
O)
u
c
o
u

-------
hybrid modeling approaches tend to broaden the areas captured in the exposure assessment, and
in doing so, the studies that utilize these methods tend to report lower mean PM2.5 concentrations
than monitor-based approaches. Further, other aspects of the method used to calculate PM2.5
concentrations (i.e. population weighting, trim mean) can also have an impact on the predicted
exposure and the related study-reported mean concentration.

2.3.3.1 Predicted Ambient PM2.5 and Exposure Based on Monitored Data

For an area to meet the NAAQS, all valid design values in that area, including the highest
annual and 24-hour values, must be at or below the levels of the standards. Because monitors are
often required in locations with high PM2.5 concentrations (section 2.2.3), areas meeting an
annual PM2.5 standard with a particular level would be expected to have long-term average PM2.5
concentrations (i.e., averaged across space and over time in the area) somewhat below that
standard level. Figure 2-28 and Figure 2-29 indicate that, based on recent air quality in U.S.
CBS As, maximum annual PM2.5 design values are often 10% to 20% higher than annual average
concentrations (i.e., averaged across multiple monitors in the same CBSA). The difference
between the maximum annual design value and average concentration in an area can be smaller
or larger than this range, likely depending on factors such as the number of monitors, monitor
siting characteristics, and the distribution of ambient PM2.5 concentrations. Given that higher
PM2.5 concentrations have been reported at some near-road monitoring sites, relative to the
surrounding area (section 2.3.2.2.2), recent requirements for PM2.5 monitoring at near-road
locations in large urban areas (section 2.2.3.3) may increase the ratios of maximum to average
annual design values in some areas. Such ratios may also depend on how the averages are
calculated (i.e., averaged across monitors versus across modeled grid cells). Compared to annual
design values, Figure 2-29 indicates a more variable relationship between maximum 24-hour
PM2.5 design values and annual average concentrations.

2-45

-------

18 -

17 -

•

16 -

/
/

15 -

/ 1:1 line

• /

14 -

•

/
/

13 -

• , •

in
O)
O

12 ¦
11 -

"ro

ZS
C
C
<

10 -
9 -

•• \ /

A™ /

CTi

• • • JT'

8 -

„ •• r

o
r\i

• ;•*
• r

7 ¦

T—1

6 -

Hawaii

r\i

• IndustMidwest

5 ¦

• Northeast

4 -

• Northwest

X
fU

3 ¦

• SoCal

• Southeast

U~)

2 -

• Southwest

co
u

1 -
0 -

1 1 1 1 1 1 1 1 1 1

• UpperMidwest

T 1 1 1 1 1 1

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
CBSA Average 2017-2019 Annual Design Value ((jg nr3)

Figure 2-28. Comparison of CBSA average annual design values and CBSA maximum
annual design values for 2017-2019. (Note: Includes all CBSAs with at least 3 valid annual
DVs.)

2-46

-------
Table 2-3. Nationwide averages of ratios of maximum annual PM2.5 design values to
average composite monitor PM2.5 concentrations across CBSAs.

Years of
Monitoring Data

Number of Monitors
per CBSA

Number
of CBSAs

Ratio of Maximum
Annual DV to CBSA
Average

Ratio of Maximum
24-hour DV to CBSA
Average

2009-2011

3 or more

1.12

1.13

4 or more

1.14

1.16

5 or more

1.17

1.19

2012-2014

3 or more

1.15

4 or more

1.17

1.18

5 or more

1.19

1.21

2015-2017

3 or more

1.16

1.19

4 or more

1.19

1.21

5 or more

1.20

1.24

2017-2019

3 or more

1.16

1.22

4 or more

1.19

1.25

5 or more

1.21

1.26

2-47

-------
E

fD
>

ai
1/1
(D

ro
~

(7*

1—1

f\l

O
r\i

(T3

<
LO
CD
U

60 n
55 -
50

45 :

40 -

30 :

25 -
20

15 :
10 :

5 -

Hawaii

• IndustMidwest

• Northeast

• Northwest

• SoCal

• Southeast

• Southwest

• UpperMidwest

• M •
• •

• • • •
•• •
• • • •
m» mm
• •
• •• •
• • ••

0 1

8 9 10 11 12 13 14 15 16 17 18

CBSA Average 2017-2019 Annual Design Value (pg rrr3)

Figure 2-29. Comparison of CBSA average annual design values and CBSA maximum
daily design values for 2017-2019. (Note: Dashed lines indicate the level of the current 24-
hour PM2.5 standard (35 |ig/m3) and the current annual PM standard (12.0 |ig/m3). Includes
all CBS As with at least 3 valid daily and 3 valid annual DVs.)

2.3.3.2 Predicted Ambient PM2.5 Based on Hybrid Modeling Approaches

Ambient concentrations of PM2.5 are often characterized using measurements from
national monitoring networks due to the accuracy and precision of the measurements and the
public availability of data. For applications requiring PM2.5 characterizations across urban areas,
data averaging techniques such as area-wide and population-weighted averaging of monitors are
sometimes used to provide complete coverage from the site measurements (U.S. EPA, 2019a,
chapter 3). Yet data averaging methods may not adequately represent the spatial heterogeneity of
PM2.5 within an area and are not practical for large unmonitored areas or time periods. As a
result, additional methods have been developed to improve PM2.5 characterizations in areas
where monitoring is relatively sparse or unavailable. Methods include interpolation of monitored
data, land-use regression models, chemical-transport models (CTMs), models based on satellite-
derived aerosol optical depth (AOD), and hybrid spatiotemporal models that combine

2-48

-------
information from the individual approaches (U.S. EPA, 2019a, chapter 3). A number of recent
studies have employed such methods to estimate PM2.5 air quality concentrations across the U.S.
and Canada, and to estimate population exposures for use in epidemiologic analyses (U.S. EPA,
2019a, sections 3.3 and 3.4). Given the increasing availability and application of these methods,
in this section we provide an overview of recently developed hybrid modeling methods, their
predictions and performance, and how predictions from various methods compare to each other.

2.3.3.2.1 Overview of Hybrid Methods

Hybrid methods are broadly classified into four categories: (1) methods based primarily
on interpolation of monitor data, (2) Bayesian statistical downscalers, (3) methods based
primarily on satellite-derived AOD, and (4) methods based on machine-learning algorithms.

Each method is discussed briefly below.

Interpolation-based methods are the simplest approach for developing spatial fields of
PM2.5 concentrations and rely on the moderate degree of spatial autocorrelation in PM2.5 in many
areas of the U.S. Interpolation methods often use inverse-distance or inverse-distance-squared
weighted averaging of monitoring data to predict PM2.5 concentrations at unmonitored receptor
points. Examples include the Voronoi neighbor averaging (VNA) approach and the enhanced
VNA approach (eVNA). The VNA approach applies weighted averaging to the concentrations
monitored in the Voronoi cells neighboring the cell containing the prediction point (Abt
Associates, 2014). In the eVNA approach, monitored data are further weighted by the ratio of
CTM predictions in the grid-cell containing the prediction point to the grid-cell containing the
monitor (Abt Associates, 2014).

Bayesian statistical modeling has been used to calibrate CTM PM2.5 predictions or
satellite-derived AOD estimates to surface measurements (Berrocal et al., 2012; Wang et al.,
2018b, Berrocal et al., 2020). This approach, commonly referred to as a Bayesian downscaler
because it "downscales" grid-cell average values to points, first regresses the PM2.5 predictions
or AOD estimates on monitoring data. The resulting relationships are then used to develop a
gridded PM2.5 field from the CTM or AOD input field. Bayesian downscalers have been applied
to develop gridded daily PM2.5 fields at 12-km resolution for the conterminous U.S. (Wang et al.,
2018b; U.S. EPA, 2017). An ensemble technique that optimally combines predictions of CTM
and AOD downscalers has also been developed to predict PM2.5 at high resolution over Colorado
during the fire season (Geng et al., 2018).

Surface PM2.5 concentrations can also be predicted based on satellite retrievals of AOD
and the relationship between surface PM2.5 and AOD from CTM simulations (van Donkelaar et
al., 2010). For example, in van Donkelaar et al. (2015a), satellite-based approaches (van
Donkelaar et al., 2010; van Donkelaar et al., 2013) were used to estimate a gridded field of

2-49

-------
global mean PM2.5 concentration for the 2001-2010 period that was combined with information
from radiometrically stable satellite instruments (Boys et al., 2014) to develop global PM2.5
fields over the 1998-2012 period (van Donkelaar et al., 2015a). Motivated by the limited use of
surface measurements in this approach, van Donkelaar et al. (2015b) developed an updated
method that incorporates additional information from PM2.5 monitoring networks to improve
performance. Specifically, geographically weighted regression (GWR) of residual PM2.5 (i.e., the
difference between monitored PM2.5 and predictions based on satellite-derived AOD) with land-
use and other variables is performed to improve PM2.5 concentration estimates in areas such as
North America where monitoring is relatively dense (van Donkelaar et al., 2019; van Donkelaar
et al., 2015b). This approach has been used to create long-term PM2.5 fields globally and for
North America at about 1-km resolution. However, the developers caution that PM2.5 gradients
may not be fully resolved at 1-km resolution due to the influence of coarser-scale data used in
the model36 and report that mean error variance decreases when averaging the 1-km fields to
coarser resolution (van Donkelaar et al., 2019).

Daily PM2.5 fields based on non-parametric (i.e., machine learning) methods have also
been developed to characterize PM2.5 over the U.S. Non-parametric methods facilitate the use of
large numbers of predictor variables that may have complex nonlinear relationships with PM2.5
concentrations that would be challenging to specify with a parametric method. For example, a
neural network algorithm was used to predict daily PM2.5 fields at 1-km resolution over the
conterminous U.S. during 2000-2012 using more than 50 predictor variables including satellite-
derived AOD, CTM predictions, satellite-derived absorbing aerosol index, meteorological data,
and land-use variables (Di et al., 2016). A random forest algorithm was also applied to develop
daily PM2.5 fields at 12-km resolution over the conterminous U.S. in 2011 and provide variable
importance information for about 40 predictor variables including CTM results and satellite-
derived AOD (Hu et al., 2017). Satellite-derived AOD and the convolution layer for nearby
PM2.5 measurements are ranked among the top five most important predictor variables for the
importance metrics considered. An ensemble model based on random forest, neural network, and
gradient boosting methods has also been recently applied to develop daily 1-km PM2.5
concentration fields over the U.S. for the 2000-2015 period (Di et al., 2019). A wide range of
parametric and non-parametric hybrid PM2.5 models have recently been reviewed in Chapter 3 of
the 2019 ISA (U.S. EPA, 2019a).

2.3.3.2.2 Performance of the Methods

36 See https://sites.wustl.edu/acag/datasets/surface-pm2-5/

2-50

-------
The performance of hybrid modeling methods is often evaluated against surface
measurements using n-fold cross validation (i.e., 1/n of the data are reserved for validation with
the rest used for model training, and the process is repeated n times). Although model evaluation
methods are not consistent across studies, ten-fold cross-validation statistics are often reported
and support use of the hybrid methods just described. For example, the neural network achieved
total R2 of 0.84 and root-mean-square error (RMSE) of 2.94 |Lxg m"3 for daily PM2.5 predictions at
sites in the conterminous U.S. during 2000-2012 (Di et al., 2016). The random forest achieved
total R2 of 0.80 and RMSE of 2.83 |Lxg m"3 for daily PM2.5 predictions at U.S. sites in 2011 (Hu et
al., 2017). The satellite-derived AOD approach with GWR yielded an R2 of 0.79 and RMSE of
1.7 |Lxg m"3 in cross validation for longer-term PM2.5 predictions at sites in North America (van
Donkelaar et al., 2015b). The Bayesian downscalers had weaker performance in cross validation
(e.g., national R2: 0.66-0.70; Wang et al., 2018b; Kelly et al., 2019a) than the other methods,
possibly due to the relatively small number of predictor variables. However, the downscalers
have advantages of simplicity, computational efficiency, and lower potential for overfitting
compared with the machine learning methods.

Although model validation analyses often report favorable performance in terms of
aggregate cross-validation statistics, studies have reported heterogeneity in performance by
season, region, and concentration range. For example, several methods had relatively high cross-
validation R2 in summer compared with other seasons (Kelly et al., 2019a ; Hu et al., 2017; Di et
al., 2016; van Donkelaar et al., 2015b). Also, studies have noted relatively weak performance in
parts of the western U.S., possibly due to the sharp concentration gradients, complex terrain, low
concentrations (and therefore signal-to-noise ratio), less dense monitoring, prevalence of
wildfire, and challenges in satellite retrievals and CTM modeling (Di et al., 2016; Wang et al.,
2018b; Hu et al., 2017; Kelly et al., 2019a). Predictive capability in terms of cross-validation R2
has also been reported to weaken with decreasing PM2.5 concentration in several studies (e.g.,
Kelly et al., 2019a; Di et al., 2016; van Donkelaar et al., 2019). This trend could be due in part to
increases in the fraction of the PM2.5 distribution that is explained by less predictable stochastic
variation as PM2.5 concentrations decrease (Just et al., 2020). Trends in model performance
associated with PM2.5 concentration (e.g., Figure 2-30) could also be due to the relatively sparse
monitoring in remote areas, where PM2.5 concentrations tend to be low. Consistent with this
hypothesis, studies have reported degradation of model performance metrics with increasing
distance to the nearest in-sample monitor, suggesting that predictions are most reliable in densely
monitored urban areas (Jin et al., 2019; Huang et al., 2018; Kelly et al., 2019a; Berrocal et al.,
2020).

2-51

-------
1.00-

0.75-

0.50-

F^l Downscaler

0.25-

^ VNA
^ eVNA

0.00-

46 171 487 301 37

I I I I I

0-3 3-6 6-9 9-12 >12

Observed PM2.5 (ng m 3)

Figure 2-30. R2 for ten-fold cross-validation of daily PM2.5 predictions in 2015 from three
methods for individual sites as a function of observed concentration. Text indicates the
number of monitors in the PM2.5 concentration range. Downscaler: Bayesian downscaler of
CMAQ predictions; VNA: Voronoi Neighbor Averaging; eVNA: enhanced-VNA. From
Kelly et al., 2019a.

A limited number of studies have intercompared concentration predictions based on
different PM2.5 characterization methods. Huang et al. (2018) compared PM2.5 concentrations
from the method of Di et al. (2016) with concentrations from the CTM-based data fusion method
of Friberg et al. (2016) and the satellite-derived AOD approach of Hu et al. (2014) for North
Carolina. They reported general agreement in concentrations among methods, with some
differences along the coast and in forested regions where monitoring is less dense. Yu et al.
(2018) compared PM2.5 concentrations from fourteen approaches of varying complexity for
developing PM2.5 spatial fields over the Atlanta, Georgia region. They reported that predictions
of the methods can differ considerably, and the hybrid approaches that incorporate CTM
predictions generally outperformed the simpler techniques (e.g., monitor interpolation). Also,
model predictions appeared to be more reliable in the urban center based on relatively low cross
validation R2 for sites away from the urban core. Jin et al. (2019) reported increasing uncertainty
in hybrid model predictions with distance to the nearest AQS monitor. Keller and Peng (2019)
reported that a prediction model incorporating CTM output outperformed a monitor averaging
approach and error reduction could be achieved by restricting the study to areas near monitors.
Diao et al. (2019) reviewed publicly available PM2.5 products and identified inconsistencies in
PM2.5 predictions from several methods. Kelly et al. (2021) reported broad agreement among
model predictions at the national scale but differences in the intra-urban variations in PM2.5
concentrations.

2.3.3.2.3 Comparison of PM2.5 Fields Across Approaches

To illustrate features of the spatial fields reported in the literature, the annual mean PM2.5
concentrations for 2011 from four methods is shown in Figure 2-31, where predictions from the
methods were averaged to a common 12-km grid. The fields were developed using a Bayesian
downscaler (downscaler, Berrocal et al., 2012), neural network (DI2016, Di et al., 2016), random

2-52

-------
forest (HU2017, Hu et al., 2017), and GWR of residuals from satellite-based PM2.5 estimates
(VD2019; van Donkelaar et al., 2019). Annual mean concentrations were developed from daily
PM2.5 predictions in the downscaler, DI2016, and HU2017 cases and from monthly PM2.5
predictions in the VD2019 case. General features of the 2011 fields are in reasonable agreement
across methods, with elevated concentrations across broad areas of the eastern U.S. and in the
San Joaquin Valley and South Coast Air Basin of California. The national mean PM2.5
concentration for the VD2019 case (7.06 |.ig m"3) is slightly lower than those of the other cases
(7.36-7.44 |ig m"3), possibly because the VD2019 fields were developed using monthly (rather
than daily) PM2.5 measurements. Use of monthly averages provides greater influence on the
annual mean of sites with less frequent monitoring that tend to be in rural areas with relatively
low concentrations. Mean PM2.5 concentrations predicted by the four methods in nine U.S.
climate regions (Karl and Koss, 1984) are provided in Table 2-4.

downscaler

Avg: 7.44 ug/m3

iter"

45-;

40-
35-
30-

Avg: 7.36 ug/m3
-120 -110

HU2017

"satr ^

D12016

; jBMHI Mr

V, f

X " 4 v

Avg: 7.38 ug/m3

Gulf Of

-100

-90

-80

-70
Longitude

VD2019

Hpr

M Gulfof 1,. .

1 1 1 1

I ,

Avg: 7.06 ug/m3

ug/m3

> 15

Figure 2-31. Comparison of 2011 annual average PM2.5 concentrations from four methods.

(Note: These four methods include: downscaler (Berrocal et al., 2012), DI2016 (Di et al.,
2016), HU2017 (Hu et al., 2017), and VD2019 (van Donkelaar et al., 2019). Predictions have
been averaged to a common 12-km grid for this comparison.)

2-53

-------
Table 2-4. Mean 2011 PM2.5 concentration by region for predictions in Figure 2-29

Region1

downscaler

HU2017

DI2016

VD2019

Northeast

8.5

8.0

8.2

7.5

Southeast

9.9

10.0

9.4

9.8

Ohio Valley

10.7

9.6

9.8

10.0

Upper Midwest

00
CO

7.9

7.1

South

00
CO

8.9

9.0

8.7

Southwest

5.0

5.3

5.2

5.1

N. Rockies & Plains

5.6

5.9

5.6

4.5

Northwest

5.0

5.3

6.1

4.9

West

5.5

5.7

6.0

6.5

1 U.S. climate reaion: https://www.ncdc.noaa.aov/monitorina-references/maps/us-climate-reaions.php.

In Figure 2-32, PM2.5 concentrations predicted by the four methods are shown at their
native resolution for regions centered on California, New Jersey, and Arizona. Predictions have
sharper spatial gradients and span a wider range of concentrations for the western regions
centered on California and Arizona (Figure 2-32, panels a and c) than the eastern region centered
on New Jersey (Figure 2-32, panel b). Despite general agreement among predictions for the
California and the eastern U.S. areas, the spatial texture of the concentration fields differs among
methods. For instance, the 12-km Bayesian downscaler produces the smoothest PM2.5
concentration field, and the 1-km neural network (DI2016) produces the field with the greatest
variance. Some of the largest differences in PM2.5 concentration among methods occurred over
southwest Arizona. The DI2016 and VD2019 methods predict higher concentrations in this area
than the downscaler and HU2017 methods, and the DI2016 approach predicts distinct spatial
features associated with Interstate 40, 10, and 8 that are not apparent in the other fields (Figure 2-
32, panel c).

2-54

-------
1 downscaler

DI2016

(b)

downscaler

D12016

^Modesto

42-
41 -

Ik - m

" m
°

fcrfresno

..^Bakersfield

40-
39-

0'ld,riS|^lla16

-114 -112 -110

Longitude

ug/m3

Figure 2-32. Comparison of 2011 annual average PM2.5 concentrations from four methods
for regions centered on the (a) California (b) New Jersey, and (c) Arizona. Predictions
are shown at their native resolution (i.e., about 1-km for DI2016 and VD2019 and 12-km for
downscaler and HU2017).

In Figure 2-33, the coefficient of variation (CV; i .e., the standard deviation divided by the
mean) among methods is shown in percentage units based on predictions that were averaged to a
common 12-km grid. The largest values occur in the western U.S. (Figure 2-33, panel a), where
spatial gradients are high, terrain is complex, wildfire is prevalent, monitoring is relatively
sparse, and PM2.5 concentrations are low on average. The distance from the grid-cell center to the
nearest monitor is greater than 100 km for broad areas of the west (Figure 2-34).

2-55

-------
(a)

? rS." t i

k_- ¦ vkB H&ks

45-P"^ 7* i C

(b)

(%)

60-
50-

> 50

ra 40-
ro

i 30-

-120 -110

Gulf of
-100 -90

Longitude

11442 13687 11302 14501

3-5 5-7 7-9 9-11

11-16

Average PM2.5 (jig m )

Figure 2-33. (a) Spatial distribution of the CV (i.e., standard! deviation divided by mean) in
percentage units for the four models in Figure 2-29. (b) Boxplot distributions of CV for
grid cells binned by the average PM2.5 concentration for the four models. (Note: The box
brackets the interquartile range (IQR), the horizontal line within the box represents the
median, the whiskers represent 1.5 times the IQR from either end of the box, and circles
represent individual values less than and greater than the range of the whiskers.)

>100

** Gulf of
-100 -90

Longitude

Figure 2-34. Distance from the center of the 12-km grid cells to the nearest PM2.5
monitoring site for PM2.5 measurements from the AQS database and IMPROVE
network.

Concentrations less than 5 jj,g/m3 occur exclusively in the western U.S. for the downscaler
and HU2017 methods, and the western U.S. plus a few areas along the northern U.S. border in
the eastern U.S. for the DI2016 and VD2019 methods (Figure 2-35, top row). Concentrations
between 5 and 7 jjg/m3 are predicted in the western U.S. and parts of New England for all
methods and over Florida by the downscaler and DI2016 approaches (Figure 2-35, second row).
The CV among methods increases with decreasing concentration (Figure 2-33 above, panel b),
and the median CV is about 15% for grid cells with mean concentrations less than 7 Lig/nr. As
illustrated by Figure 2-33 and Figure 2-35, the low-concentration areas with relatively large CVs
are in the western U.S. and along the northern and southern border of the eastern U.S.

2-56

-------
downscaler

HU2017

VD2019

-120-110-100 -90 -80 -70-120-110-100-90 -80 -70-120-110-100 -90 -80 -70-120-110-100 -90 -80 70

Longitude

Figure 2-35. Location of PM2.5 predictions by range in annual average concentration for
the four prediction methods at their native resolution. (Note: Concentration ranges: < 5
Lig/ro 3, 5-7 ng/m3, 7-9 ug/m \ 9-11 |Jg/m3, and >11 ug/m\)

The comparison of PM2.5 concentrations across approaches was based on the 2011 period
due to the availability of predictions from multiple methods for that year. As discussed earlier in
this chapter, PM2.5 concentrations have declined over the U.S. in the last several decades. Annual
mean PM2.5 concentrations predicted by the VD2019 method for 2011 are compared with
predictions for 2001, 2006, and 2016 in Figure 2-36. The VD2019 fields capture the trend of
decreasing PM2.5 over the U.S. during this period, and the areas with annual mean PM2.5
concentration greater than 11 ug/m3 in 2016 are limited to California and southwest Arizona.

2-57

-------
2001 2006 2011 2016 Sms

!§-- nr • m- I

25~' r "i " i i i i 'T'" i i i i i i i i i i i i i i

-120-110-100-90 -80 -70-120-110-100 -90 -80 -70 -120-110-100 -90 -80 -70 -120-110-100 -90 -80 -70

Longitude 0

Figure 2-36. Annual mean PM2.5 from the VD2019 method (van Donkelaar et al., 2019) for
2001, 2006, 2011, and 2016.

2.3.3.2.4 Comparison of PM2.5 Fields in Estimating Exposure and Relative to
Design Values

Two types of hybrid approaches that have been utilized in several key PM2.5
epidemiologic studies in the 2019 ISA and ISA Supplement include neural network approaches
and use of GWR of residual PM2.5 with land-use and other variables to improve estimates of
PM2.5 concentration in the US. As such, we further compare these two types of approaches
across various scales and taking into account population weighting approaches utilized in
epidemiologic studies when estimating PM2.5 exposure. Additionally, we assess how average
PM2.5 concentrations computed using these hybrid surfaces compare to the maximum design
values measured at ground-based monitors. For this assessment, we evaluate the DI20193 ' and
HA202038 surfaces. This analysis may help to inform how the magnitude of the overall study
reported mean PM2.5 concentrations in epidemiologic studies may be influenced by the approach
used to compute that mean and how that value might compare to monitor reported
concentrations.

In estimating exposure, some studies focus on estimating concentrations in urban areas,
while others examine the entire U.S. or large portions of the country. Figure 2-37 shows the
spatial distribution of the annual average PM2.5 concentrations for 2015 using the DI2019 surface
nationwide (panel A) and for CBSAs only (panel B). As shown in the figure, the geographic
coverage is much less when estimating the annual average PM2.5 concentrations at the CBSA
scale compared to the national scale and tends to be primarily representative of areas that are

37 This analysis includes an updated version of the surface used in Di et al. (2016). Predictions in Di et al. (2016)
were for 2000 to 2012 using a neural network model. The Di et al. (2019) study improved on that effort in several
ways. First, a generalized additive model was used that accounted for geographic variations in performance to
combine predictions from three models (neural network, random forest, and gradient boosting) to make the final
optimal PM2.5 predictions. Second, the datasets were updated that were used in model training and included
additional variables such as 12-km CMAQ modeling as predictors. Finally, more recent years were included in
the Di et al. (2019) study.

38 The HA2020 field is based on the V4.NA.03 product available at: https://sites.wustl.edu/acag/datasets/surface-
pm2-5/. The name "HA2020" comes from the references for this product (Hammer et al., 2020; van Donkelaar et
al., 2019).

2-58

-------
more urban or densely populated. Further, the areas that are not included in the CBSA-only
analysis tend to have lower PM2.5 concentrations. These areas tend to be more rural or less
densely populated areas, and likely correspond to those locations where monitoring data
availability is limited or nonexistent.

-120 -110 -100 -90 -80 -70

Longitude

Figure 2-37. Spatial distribution of the annual average P1M2.5 concentrations for 2015 using
the 1)12019 surface nationwide (panel A) and for CBSAs only (panel B).

Using the DI2019 and HA2020 surfaces, for each year of available data, the 1 km x 1 km
grid cells for each modeled surface within a CBSA were averaged, resulting in an estimated
average annual PM2.5 concentration at the CBSA spatial resolution. In addition, for each surface,
all 1 km x 1 km grid cells were averaged over the conterminous U.S., resulting in an estimated
average annual PM2.5 concentration at the national scale. These average annual PM2.5

2-59

-------
concentrations for each year from 2000-2016 for the DI2019 and HA2020 surfaces are shown in
Table 2-5. In addition, we also examined the average annual PM2.5 concentrations nationwide
and in CBSAs in terms of a 3-year average, which is the averaging time of the annual standard.
These averages are shown in Table 2-6.

Table 2-5. Average Annual PM2.5 Concentration (jig/m3) by Year.

Year

DI2019

HA2020

Nationwidea

CBSAs b

Nationwidea

CBSAsb

2000

8.36

8.96

7.37

7.83

2001

7.88

8.49

7.08

7.61

2002

7.99

8.59

7.37

7.98

2003

8.25

8.72

7.03

7.51

2004

7.62

8.18

6.59

7.13

2005

7.98

8.51

7.34

7.92

2006

7.68

8.13

6.72

7.21

2007

7.90

8.41

7.26

7.69

2008

7.13

7.59

6.51

7.00

2009

6.52

6.94

6.02

6.45

2010

6.71

7.10

6.09

6.47

2011

6.72

7.13

6.31

6.74

2012

6.69

6.95

6.24

6.47

2013

6.15

6.50

5.75

6.14

2014

6.08

6.41

5.61

6.04

2015

6.00

6.25

5.43

5.76

2016

5.29

5.56

4.98

5.36

a Nationwide average annual PM2.5 concentrations include all 1 km x 1 km grid cells of the modeling surface.
b CBSA average annual PM2.5 concentrations include only those 1 km x 1 km grid cells that were located within a CBSA.

2-60

-------
Table 2-6. Three-Year Average of the Average Annual PM2.5 Concentrations (jig/m3).

Year

DI2019

HA2020

Nationwidea

CBSAsb

Nationwidea

CBSAsb

2000-2002

8.08

8.68

7.27

7.81

2001-2003

8.04

8.60

7.16

7.70

2002-2004

7.95

8.50

7.00

7.54

2003-2005

7.95

8.47

6.99

7.52

2004-2006

7.96

8.28

6.88

7.42

2005-2007

7.85

8.35

7.11

7.61

2006-2008

7.57

8.04

6.83

7.30

2007-2009

7.18

7.65

6.60

7.04

2008-2010

6.78

7.21

6.21

6.64

2009-2011

6.65

7.05

6.14

6.55

2010-2012

6.71

7.06

6.21

6.56

2011-2013

6.52

6.86

6.10

6.45

2012-2014

6.31

6.62

5.87

6.22

2013-2015

6.08

6.38

5.60

5.98

2014-2016

5.79

6.07

5.34

5.72

At the national scale, the average annual PM2.5 concentrations are slightly higher when
using the DI2019 surface compared to the HA2020 surface but are generally similar. The
average annual PM2.5 concentrations are also slightly lower using the HA2020 surface compared
to the DI2019 surface when the analyses are conducted for CBS As. However, regardless of
which surface is used, the average annual PM2.5 concentrations for the CBSA-only analyses are
somewhat higher than for the nationwide analyses (4-8% higher), likely reflecting the more
urban or densely populated areas in the CBSA-only analyses that typically have higher PM2.5 in
ambient air compared to more rural or less densely populated areas captured in the nationwide
analyses.

Similarly, as shown in Table 2-6, for both the DI2019 and HA2020 surfaces, the
nationwide average annual PM2.5 concentrations, averaged over three years, are lower than the
CBS A only average annual PM2.5 concentrations, averaged over three years. For the national
scale, 3-year averages of the average annual PM2.5 concentrations generally range from about 5.3
|ig/m3 to 8.1 |ig/m3, compared to the CBSA scale, which ranges from 5.7 |ig/m3 to 8.7 |ig/m3.

Overall, these analyses suggest that there are slight differences in the average annual
PM2.5 concentrations depending on the modeling method employed in a hybrid modeling study.

2-61

-------
It is important to recognize that the use of different methods in the hybrid modeling studies to
estimate mean PM2.5 concentrations may influence the comparability across studies

We next evaluate how the averages of the model surfaces compare to regulatory design
values and how population weighting influences the averages. For this analysis, we include
CBSAs with three or more valid design values for the 3-year period.39 The regulatory design
values for the CBSAs were calculated for each 3-year period for the CBSAs with 3 or more
design values in each of the 3-year periods. Using the maximum design value for each CBSA
and by each 3-year period, the ratio of maximum design values to modeled average annual PM2.5
concentrations were calculated, for each 3-year period. In addition, we evaluated the influence of
population weighting on the average annual PM2.5 concentrations using both the DI2019 and
HA2020 surfaces for 3-year periods in CBSAs that also have available regulatory design value
data. These data are shown in Table 2-7.

Table 2-7. Average Annual PM2.5 Concentrations and Ratios to Regulatory Design Values.

Years of
Monitoring
Data

No. of
CBSAs3

Average
Annual PM2.5
Concentration
(|jg/m3)b

Population
Weighted
Average Annual

PM2.5
Concentration
(|jg/m3)b

Average
Maximum
Annual
DVs
(|jg/m3)b

Average Ratio
of Maximum
Annual DVs to
Average Annual

PM2.5
Concentrations0

Average Ratio
of Maximum
Annual DVs to
Population
Weighted
Average Annual

PM2.5
Concentrations0

DI2019 Surface from Di et al. (2019)

2008-2010

8.61

10.17

11.67

1.48

1.15

2011-2013

8.10

9.37

10.91

1.47

1.17

2014-2016

7.22

8.26

9.57

1.41

1.17

HA2020 Surface from Hammer et al. (2020) and van Donkelaar et al. (2019)

2008-2010

8.25

9.93

11.67

1.50

1.18

2011-2013

7.92

9.34

10.91

1.43

1.17

2014-2016

6.98

8.19

9.57

1.43

1.18

a The number of CBSAs with 3 or more valid design values for the 3-year period
b Averaged across CBSAs

c Due to the order of operations, the averages of the ratios may not match the ratio of the averages across CBSAs

As shown in Table 2-7, the results using the DI2019 and HA2020 surfaces are similar for
the average annual PM2.5 concentrations, by each 3-year period. When population weighting is
not applied, the average annual PM2.5 concentrations generally range from 7.0 to 8.6 |ig/m3.
When population weighting is applied, the average annual PM2.5 concentrations are slightly
higher, ranging from 8.2 to 10.2 |ig/m3. As with CBSAs versus the national comparison above,

39 More details about the analytical methods used for this analysis are described in section A. 6 of Appendix A.

2-62

-------
population weighting results in a higher average PM2.5 concentration than when population
weighting is not applied.

For the CBS As included in the population weighted analyses, the average maximum
annual design values generally range from 9.5 to 11.7 |ig/m3. As shown in Table 2-7, these
analyses show that the results are similar for both the DI2019 and HA2020 surfaces and the
maximum annual PM2.5 design values are often 40% to 50% higher than average annual PM2.5
concentrations when population weighting is not applied. However, when population weighting
is applied, the ratio of the maximum annual PM2.5 design values to the average annual PM2.5
concentrations are lower than when not population weighted, and generally range from 15% to
18%.

2.3.3.2.5 Summary

Hybrid PM2.5 modeling methods have improved the ability to estimate PM2.5 exposure for
populations throughout the conterminous U.S. compared with the earlier approaches based on
monitoring data alone. Excellent performance in cross-validation tests suggests that hybrid
methods are reliable for estimating PM2.5 exposure in many applications. As discussed in
Chapter 3 of this PA, good agreement in health study results between monitor- and model-based
methods for urban areas (McGuinn et al., 2017) and general consistency in results for the
conterminous U.S. (Jerrett et al., 2017; Di et al., 2016) also suggests that the fields are reliable
for use in health studies. However, there are also important limitations associated with the
modeled fields. First, performance evaluations for the methods are weighted toward densely
monitored urban areas at the scales of representation of the monitoring networks. Predictions at
different scales or in sparsely monitored areas are relatively untested. Second, studies have
reported heterogeneity in performance with relatively weak performance in parts of the western
U.S., at low concentrations, at greater distance to monitors, and under conditions where the
reliability and availability of key input datasets (e.g., satellite retrievals and air quality modeling)
are limited. Differences in predictions among different hybrid methods have also been reported
and tend to be most important under conditions with the performance issues just noted.
Differences in predictions could also be related to the different approaches used to create long-
term PM2.5 fields (e.g., averaging daily PM2.5 fields vs. developing long-term average fields),
which is important due to variable monitoring schedules. More work is warranted on identifying
the most appropriate model performance metrics and comprehensively characterizing model
performance to further inform our understanding of the implications of using these fields to
estimate PM2.5 exposures in health studies.

When additional analyses are done to further compare the DI2019 and HA2020 surfaces,
the results suggest the DI2019 and HA2020 surfaces predict similar average annual PM2.5

2-63

-------
concentrations at the national scale and on average across all CBSAs in the U.S. The spatial scale
can affect the magnitude of the average annual PM2.5 concentration with somewhat higher
concentrations (4-8% higher) resulting from averaging across all CBSAs in the U.S. versus
averaging across the entire U.S. Additionally, when average annual PM2.5 concentrations from
the hybrid modeled surfaces are compared to the average maximum annual design value
measured at ground-based monitors in a subset of CBSAs, the average of the maximum annual
design values tends to be a 40-50% higher than the average annual PM2.5 concentration estimated
from the hybrid modeling surfaces. When population weighting is introduced, the average of the
maximum annual design values tends to only be 15-18%) higher than the average annual PM2.5
concentration estimated from the hybrid modeling surfaces. This analysis may help better
explain why reported study means from different epidemiologic studies can vary and why these
mean values tend to be lower than concentrations reported at ground-based monitors. However,
it is important to recognize that these results only reflect two surfaces and two types of
approaches and that the use of different hybrid methods to estimate mean PM2.5 concentrations
may influence the comparability across studies.

2.4 BACKGROUND PM

For the purposes of this assessment, we define background PM as all particles that are
formed by sources or processes that cannot be influenced by actions within the jurisdiction of
concern. For this document, U.S. background PM is defined as any PM formed from emissions
other than U.S. anthropogenic (i.e., manmade) emissions. Potential sources of U.S. background
PM include both natural sources (i.e., PM that would exist in the absence of any anthropogenic
emissions of PM or PM precursors) and transboundary sources originating outside U.S. borders.

Ambient monitoring networks provide long-term records of speciated PM concentrations
across the U.S., which can inform estimates of individual source contributions to background PM
levels in different parts of the country. However, even the most remote monitors within the U.S.
can be periodically affected by U.S. anthropogenic emissions. Monitor data are also limited in
more remote areas due to a sparser monitoring network where PM concentrations are more likely
influenced by background sources. Chemical transport models (CTMs) offer complementary
information to ambient monitor networks by providing more spatially and temporally
comprehensive estimates of atmospheric composition. CTMs can also be applied to isolate
contributions from specific emission sources to PM concentrations in different areas via source
apportionment or "zero-out" modeling (i.e., estimating what the residual concentrations would be
were emissions from the emission source of interest to be entirely removed).

At annual and national scales, estimated background PM concentrations in the U.S. are
small compared to contributions from domestic anthropogenic emissions. For example, based on

2-64

-------
zero-out modeling in the 2012 review of the PM NAAQS, annual background PM2.5
concentrations were estimated to range from 0.5 to 3 |ig/m3 across the sites examined. The
magnitude and sources of background PM can vary widely by region and time of year. Coastal
sites may experience a consistent contribution of PM from sea spray aerosol, while other areas
covered with dense vegetation may be impacted by biogenic aerosol production during the
summertime. Sources of background PM also operate across a range of time scales. While some
sources like biogenic aerosol vary at monthly to seasonal scales, many sources of background
PM are episodic in nature. These episodic sources (e.g., large wildfires) can be characterized by
infrequent contributions to high-concentration events occurring over shorter periods of time (e.g.,
hours to several days). Such episodic events are sporadic and do not necessarily occur in all
years. While these exceptional episodes can lead to violations of the daily PM2.5 standard (35
|ig/m3) in some cases (Schweizer et al., 2017), such events are routinely screened for and usually
identifiable in the monitoring data. As described further below, contributions to background PM
in the U.S. result mainly from sources within North America. Contributions from
intercontinental events have also been documented (e.g., transport from dust storms occurring in
deserts in North Africa and Asia), but these events are less common and represent a relatively
small fraction of background PM in most places.

While the potential sources of background PM discussed above include sources of both
fine (PM2.5) and coarse (PM10-2.5) particles, background contributions to ambient UFP are less
well characterized and are not discussed here due to lack of information. Section 2.4.1 below
further discusses background PM from natural sources inside the U.S. Section 2.4.2 characterizes
the role of international transport of PM from sources outside U.S. borders.

2.4.1 Natural Sources

As noted in section 2.1.1, sources that contribute to natural background PM include dust
from the wind erosion of natural surfaces, sea salt, wildland fires, primary biological aerosol
particles (PBAP) such as bacteria and pollen, oxidation of biogenic hydrocarbons such as
isoprene and terpenes to produce SOA, and geogenic sources such as sulfate formed from
volcanic production of SO2 and oceanic production of dimethyl-sulfide (DMS). While most of
the above sources release or contribute predominantly to fine aerosol, some sources including
windblown dust, and sea salt also produce particles in the coarse size range (U.S. EPA, 2019a,
section 2.3.3).

Biogenic emissions from plants are perhaps the most ubiquitous sources of background
PM in the U.S. Certain species of plants and trees can release large amounts of VOCs such as
isoprene and monoterpenes that are oxidized in the atmosphere to form organic aerosol. SOA
production from biogenic emissions is largest in the southeastern U.S., where conditions are

2-65

-------
warm, humid, and sunny for much of the year. Many of the processes involved with biogenic
SOA formation are complex and remain highly uncertain. Results from radiocarbon techniques
applied to distinguish modern (biogenic or fires) from fossil (anthropogenic) carbon fractions in
organic aerosol have suggested comparable contributions from both carbon types in the
Southeast where SOA concentrations are high (Schichtel et al., 2008). However, SOA formation
from biogenic emission sources can also be facilitated by the presence of anthropogenic
precursors (Marais et al., 2016; Xu et al., 2015). More work characterizing the interactions of
anthropogenic and biogenic emissions is needed to determine the implications of such processes
for background PM concentrations.

Soil dust and sea salt have been estimated to account for less than 10% of urban PM2.5 on
average in the U.S. (Karagulian et al., 2015), although episodic contributions from these sources
can be much higher in some locations. For example, during a dust storm affecting Phoenix in
July of 2011, peak hourly average PM10 concentrations were greater than 5,000 |ig/m3, with area-
wide average hourly concentrations ranging from a few hundred to a few thousand |ig/m3
(Vukovic et al., 2014). Dust can also account for much of the PM that originates from outside the
U.S., which we discuss further below (U.S. EPA, 2019a, section 2.5.4.2). In addition to sea salt
aerosol, biological production of the sulfate precursor DMS can also occur in some marine
environments, although the impact of DMS emissions on annual mean sulfate concentrations is
likely very small in the U.S. (<0.2 |ig/m3) and confined to coastal areas (Sarwar et al., 2018).

Wildfires release large amounts of particles and gaseous PM precursors. Invasive species,
historical fire management practices, frequency of drought, and extreme heat have resulted in
longer fire seasons (Jolly et al., 2015) and more large fires (Dennison et al., 2014) over time. In
addition to emissions from fires in the U.S., emissions from fires in other countries can be
transported to the U.S. Transport of smoke from fires in Canada, Mexico, Central America, and
Siberia have been documented in multiple studies (U.S. EPA, 2009). According to the NEI,
wildfire smoke contributes between 10 and 20% of primary PM emissions in the U.S. per year
(U.S. EPA, 2019a, section 2.3.1), with much higher localized contributions near fire-affected
areas.

To illustrate how episodic impacts from a large natural source can affect PM
concentrations in the U.S., Figure 2-38 and Figure 2-39 show an example from a recent wildfire
event. In summer 2017, smoke from wildfires in British Columbia, Canada led to severe air
quality degradation in parts of the Pacific Northwest. A NASA Worldview40 image from August
4, 2017 (Figure 2-38) shows smoke from multiple fire detections across southern British
Columbia crossing into northern Washington state. Smoke from these fires was also captured at

40 Available from https://worldview.earthdata.nasa.gov.

2-66

-------
the North Cascades IMPROVE monitor (Figure 2-39), where daily fine PM concentrations were
increased from a typical baseline of less than 10 lag/in ' to -100 |ig/m3 during this time.

Figure 2-38. Smoke and fire detections observed by the MODIS instrument onboard the
Aqua satellite on August 4th, 2017 accessed through NASA Worldview.

100

-------
Later in August and September 2017, many other wildfires occurred in Washington state
and Oregon, making this fire season one of the worst for the Pacific Northwest in recent history.
The severe fires in British Columbia, Washington and Oregon during 2017 have been linked to
the combination of usually hot temperatures in August/September in the region following a very
wet preceding winter season. While many of the most severe wildfire events in the U.S. occur in
the western part of the country during the late summer, most of the contiguous U.S. is affected
by wildfire smoke during some part of the year (Kaulfus et al., 2017).

2.4.2 International Transport

Background PM contributions from international sources include PM that is both natural
and anthropogenic in origin crossing U.S. borders from Canada and Mexico or from longer range
intercontinental transport. While in general the biggest contributions to U.S. background PM
from international sources come from nearby Canada and Mexico, large episodic events from
intercontinental sources can sometimes occur (e.g., windblown dust from Asia or Africa). This
section discusses transboundary PM transport within North America (section 2.4.2.1) as well as
long range intercontinental transport from anthropogenic (section 2.4.2.2) and natural (section
2.4.2.3) sources.

2.4.2.1 Transboundary Transport in North America

As discussed above, some of the largest potential international sources of U.S.
background PM originate elsewhere in North America. PM produced from fires in both Canada
and Mexico can affect air quality in the U.S., particularly in border states (Park et al., 2007;
Miller et al., 2011; Wang et al., 2018a). Anthropogenic emissions from Canada and Mexico can
also influence U.S. PM air quality. An inverse modeling study by Henze et al. (2009) estimated
that in 2001 anthropogenic SOx emissions from Canada and Mexico accounted for 6% and 4%
respectively of total daily inorganic PM2.5 in the U.S. These authors also estimated that SOx
emissions related to international shipping accounted for approximately 2% of total inorganic
PM in the U.S.

2.4.2.2 Long Range Transport from Anthropogenic Sources

Due to the relatively short atmospheric lifetime of particles (-days to weeks), long range
transport of aerosols does not contribute significant PM mass to the U.S. Heald et al. (2006)
estimated that transport from Asia accounted for less than 0.2 |ig/m3 of sulfate PM2.5 in the
Northwestern U.S. in spring, and Leibensperger et al. (2011) estimated intercontinental
contributions from Asian anthropogenic SO2 andNOx emissions of 0.1 - 0.25 |ig/m3 annually in
the western U.S. Leibensperger et al. (2011) also concluded that much of the intercontinental
influence captured by the GEOS-Chem model was in fact local PM production attributable to

2-68

-------
domestic emissions in receptor countries arising from changes in global oxidant budgets, rather
than impacts from PM directly transported across geopolitical boundaries. The studies above are
also consistent with findings from other analyses. A report from the United Nations on global air
quality synthesizing results across many studies estimated an annual average contribution of
approximately 0.1 |ig/m3 sulfate PM in North America due to transport from East Asia
(TFHTAP, 2006).

2.4.2.3 Long Range Transport from Natural Sources

Long range transport of dust from both Asia (Vancuren and Cahill, 2002; Yu et al., 2008)
and North Africa (Prospero, 1999a; Prospero, 1999b; Chiapello et al., 2005; McKendry et al.,
2007) has been shown to occasionally contribute to surface PM concentrations in some regions
of the U.S. The likelihood of such long-range dust transport events depends on large-scale
meteorological patterns, which can vary significantly across seasons and between years. Yu et al.
(2015) found that the transport of North African dust across the Atlantic Ocean is strongly
negatively correlated with precipitation in the Sahel during the preceding year. Dust from Africa
has also shown a decreasing trend of approximately 10% per decade from 1982 to 2008 based on
measurements of aerosol optical depth and surface concentrations in Barbados. This trend was
attributed to a corresponding decrease in surface winds over source regions (Ridley et al., 2014).
Variability in springtime Asian dust transport to the U.S. has been linked to north-south shifts in
trans-Pacific flow modulated by the El Nino-Southern Oscillation (Achakulwisut et al., 2017), as
well as to variations in regional precipitation affecting both dust emissions in Asia and
atmospheric residence times during transport (Fischer et al., 2009).

On average, intercontinental dust transport is estimated to contribute about 1-2 |ig/m3 to
annual PM2.5 at some U.S. sites (Jaffe et al., 2005; TFHTAP, 2006; Creamean et al., 2014).
However, daily concentrations can be substantially larger for individual events, especially for
coarser particles. For example, Jaffe et al. (2003) found evidence of Asian dust events in 1998
and 2001 contributing 30-40 |ig/m3 to daily PM10 at sites throughout the U.S., although the
authors also note that large events of this scale are rare and only occurred twice during their 15-
year study period. Similar magnitudes have also been reported for individual North African
events; analysis of a multidecadal record of African dust reaching Miami indicated
concentrations of PMranging from -10 to 120 |ig/m3 (Prospero, 1999b; Prospero, 1999a).42 In
June 2020 a large dust transport episode originating in North Africa may have contributed up to
50 |ig/m3 for several days at multiple sites in the southeastern U.S. (Pu and Jin, 2021).

42 Sample collection began in 1974, before network PM10 and PM2 5 samplers were developed, and no size cut was
specified (Prospero, 1999b).

2-69

-------
2.4.3 Estimating Background PM with Recent Data

As discussed above, the 2009 PM ISA estimated background PM concentrations at
several remote IMPROVE sites in different regions of the U.S. for 2004 using a combination of
monitor data and zero-out air quality modeling. Revisiting the speciated IMPROVE PM data at
the monitors included in the 2009 ISA assessment provides some insights into how contributions
from different PM sources may have changed, and what those changes (or lack thereof) mean for
our current understanding of background PM in the U.S.

Figure 2-40 shows observed annual average PM2.5 in 2004 and 2016 at the same remote
monitors examined in the 2009 ISA. The comparisons show decreases in both total PM2.5 and
ammonium sulfate across all sites examined, consistent with decreases in anthropogenic SO2 and
other PM precursors observed over this time period. It is likely that most of the remaining
ammonium sulfate observed at these sites is also a result of domestic anthropogenic emissions
and therefore not relevant for assessments of background PM.

Sea salt and dust aerosol are likely natural in origin at these remote sites. With the
exception of REDW1, a coastal site in California, soil and sea salt aerosol together account for
less than about 0.5 |ig/m3 of the annual average PM2.5 at all monitors examined here, which is
below the values cited from the literature for long range dust contributions discussed above.
Contributions from ammonium nitrate and elemental carbon could be from either anthropogenic
or natural sources, but together represent less than about 0.5 |ig/m3 at most of the sites in 2016.
The largest contribution from nitrate occurs at the BRIG1 monitor in New Jersey and is likely
anthropogenic given the high density of NOx from vehicle emissions in that region.

After ammonium sulfate, the next largest contributing species for most of the sites is
organic matter, which for many of the monitors in Figure 2-40 represents 50% or more of total
PM in both 2004 and 2016. In addition to the IMPROVE sites from the 2019 ISA, Figure 2-40
also shows comparisons for three sites in the Southeast U.S. As a region, the Southeast has the
highest levels of biogenic aerosol production in the country, so the organic matter contribution at
these three sites likely represents an upper bound for the country of what natural biogenic
organic aerosol production could be under present atmospheric conditions. The organic aerosol
components shown in Figure 2-40 will also include the influence of fires for some monitors. The
average organic matter contribution in the IMPROVE data during the full 2004-2016 period for
almost all of the sites shown in Figure 2-40,is approximately 2 |ig/m3 or lower. The SIPS1 and
OKEF1 monitors in the Southeast both experience slightly higher average organic matter
concentrations around 3 |ig/m3 for this same period, values that include contributions from
wildfire smoke in some years.

While contributions from ammonium sulfate have decreased substantially at some of the
monitors, particularly the eastern sites, contributions from organic aerosol are roughly consistent

2-70

-------
between 2004 and 2016, as are the contributions from the other species assumed to be mostly
natural in origin (soil and sea salt). Although in Figure 2-40 the organic matter concentrations
appear to be decreasing between 2004 and 2016 for some of the sites, looking at all intervening
year during the 2004-2016 period shows no statistically significant trend for any of the sites
except for DOSOl (downward trend of-0.05 |ig/m3 per year). Therefore, while no new zero-out
modeling was done for the reconsideration, revisiting these monitors with more recent data
suggests that estimates of background concentrations at these monitors are still around 1-3 |ig/m3
and have not changed significantly since the 2012 PM NAAQS review.

While estimates of total annual background concentrations have generally not changed
significantly since the 2012 review, our scientific understanding of organic aerosol formation has
evolved. Organic aerosol can be produced from a variety of natural and anthropogenic processes,
which presents a challenge for source attribution techniques. Additionally, new research over the
past decade has identified a host of new sources and chemical pathways for SOA formation that
have only recently begun to be implemented into CTMs. Further research implementing these
new sources and pathways into CTMs is needed to understand 1) the behavior of these different
algorithms under a range of possible atmospheric conditions, and 2) what the implications are for
understanding SOA formation in the U.S.

2-71

-------
ACAD1

BRIG1

D0SO1

V0YA2

2054 2016

BRID1

15"

2034 25316

REDW1

2034 2516

2004 2016

CANY 1

2004 2310

GICL1

15-

15-
10"

0-1

2004 2010

OKEF1

2034 2310

SHR01

2034

2310

Amm. nit.

Amm. sulf.

2034 2016

2304 2010

GLAC1

15-

2004 2010

SIPS1

2034 2010

Figure 2-40. Speciated annual average IMPROVE PM2.5 in jig/m3 at select remote monitors
during 2004 and 2016. (Note: Monitor locations are shown in Figure 2-41.)

Figure 2-41. Site locations for the IMPROVE monitors in Figure 2-40. (Note: Monitors also
assessed in the 2009 ISA are shown in blue. Monitors only examined in this assessment are
shown in red.)

2-72

-------
REFERENCES

Abt Associates, Inc. (2014). User's guide: modeled attainment test software. U.S. Environmental
Protection Agency. Research Triangle Park, NC.

Achakulwisut, P, Shen, L and Mickley, LJ (2017). What controls springtime fine dust variability
in the western United States? Investigating the 2002-2015 increase in fine dust in the US
southwest. Journal of Geophysical Research: Atmospheres 122(22): 12449-12467.

Berrocal, VJ, Gelfand, AE and Holland, DM (2012). Space-time data fusion under error in

computer model output: An application to modeling air quality. Biometrics 68(3): 837-
848.

Berrocal, VJ, Guan, Y, Muyskens, A, Wang, H, Reich, BJ, Mulholland, JA and Chang, HH

(2020). A comparison of statistical and machine learning methods for creating national
daily maps of ambient PM2. 5 concentration. Atmospheric Environment 222: 117130.

Boys, BL, Martin, RV, van Donkelaar, A, MacDonell, RJ, Hsu, NC, Cooper, MJ, Yantosca, RM,
Lu, Z, Streets, DG, Zhang, Q and Wang, SW (2014). Fifteen-year global time series of
satellite-derived fine particulate matter. Environmental Science and Technology 48(19):
11109-11118.

Budisulistiorini, SH, Li, X, Bairai, ST, Renfro, J, Liu, Y, Liu, YJ, McKinney, KA, Martin, ST,
McNeill, VF, Pye, HOT, Nenes, A, Neff, ME, Stone, EA, Mueller, S, Knote, C, Shaw,
SL, Zhang, Z, Gold, A and Surratt, JD (2015). Examining the effects of anthropogenic
emissions on isoprene-derived secondary organic aerosol formation during the 2013
Southern Oxidant and Aerosol Study (SOAS) at the Look Rock, Tennessee ground site.
Atmos. Chem. Phys. 15(15): 8871-8888.

Carlton, AG, Pinder, RW, Bhave, PV and Pouliot, GA (2010). To What Extent Can Biogenic
SOA be Controlled? Environmental Science & Technology 44(9): 3376-3380.

Carlton, AG, Pye, HOT, Baker, KR and Hennigan, CJ (2018). Additional Benefits of Federal
Air-Quality Rules: Model Estimates of Controllable Biogenic Secondary Organic
Aerosol. Environmental Science & Technology 52(16): 9254-9265.

Chan, EAW, Gantt, B and McDow, S (2018). The reduction of summer sulfate and switch from
summertime to wintertime PM2.5 concentration maxima in the United States.

Atmospheric Environment 175: 25-32.

Chiapello, I, Moulin, C and Prospero, JM (2005). Understanding the long-term variability of
African dust transport across the Atlantic as recorded in both Barbados surface
concentrations and large-scale Total Ozone Mapping Spectrometer (TOMS) optical
thickness. Journal of Geophysical Research 110(D18): D18S10.

Creamean, JM, Spackman, JR, Davis, SM and White, AB (2014). Climatology of long-range

transported Asian dust along the West Coast of the United States. Journal of Geophysical
Research: Atmospheres 119(21): 12171-12185.

2-73

-------
Day, M, Pouliot, G, Hunt, S, Baker, KR, Beardsley, M, Frost, G, Mobley, D, Simon, H,

Henderson, B and Yelverton, T (2019). Reflecting on progress since the 2005 NARSTO
emissions inventory report. Journal of the Air Waste Management Association 69(9):
1023-1048.

Dennison, PE, Brewer, SC, Arnold, JD and Moritz, MA (2014). Large wildfire trends in the
western United States, 1984-2011. Geophysical Research Letters 41(8): 2928-2933.

DeWinter, J, Brown, S, Seagram, A, Landsberg, K and Eisinger, D (2018). A national-scale

review of air pollutant concentrations measured in the U.S. near-road monitoring network
during 2014 and 2015. Atmospheric Environment 183.

DHEW (1969). Air Quality Criteria for Particulate Matter. National Air Pollution Control
Administration. Washington, D.C. U.S. Department of Health. January 1969.

Di, Q, Amini, H, Shi, L, Kloog, I, Silvern, R, Kelly, J, Sabath, MB, Choirat, C, Koutrakis, P and
Lyapustin, A (2019). An ensemble-based model of PM2.5 concentration across the
contiguous United States with high spatiotemporal resolution. Environment International
130: 104909.

Di, Q, Kloog, I, Koutrakis, P, Lyapustin, A, Wang, Y and Schwartz, J (2016). Assessing PM2.5
exposures with high spatiotemporal resolution across the Continental United States.
Environmental Science and Technology 50(9): 4712-4721.

Diao, M, Holloway, T, Choi, S, O'Neill, SM, Al-Hamdan, MZ, Van Donkelaar, A, Martin, RV,
Jin, X, Fiore, AM and Henze, DK (2019). Methods, availability, and applications of
PM2. 5 exposure estimates derived from ground measurements, satellite, and atmospheric
models. Journal of the Air & Waste Management Association 69(12): 1391-1414.

Fischer, EV, Hsu, NC, Jaffe, DA, Jeong, MJ and Gong, SL (2009). A decade of dust: Asian dust
and springtime aerosol load in the US Pacific Northwest. Geophysical Research Letters
36.

Friberg, MD, Zhai, X, Holmes, HA, Chang, HH, Strickland, MJ, Sarnat, SE, Tolbert, PE,

Russell, AG and Mulholland, JA (2016). Method for fusing observational data and
chemical transport model simulations to estimate spatiotemporally resolved ambient air
pollution. Environmental Science and Technology 50(7): 3695-3705.

Gantt, B, Owen, RC and Watkins, N (2021). Characterizing Nitrogen Oxides and Fine

Particulate Matter near Major Highways in the United States Using the National Near-
Road Monitoring Network. Environmental science & technology 55(5): 2831-2838.

Geng, G, Murray, NL, Tong, D, Fu, JS, Hu, X, Lee, P, Meng, X, Chang, HH and Liu, Y (2018).
Satellite-based daily PM2.5 estimates during fire seasons in Colorado. Journal of
Geophysical Research: Atmospheres 123(15): 8159-8171.

Hammer, MS, van Donkelaar, A, Li, C, Lyapustin, A, Sayer, AM, Hsu, NC, Levy, RC, Garay,
MJ, Kalashnikova, OV and Kahn, RA (2020). Global estimates and long-term trends of

2-74

-------
fine particulate matter concentrations (1998-2018). Environmental Science &
Technology 54(13): 7879-7890.

Hand, JL, Schichtel, BA, Malm, WC and Pitchford, ML (2012). Particulate sulfate ion

concentration and S02 emission trends in the United States from the early 1990s through
2010. Atmospheric Chemistry and Physics 12(21): 10353-10365.

Heald, CL, Jacob, DJ, Park, RJ, Alexander, B, Fairlie, TD, Yantosca, RM and Chu, DA (2006).
Transpacific transport of Asian anthropogenic aerosols and its impact on surface air
quality in the United States. Journal of Geophysical Research: Atmospheres 111(D14).

Henze, DK, Seinfeld, JH and Shindell, DT (2009). Inverse modeling and maing US air quality
influences of inorganic PM(2.5) precursor emissions using the adjoint of GEOS-Chem.
Atmospheric Chemistry and Physics 9(16): 5877-5903.

Hu, X, Belle, JH, Meng, X, Wildani, A, Waller, LA, Strickland, MJ and Liu, Y (2017).

Estimating PM2.5 Concentrations in the Conterminous United States Using the Random
Forest Approach. Environmental Science and Technology 51(12): 6936-6944.

Hu, X, Waller, LA, Lyapustin, A, Wang, Y and Liu, Y (2014). 10-year spatial and temporal
trends of PM2.5 concentrations in the southeastern US estimated using high-resolution
satellite data. Atmospheric Chemistry and Physics 14(12): 6301-6314.

Huang, Ra, Zhai, X, Ivey, CE, Friberg, MD, Hu, X, Liu, Y, Di, Q, Schwartz, J, Mulholland, JA
and Russell, AG (2018). Air pollutant exposure field modeling using air quality model-
data fusion methods and comparison with satellite AOD-derived fields: application over
North Carolina, USA. Air Quality, Atmosphere and Health 11(1): 11-22.

Jaffe, D, Snow, J and Cooper, O (2003). The 2001 Asian Dust Events: Transport and impact on
surface aerosol concentrations in the U.S. Eos 84(46): 501-506.

Jaffe, D, Tamura, S and Harris, J (2005). Seasonal cycle and composition of background fine
particles along the west coast of the US. Atmospheric Environment 39(2): 297-306.

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 and
Burnett, RT (2017). Comparing the health effects of ambient particulate matter estimated
using ground-based versus remote sensing exposure estimates. Environmental health
perspectives 125(4): 552-559.

Jin, X, Fiore, AM, Civerolo, K, Bi, J, Liu, Y, van Donkelaar, A, Martin, RV, Al-Hamdan, M,
Zhang, Y, Insaf, TZ, Kioumourtzoglou, M-A, He, MZ and Kinney, PL (2019).
Comparison of multiple PM2.5 exposure products for estimating health benefits of
emission controls over New York State, USA. Environmental Research Letters 14(8):
084023.

2-75

-------
Jolly, WM, Cochrane, MA, Freeborn, PH, Holden, ZA, Brown, TJ, Williamson, GJ and

Bowman, DMJS (2015). Climate-induced variations in global wildfire danger from 1979
to 2013. Nature Communications 6.

Just, AC, Arfer, KB, Rush, J, Dorman, M, Shtein, A, Lyapustin, A and Kloog, I (2020).

Advancing methodologies for applying machine learning and evaluating spatiotemporal
models of fine particulate matter (PM2. 5) using satellite data over large regions.
Atmospheric Environment 239: 117649.

Karagulian, F, Belis, CA, Dora, CFC, Pruess-Ustuen, AM, Bonjour, S, Adair-Rohani, H and

Amann, M (2015). Contributions to cities' ambient particulate matter (PM): A systematic
review of local source contributions at global level. Atmospheric Environment 120: 475-
483.

Karl, T and Koss, WJ (1984). Regional and national monthly, seasonal, and annual temperature
weighted by area, 1895-1983. National Climatic Data Center. Asheville, NC.

Kaulfus, AS, Nair, U, Jaffe, D, Christopher, SA and Goodrick, S (2017). Biomass burning smoke
climatology of the United States: implications for particulate matter air quality.
Environmental Science and Technology 51(20): 11731-11741.

Keller, JP and Peng, RD (2019). Error in estimating area-level air pollution exposures for
epidemiology. Environmetrics 0(0): e2573.

Kelly, JT, Jang, C, Timin, B, Di, Q, Schwartz, J, Liu, Y, van Donkelaar, A, Martin, RV,

Berrocal, V and Bell, ML (2021). Examining PM2.5 concentrations and exposure using
multiple models. Environmental Research 196: 110432.

Kelly, JT, Jang, CJ, Timin, B, Gantt, B, Reff, A, Zhu, Y, Long, S and Hanna, A (2019a). A

system for developing and projecting PM2.5 spatial fields to correspond to just meeting
National Ambient Air Quality Standards. Atmospheric Environment: X 2: 100019.

Kelly, JT, Koplitz, SN, Baker, KR, Holder, AL, Pye, HOT, Murphy, BN, Bash, JO, Henderson,
BH, Possiel, NC, Simon, H, Eyth, AM, Jang, CJ, Phillips, S and Timin, B (2019b).
Assessing PM2.5 model performance for the conterminous U.S. with comparison to model
performance statistics from 2007-2015. Atmospheric Environment 214: 116872.

Leibensperger, EM, Mickley, LJ, Jacob, DJ and Barrett, SRH (2011). Intercontinental influence
of NO(x) and CO emissions on particulate matter air quality. Atmospheric Environment
45(19): 3318-3324.

Li, Y, Mickley, LJ and Kaplan, JO (2021). Response of dust emissions in southwestern North
America to 21st century trends in climate, C02 fertilization, and land use: implications
for air quality. Atmos. Chem. Phys. 21(1): 57-68.

Liu, Y, Goodrick, S and Heilman, W (2014). Wildland fire emissions, carbon, and climate:
Wildfire-climate interactions. Forest Ecology and Management 317: 80-96.

2-76

-------
Marais, EA, Jacob, DJ, Jimenez, JL, Campuzano-Jost, P, Day, DA, Hu, W, Krechmer, J, Zhu, L,
Kim, PS, Miller, CC, Fisher, JA, Travis, K, Yu, K, Hanisco, TF, Wolfe, GM, Arkinson,
HL, Pye, HOT, Froyd, KD, Liao, J and McNeill, VF (2016). Aqueous-phase mechanism
for secondary organic aerosol formation from isoprene: application to the southeast
United States and co-benefit of S02 emission controls. Atmos. Chem. Phys. 16(3): 1603-
1618.

McGuinn, LA, Ward-Caviness, C, Neas, LM, Schneider, A, Di, Q, Chudnovsky, A, Schwartz, J,
Koutrakis, P, Russell, AG, Garcia, V, Kraus, WE, Hauser, ER, Cascio, W, Diaz-Sanchez,
D and Devlin, RB (2017). Fine particulate matter and cardiovascular disease:

Comparison of assessment methods for long-term exposure. Environmental Research
159: 16-23.

McKendry, IG, Strawbridge, KB, O'Neill, NT, MacDonald, AM, Liu, PSK, Leaitch, WR,

Anlauf, KG, Jaegle, L, Fairlie, TD and Westphal, DL (2007). Trans-Pacific transport of
Saharan dust to western North America: A case study. Journal of Geophysical Research
112(D1): D01103.

Miller, DJ, Sun, K, Zondlo, MA, Kanter, D, Dubovik, O, Welton, EJ, Winker, DM and Ginoux,
P (2011). Assessing boreal forest fire smoke aerosol impacts on U.S. air quality: A case
study using multiple data sets. Journal of Geophysical Research 116(D22): D22209.

Park, RJ, Jacob, DJ and Logan, JA (2007). Fire and biofuel contributions to annual mean aerosol
mass concentrations in the United States. Atmospheric Environment 41(35): 7389-7400.

Prospero, JM (1999a). Long-range transport of mineral dust in the global atmosphere: Impact of
African dust on the environment of the southeastern United States. Proceedings of the
National Academy of Sciences 96(7): 3396-3403.

Prospero, JM (1999b). Long-term measurements of the transport of African mineral dust to the
southeastern United States: implications for regional air quality. Journal of Geophysical
Research 104(D13): 15917-15927.

Pu, B and Ginoux, P (2017). Projection of American dustiness in the late 21st century due to
climate change. Scientific Reports 7(1): 5553.

Pu, B and Jin, Q (2021). A Record-Breaking Trans-Atlantic African Dust Plume Associated with
Atmospheric Circulation Extremes in June 2020 %J Bulletin of the American
Meteorological Society. 102(7): E1340-E1356.

Ridley, DA, Heald, CL and Prospero, JM (2014). What controls the recent changes in African
mineral dust aerosol across the Atlantic? Atmospheric Chemistry and Physics 14(11):
5735-5747.

Saha, PK, Robinson, ES, Shah, RU, Zimmerman, N, Apte, JS, Robinson, AL and Presto, AA
(2018). Reduced ultrafine particle concentration in urban air: changes in nucleation and
anthropogenic emissions. Environmental Science and Technology 52(12): 6798-6806.

2-77

-------
Sarwar, G, Xing, Ji, Fahey, K, Foley, K, Wong, D, Mathur, R, Gan, CM, Gantt, B and Simon, H
(2018). Dimethylsulfide Chemistry: Annual, Seasonal, and Spatial Impacts on Sulfate.
International Technical Meeting on Air Pollution Modelling and its Application: 347-
352.

Schichtel, BA, Malm, WC, Bench, G, Fallon, S, McDade, CE, Chow, JC and Watson, JG (2008).
Fossil and contemporary fine particulate carbon fractions at 12 rural and urban sites in the
United States. Journal of Geophysical Research: Atmospheres 113(D2).

Schoennagel, T, Balch, JK, Brenkert-Smith, H, Dennison, PE, Harvey, BJ, Krawchuk, MA,
Mietkiewicz, N, Morgan, P, Moritz, MA, Rasker, R, Turner, MG and Whitlock, C
(2017). Adapt to more wildfire in western North American forests as climate changes.
Proceedings of the National Academy of Sciences 114(18): 4582.

Schweizer, D, Cisneros, R, Traina, S, Ghezzehei, TA and Shaw, G (2017). Using National

Ambient Air Quality Standards for fine particulate matter to assess regional wildland fire
smoke and air quality management. Journal of Environmental Management 201: 345-
356.

Sun, K, Tao, L, Miller, DJ, Khan, MA and Zondlo, MA (2014). On-Road Ammonia Emissions
Characterized by Mobile, Open-Path Measurements. Environmental Science &
Technology 48(7): 3943-3950.

TFHTAP (2006). Hemispheric transport of air pollution 2010. Part A: Ozone and particulate

matter. Air Pollution Studies No. 17. United Nations Publications. Geneva, Switzerland.

U.S. EPA (2009). Integrated Science Assessment for Particulate Matter (Final Report). Office of
Research and Development, National Center for Environmental Assessment. Research
Triangle Park, NC. U.S. EPA. EPA-600/R-08-139F. December 2009. Available at:
https://cfpub.epa. gov/ncea/risk/recordisplav.cfm?deid=216546.

U.S. EPA (2012). Near-road N02 monitoring technical assistance document. U.S.

Environmental Protection Agency, Office of Air Quality Planning and Standards.
Research Triangle Park, NC. U.S. EPA. EPA-454/B-12-002. June 2012. Available at:
http://nepis.epa. gov/exe/ZvPURL.cgi?Dockev=P100EPRZ.txt.

U.S. EPA (2015). Pilot study on coarse PM monitoring. EPA-454/R-15-001. U.S. EPA Office of
Air Quality Planning and Standards. Research Triangle Park, NC.

U.S. EPA (2017). Bayesian space-time downscaling fusion model (downscaler) -Derived

Estimates of Air Quality for 2013. Office of Air Quality Planning and Standards, Office
of Air and Radiation. Research Triangle Park, NC. U.S. EPA. EPA-450/R-17-001.
Available: https://www.epa.gov/hesc/rsig-related-downloadable-data-files.

U.S. EPA (2018). Integrated Science Assessment (ISA) for Particulate Matter (External Review
Draft). U.S. Environmental Protection Agency, Office of Research and Development,
National Center for Environmental Assessment. Washington, DC. U.S. EPA.

2-78

-------
EPA/600/R-18/179. October 2018. Available at:
https://cfpub.epa.gov/ncea/isa/recordisplav.cfm?deid=341593.

U.S. EPA (2019a). Integrated Science Assessment (ISA) for Particulate Matter (Final Report).
U.S. Environmental Protection Agency, Office of Research and Development, National
Center for Environmental Assessment. Washington, DC. U.S. EPA. EPA/600/R-19/188.
December 2019. Available at: https://www.epa.gov/naaqs/particulate-matter-pm-
standards-integrated-science-assessments-current-review.

U.S. EPA (2019b). Review of Sunset Organic and Elemental Carbon (OC and EC)

Measurements During EPA's Sunset Carbon Evaluation Project. U.S. Environmental
Protection Agency, Office of Air Quality Planning and Standards. Research Triangle
Park, NC. U.S. EPA. EPA-454/R-19-005. May 2019. Available at:
https://www.epa.gov/sites/production/files/2019-05/documents/sunset epa-454r-19-
005 l.pdf.

U.S. EPA (2020). Addendum SPECIATE Verison 5.1 Database Development Documentation.

U.S. Environmental Protection Agency. Washington, DC. U.S. EPA. EPA-600/R-20/189.
June 2020. Available at: https://www.epa.gov/sites/default/files/202Q-
07/documents/speciate 5.l.pdf.

U.S. EPA (2021). 2017 National Emissions Inventory: January 2021 Updated Release, Technical
Support Document. U.S. Environmental Protection Agency, Office of Air Quality
Planning and Standards. Research Triangle Park, NC. U.S. EPA. February 2021.
Available at: https://www.epa.gov/sites/production/files/2Q21-
02/documents/nei2017 tsd full ian2021.pdf.

van Donkelaar, A, Martin, RV, Brauer, M and Boys, BL (2015a). Use of satellite observations
for long-term exposure assessment of global concentrations of fine particulate matter.
Environmental Health Perspectives 123(2): 135-143.

van Donkelaar, A, Martin, RV, Brauer, M, Kahn, R, Levy, R, Verduzco, C and Villeneuve, PJ
(2010). Global estimates of ambient fine particulate matter concentrations from satellite-
based aerosol optical depth: development and application. Environmental Health
Perspectives 118(6): 847-855.

van Donkelaar, A, Martin, RV, Li, C and 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. Environmental Science &
Technology 53(5).

van Donkelaar, A, Martin, RV, Spurr, RJ and Burnett, RT (2015b). High-resolution satellite-
derived PM2.5 from optimal estimation and geographically weighted regression over
North America. Environmental Science and Technology 49(17): 10482-10491.

van Donkelaar, A, Martin, RV, Spurr, RJD, Drury, E, Remer, LA, Levy, RC and Wang, J (2013).
Optimal estimation for global ground-level fine particulate matter concentrations. Journal
of Geophysical Research: Atmospheres 118(11): 5621-5636.

2-79

-------
Vancuren, RA and Cahill, TA (2002). Asian aerosols in North America: Frequency and
concentration of fine dust. Journal of Geophysical Research 107(D24): 4804.

Vu, TV, Delgado-Saborit, JM and Harrison, RM (2015). Review: Particle number size

distributions from seven major sources and implications for source apportionment
studies. Atmospheric Environment 122: 114-132.

Vukovic, A, Vujadinovic, M, Pejanovic, G, Andric, J, Kumjian, MR, Djurdjevic, V, Dacic, M,

Prasad, AK, El-Askary, HM, Paris, BC, Petkovic, S, Nickovic, S and Sprigg, WA (2014).
Numerical simulation of "an American haboob". Atmospheric Chemistry and Physics
14(7): 3211-3230.

Wang, S-C, Wang, Y, Estes, M, Lei, R, Talbot, R, Zhu, L and Hou, P (2018a). Transport of
Central American fire emissions to the U.S. Gulf Coast: climatological pathways and
impacts on ozone and PM2.5. Journal of Geophysical Research: Atmospheres 123(15):
8344-8361.

Wang, Y, Hopke, PK, Chalupa, DC and Utell, MJ (2011). Long-term study of urban ultrafine
particles and other pollutants. Atmospheric Environment 45(40): 7672-7680.

Wang, Y, Hu, X, Chang, H, Waller, L, Belle, J and Liu, Y (2018b). A Bayesian downscaler
model to estimate daily PM2.5 levels in the conterminous U.S. International Journal of
Environmental Research and Public Health 15(9): 1999.

Whitby, KT, Husar, RB and Liu, BYH (1972). The aerosol size distribution of Los Angeles
smog. Journal of Colloid and Interface Science 39: 177-204.

Xu, L, Guo, H, Boyd, CM, Klein, M, Bougiatioti, A, Cerully, KM, Hite, JR, Isaacman-Vanwertz,
G, Kreisberg, NM, Knote, C, Olson, K, Koss, A, Goldstein, AH, Hering, SV, de Gouw, J,
Baumann, K, Lee, SH, Nenes, A, Weber, RJ and Ng, NL (2015). Effects of
anthropogenic emissions on aerosol formation from isoprene and monoterpenes in the
southeastern United States. Proceedings of the National Academy of Sciences 112(1): 37-
42.

Yu, H, Chin, M, Bian, H, Yuan, T, Prospero, JM, Omar, AH, Remer, LA, Winker, DM, Yang, Y,
Zhang, Ya and Zhang, Z (2015). Quantification of trans-Atlantic dust transport from
seven-year (2007-2013) record of CALIPSO lidar measurements. Remote Sensing of
Environment 159: 232-249.

Yu, H, Remer, LA, Chin, M, Bian, H, Kleidman, RG and Diehl, T (2008). A satellite-based
assessment of transpacific transport of pollution aerosol. Journal of Geophysical
Research 113(D14): 1-15.

Yu, H, Russell, A, Mulholland, J, Odman, T, Hu, Y, Chang, HH and Kumar, N (2018). Cross-
comparison and evaluation of air pollution field estimation methods. Atmospheric
Environment 179: 49-60.

2-80

-------
3 RECONSIDERATION OF THE PRIMARY
STANDARDS FOR PM2 5

This chapter presents and evaluates the policy implications of the key aspects of the
scientific and technical information pertaining to this reconsideration of the primary PM2.5
standards. In so doing, this chapter presents key aspects of the evidence of health effects of
PM2.5, as documented in the 2019 ISA (U.S. EPA, 2019) and the ISA Supplement (U.S. EPA,
2022),1 with support from the prior ISA and AQCDs, and associated public health implications.
It also presents key aspects of updated quantitative risk analyses conducted for this
reconsideration, as detailed in the appendices associated with this chapter. Together this
information provides the basis for our evaluation of the scientific information regarding health
effects of PM2.5 in ambient air and the potential for effects to occur under air quality conditions
associated with the existing standard (or any alternatives considered), as well as the associated
implications for public health. Our evaluation is focused on key policy-relevant questions
derived from the IRP (U.S. EPA, 2016, section 2.1) for the review completed in 2020, and also
takes into account conclusions reached in previous reviews. In this way we identify key policy-
relevant considerations and summary conclusions regarding the public health protection provided
by the current standards for the Administrator's consideration in this reconsideration of the 2020
final decision on the primary PM2.5 standards.

Within this chapter, background information on the current standards is summarized in
section 3.1. The general approach for considering the available information in this
reconsideration, including policy-relevant questions identified to frame our policy evaluation, is
summarized in section 3.2. Key aspects of the available health effects evidence and associated
public health implications and uncertainties are addressed in section 3.3, and the current air
quality and risk information, with associated uncertainties, is addressed in section 3.4. Section
3.5 summarizes CASAC advice and public comments, and section 3.6 summarizes the key
evidence- and risk-based considerations identified in our evaluation and also presents associated
conclusions on the adequacy of the current standards. Key remaining uncertainties and areas for
future research are identified in section 3.7.

1 As described in detail in section 1.4.2 above and section 3.3 below, the ISA Supplement focuses on a thorough
evaluation of some studies that became available after the literature cutoff date of the 2019 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 ISA (U.S. EPA, 2022). The selection of the health effects to evaluate within
the ISA Supplement was based on the health effects for which the evidence in the 2019 ISA supported a "causal
relationship" and the subsequent use of scientific evidence in the 2020 PA. Specifically, for PM2 5-related health
effects, the focus within the ISA Supplement is on mortality and cardiovascular effects. The ISA Supplement
does not include an evaluation of studies for other PM2 5-related health effects (U.S. EPA, 2022).

3-1

-------
3.1 BACKGROUND ON THE CURRENT STANDARDS

The current primary PM2.5 standards were retained in 2020 based on the Administrator's
judgments regarding the available scientific evidence, the available risk information regarding
the risk that may be allowed by such standards, and the appropriate degree of public health
protection provided by the existing standards (85 FR 82718, December 18, 2020). With the 2020
final decision, the EPA retained the primary 24-hour PM2.5 standard, with its level of 35 |ig/m3,
and the primary annual PM2.5 standard, with its level of 12.0 |ig/m3. This decision drew upon the
scientific evidence assessed in the 2019 ISA, the evidence and quantitative risk information in
the 2020 PA, the advice and recommendations of the CASAC, and public comments on the
proposed decision (85 FR 24094, April 30, 2020).

The health effects evidence base available in the 2020 review included extensive
evidence from previous reviews as well as the evidence that had emerged since the prior review
had been completed in 2012. This evidence base, spanning several decades, documents the
relationship between short- and long-term PM2.5 exposure and mortality or serious morbidity
effects. The evidence available in the 2019 ISA reaffirmed, and in some cases strengthened, the
conclusions from the 2009 ISA regarding the health effects of PM2.5 exposures (U.S. EPA,
2009). Much of the evidence came from epidemiologic studies conducted in North America,
Europe, or Asia that demonstrated generally positive, and often statistically significant, PM2.5
health effect associations. Such studies reported associations between estimated PM2.5 exposures
and non-accidental, cardiovascular, or respiratory mortality; cardiovascular or respiratory
hospitalizations or emergency department visits; and other mortality/morbidity outcomes (e.g.,
lung cancer mortality or incidence, asthma development). Experimental evidence, as well as
evidence from panel studies, strengthened support for potential biological pathways through
which PM2.5 exposures could lead to health effects reported in many population-based
epidemiologic studies, including support for pathways that could lead to cardiovascular,
respiratory, nervous system, and cancer-related effects (U.S. EPA, 2019). Based on this
evidence, the 2019 ISA concludes there to be a causal relationship between long- and short-term
PM2.5 exposure and mortality and cardiovascular effects, as well as likely to be causal
relationships between long- and short-term PM2.5 exposures and respiratory effects, and between
long-term PM2.5 exposures and cancer and nervous system effects (U.S. EPA, 2019, section 1.7).

Epidemiologic studies reported PM2.5 health effect associations with mortality and/or
morbidity across multiple U.S. cities and in diverse populations, including in studies examining
populations and lifestages that may be at comparatively higher risk of experiencing a PM2.5-
related health effect (e.g., older adults, children). The 2019 ISA cited extensive evidence
indicating that "both the general population as well as specific populations and lifestages are at

3-2

-------
risk for PIVh.s-related health effects" (U.S. EPA, 2019, p. 12-1). In support of the causal and
likely to be causal determinations, the 2019 ISA cites substantial evidence for:

• PM-related mortality and cardiovascular effects in older adults (U.S. EPA, 2019, sections
11.1, 11.2, 6.1, and 6.2);

• PM-related cardiovascular effects in people with pre-existing cardiovascular disease (U.S.
EPA, 2019, section 6.1);

• PM-related respiratory effects in people with pre-existing respiratory disease, particularly
asthma (U.S. EPA, 2019, section 5.1);

• PM-related impairments in lung function growth and asthma development in children (U.S.
EPA, 2019, sections 5.1, 5.2, and 12.5.1.1).

The 2019 ISA also noted that stratified analyses (i.e., analyses that allow for comparison of PM-
related health effects in subgroups to health effects for full populations) provided strong
evidence for racial and ethnic differences in PM2.5 exposures and PM2.5-related health risk. Such
analyses indicated that certain racial and ethnic groups such as Hispanic and non-Hispanic Black
populations have higher PM2.5 exposures than non-Hispanic White populations, thus contributing
to risk of adverse health effects in minority populations (U.S. EPA, 2019, section 12.5.4).
Stratified analyses focused on other groups also suggested that populations with pre-existing
cardiovascular or respiratory disease, populations that are overweight or obese, populations that
have particular genetic variants, and populations that are of low socioeconomic status could be at
increased risk for PM2.5-related adverse health effects (U.S. EPA, 2019, chapter 12).

The risk information available in the 2020 review included risk estimates for air quality
conditions just meeting the existing primary PM2.5 standards, and also for air quality conditions
just meeting potential alternative standards. The general approach to estimating PM2.5-associated
health risks combined concentration-response functions from epidemiologic studies with model-
based PM2.5 air quality surfaces, baseline health incidence data, and population demographics for
47 urban areas (U.S. EPA, 2020, section 3.3, Figure 3-10, Appendix C). The risk assessment
estimated that the existing primary PM2.5 standards could allow a substantial number of PM2.5-
associated deaths in the U.S. Uncertainty in risk estimates (e.g., in the size of risk estimates) can
result from a number of factors, including assumptions about the shape of the concentration-
response relationship with mortality at low ambient PM concentrations, the potential for
confounding and/or exposure measurement error, and the methods used to adjust PM2.5 air
quality. In light of the limitations and uncertainties, these risk estimates were given little weight
by the Administrator in his decision on the standards (85 FR 82717, December 18. 2020).

Consistent with the general approach routinely employed in NAAQS reviews, the initial
consideration in the 2020 review of the primary PM2.5 standards was with regard to the adequacy

3-3

-------
of protection provided by the then-existing standards. Key aspects of that consideration are
summarized in section 3.1.1 below.

3.1.1 Considerations Regarding the Adequacy of the Existing Standards in the 2020
Review

With the 2020 final decision, the EPA retained the primary 24-hour PM2.5 standard, with
its level of 35 |ig/m3, and the primary annual PM2.5 standard, with its level of 12.0 |ig/m3. The
Administrator's conclusions regarding the adequacy of the primary PM2.5 standards at the time of
the 2020 review was based on consideration of the evidence, analyses and conclusions contained
in the 2019 ISA; the quantitative risk assessment in the 2020 PA; advice from the CAS AC; and
public comments. Key considerations informing the Administrator's decision to retain the
standards that were promulgated in the 2012 review are summarized below.

As an initial matter, the Administrator considered the range of scientific evidence
evaluating these effects, including studies of at-risk populations, to inform his review of the
primary PM2.5 standards, placing the greatest weight on evidence of effects for which the 2019
ISA determined there to be a causal or likely to be causal relationship with long- and short-term
PM2.5 exposures (85 FR 82714-82715, December 18, 2020).

With regard to indicator, the Administrator recognized that, consistent with the evidence
available in prior reviews, the scientific evidence in the 2020 review continued to provide strong
support for health effects following short- and long-term PM2.5 exposures. He noted the 2020 PA
conclusions that the information continued to support the PM2.5 mass-based indicator and
remained too limited to support a distinct standard for any specific PM2.5 component or group of
components, and too limited to support a distinct standard for the ultrafine fraction. Thus, the
Administrator concluded that it was appropriate to retain PM2.5 as the indicator for the primary
standards for fine particles (85 FR 82715, December 18, 2020).

With respect to averaging time and form, the Administrator noted that the scientific
evidence continued to provide strong support for health effects associations with both long-term
(e.g., annual or multi-year) and short-term (e.g., mostly 24-hour) exposures to PM2.5, consistent
with the conclusions in the 2020 PA. In the 2019 ISA, epidemiologic and controlled human
exposure studies examined a variety ofPlVh.s exposure durations. Epidemiologic studies
continued to provide strong support for health effects associated with short-term PM2.5 exposures
based on 24-hour PM2.5 averaging periods, and the EPA noted that associations with sub-daily
estimates are less consistent and, in some cases, smaller in magnitude (U.S. EPA, 2019, section
1.5.2.1; U.S. EPA, 2020, section 3.5.2.2). In addition, controlled human exposure and panel-
based studies of sub-daily exposures typically examined subclinical effects, rather than the more
serious population-level effects that have been reported to be associated with 24-hour exposures

3-4

-------
(e.g., mortality, hospitalizations). Taken together, the 2019 ISA concludes that epidemiologic
studies did not indicate that sub-daily averaging periods were more closely associated with
health effects than the 24-hour average exposure metric (U.S. EPA, 2019, section 1.5.2.1).
Additionally, while controlled human exposure studies provided consistent evidence for
cardiovascular effects following PM2.5 exposures for less than 24 hours (i.e., < 30 minutes to 5
hours), exposure concentrations in the studies were well-above the ambient concentrations
typically measured in locations meeting the existing standards (U.S. EPA, 2020, section 3.2.3.1).
Thus, these studies also did not suggest the need for additional protection against sub-daily PM2.5
exposures (U.S. EPA, 2020, section 3.5.2.2). Therefore, the Administrator judged that the 24-
hour averaging time remained appropriate (85 FR 82715, December 18, 2020).

With regard to the form of the 24-hour standard (98th percentile, averaged over three
years), the Administrator noted that epidemiologic studies continued to provide strong support
for health effect associations with short-term (e.g., mostly 24-hour) PM2.5 exposures (U.S. EPA,
2020, section 3.5.2.3) and that controlled human exposure studies provided evidence for health
effects following single short-term "peak" PM2.5 exposures. Thus, the evidence supported
retaining a standard focused on providing supplemental protection against short-term peak
exposures and supported a 98th percentile form for a 24-hour standard. The Administrator further
noted that this form also provided an appropriate balance between limiting the occurrence of
peak 24-hour PM2.5 concentrations and identifying a stable target for risk management programs
(U.S. EPA, 2020, section 3.5.2.3). As such, the Administrator concluded that the available
information supported retaining the form and averaging time of the current 24-hour standard
(98th percentile, averaged over three years) and annual standard (annual average, averaged over
three years) (85 FR 82715, December 18, 2020).

With regard to the level of the standards, in reaching his final decision, the Administrator
considered the large body of evidence presented and assessed in the 2019 ISA (U.S. EPA, 2019),
the policy-relevant and risk-based conclusions and rationales as presented in the 2020 PA (U.S.
EPA, 2020), advice from the CASAC, and public comments. In particular, in considering the
2019 ISA and 2020 PA, he considered key epidemiologic studies that evaluated associations
between PM2.5 air quality distributions and mortality and morbidity, including key accountability
studies; the availability of experimental studies to support biological plausibility; controlled
human exposure studies examining effects following short-term PM2.5 exposures; air quality
analyses; and the important uncertainties and limitations associated with the information (85 FR
82715, December 18, 2020).

As an initial matter, the Administrator considered the protection afforded by both the
annual and 24-hour standards together against long- and short-term PM2.5 exposures and health
effects. The Administrator recognized that the annual standard was most effective in controlling

3-5

-------
"typical" PM2.5 concentrations near the middle of the air quality distribution (i.e., around the
mean of the distribution), but also provided some control over short-term peak PM2.5
concentrations. On the other hand, the 24-hour standard, with its 98th percentile form, was most
effective at limiting peak 24-hour PM2.5 concentrations, but in doing so also had an effect on
annual average PM2.5 concentrations. Thus, while either standard could be viewed as providing
some measure of protection against both average exposures and peak exposures, the 24-hour and
annual standards were not expected to be equally effective at limiting both types of exposures.
Thus, consistent with previous reviews, the Administrator's consideration of the public health
protection provided by the existing primary PM2.5 standards was based on his consideration of
the combination of the annual and 24-hour standards. Specifically, he recognized that the annual
standard was more likely to appropriately limit the "typical" daily and annual exposures that are
most strongly associated with the health effects observed in epidemiologic studies. The
Administrator concluded that an annual standard (as the arithmetic mean, averaged over three
years) remained appropriate for targeting protection against the annual and daily PM2.5 exposures
around the middle portion of the PM2.5 air quality distribution. Further, recognizing that the 24-
hour standard (with its 98th percentile form) was more directly tied to short-term peak PM2.5
concentrations, and more likely to appropriately limit exposures to such concentrations, the
Administrator concluded that the current 24-hour standard (with its 98th percentile form,
averaged over three years) remained appropriate to provide a balance between limiting the
occurrence of peak 24-hour PM2.5 concentrations and identifying a stable target for risk
management programs. However, the Administrator recognized that changes in PM2.5 air quality
to meet an annual standard would likely result not only in lower short- and long-term PM2.5
concentrations near the middle of the air quality distribution, but also in fewer and lower short-
term peak PM2.5 concentrations. The Administrator further recognized that changes in air quality
to meet a 24-hour standard, with a 98th percentile form, would result not only in fewer and lower
peak 24-hour PM2.5 concentrations, but also in lower annual average PM2.5 concentrations (85
FR 82715-82716, December 18, 2020).

Thus, in considering the adequacy of the 24-hour standard, the Administrator noted the
importance of considering whether additional protection was needed against short-term
exposures to peak PM2.5 concentrations. In examining the scientific evidence, he noted the
limited utility of the animal toxicologic studies in directly informing conclusions on the
appropriate level of the standard given the uncertainty in extrapolating from effects in animals to
those in human populations. The Administrator noted that controlled human exposure studies
provided evidence for health effects following single, short-term PM2.5 exposures that
corresponded best to exposures that might be experienced in the upper end of the PM2.5 air
quality distribution in the U.S. (i.e., "peak" concentrations). However, most of these studies

3-6

-------
examined exposure concentrations considerably higher than are typically measured in areas
meeting the standards (U.S. EPA, 2020, section 3.2.3.1). In particular, controlled human
exposure studies often reported statistically significant effects on one or more indicators of
cardiovascular function following 2-hour exposures to PM2.5 concentrations at and above 120
[j,g/m3 (at and above 149 [j,g/m3 for vascular impairment, the effect shown to be most consistent
across studies). To provide insight into what these studies may indicate regarding the primary
PM2.5 standards, the 2020 PA (U.S. EPA, 2020, p. 3-49) noted that 2-hour ambient
concentrations of PM2.5 at monitoring sites meeting the current standards almost never exceeded
32 [j,g/m3. In fact, even the extreme upper end of the distribution of 2-hour PM2.5 concentrations
at sites meeting the primary PM2.5 standards remained well-below the PM2.5 exposure
concentrations consistently shown in controlled human exposure studies to elicit effects (i.e.,
99.9th percentile of 2-hour concentrations at these sites is 68 [j,g/m3 during the warm season).
Thus, the available experimental evidence did not indicate the need for additional protection
against exposures to peak PM2.5 concentrations, beyond the protection provided by the
combination of the 24-hour and the annual standards (U.S. EPA, 2020, section 3.2.3.1; 85 FR
82716, December 18, 2020).

With respect to the epidemiologic evidence, the Administrator noted that the studies did
not indicate that associations in those studies were strongly influenced by exposures to peak
concentrations in the air quality distribution and thus did not indicate the need for additional
protection against short-term exposures to peak PM2.5 concentrations (U.S. EPA, 2020, section
3.5.1). The Administrator noted that this was consistent with CAS AC consensus support for
retaining the current 24-hour standard. Thus, the Administrator concluded that the 24-hour
standard with its level of 35 |ig/m3 was adequate to provide supplemental protection (i.e., beyond
that provided by the annual standard alone) against short-term exposures to peak PM2.5
concentrations (85 FR 82716, December 18, 2020).

With regard to the level of the annual standard, the Administrator recognized that the
annual standard, with its form based on the arithmetic mean concentration, was most
appropriately meant to limit the "typical" daily and annual exposures that were most strongly
associated with the health effects observed in epidemiologic studies. However, the Administrator
also noted that while epidemiologic studies examined associations between distributions of PM2.5
air quality and health outcomes, they did not identify particular PM2.5 exposures that cause
effects and thus, they could not alone identify a specific level at which the standard should be
set, as such a determination necessarily required the Administrator's judgment. Thus, consistent
with the approaches in previous NAAQS reviews, the Administrator recognized that any
approach that used epidemiologic information in reaching decisions on what standards are
appropriate necessarily required judgments about how to translate the information from the

3-7

-------
epidemiologic studies into a basis for appropriate standards. This approach included
consideration of the uncertainties in the reported associations between daily or annual average
PM2.5 exposures and mortality or morbidity in the epidemiologic studies. Such an approach is
consistent with setting standards that are neither more nor less stringent than necessary,
recognizing that a zero-risk standard is not required by the Clean Air Act (CAA) (85 FR 82716,
December 18, 2020).

The Administrator emphasized uncertainties and limitations that were present in
epidemiologic studies in previous reviews and persisted in the 2020 review. These uncertainties
included exposure measurement error, potential confounding by copollutants, increasing
uncertainty of associations at lower PM2.5 concentrations, and heterogeneity of effects across
different cities or regions (85 FR 82716, December 18, 2020). The Administrator also noted the
advice given by the CASAC on this matter. The CASAC members who supported retaining the
annual standard expressed their concerns with the epidemiologic studies, asserting that these
studies did not provide a sufficient basis for revising the existing standards. They also identified
several key concerns regarding the associations reported in epidemiologic studies and concluded
that "while the data on associations should certainly be carefully considered, this data should not
be interpreted more strongly than warranted based on its methodological limitations" (Cox, 2019,
p. 8 consensus responses).

Taking into consideration the views expressed by the CASAC members who supported
retaining the annual standard, the Administrator recognized that epidemiologic studies examined
associations between distributions of PM2.5 air quality and health outcomes, and they did not
identify particular PM2.5 exposures that cause effects (U.S. EPA, 2020, section 3.1.2). While the
Administrator remained concerned about placing too much weight on epidemiologic studies to
inform conclusions on the adequacy of the primary standards, he noted the approach to
considering such studies in the 2012 review. In the 2012 review, it was noted that the evidence of
an association in any epidemiologic study was "strongest at and around the long-term average
where the data in the study are most concentrated" (78 FR 3140, January 15, 2013). In
considering the characterization of epidemiologic studies, the Administrator viewed that when
assessing the mean concentrations of the key short-term and long-term epidemiologic studies in
the U.S. that use ground-based monitoring (i.e., those studies where the mean is most directly
comparable to the current annual standard), the majority of studies had mean concentrations at or
above the level of the existing annual standard, with the mean of the study-reported means or
medians equal to 13.5 |ig/m3, a concentration level above the existing level of the primary annual
standard of 12 |ig/m3. The Administrator further noted his caution in directly comparing the
reported study mean values to the standard level given that study-reported mean concentrations,
by design, are generally lower than the design value of the highest monitor in an area, which

3-8

-------
determines compliance. In the 2020 PA, analyses of recent air quality in U.S. CBS As indicated
that maximum annual PM2.5 design values for a given three-year period were often 10% to 20%
higher than average monitored concentrations (i.e., averaged across multiple monitors in the
same CBSA) (U.S. EPA, 2020, Appendix B, section B.7). He further noted his concern in
placing too much weight on any one epidemiologic study but instead judged that it was more
appropriate to focus on the body of studies together and therefore noted the calculation of the
mean of study-reported means (or medians). Thus, while the Administrator was cautious in
placing too much weight on the epidemiologic evidence alone, he noted that: (1) the reported
mean concentration in the majority of the key U.S. epidemiologic studies using ground-based
monitoring data were above the level of the existing annual standard; (2) the mean of the
reported study means (or medians) (i.e., 13.5 |ig/m3) was above the level of the current standard;2
(3) air quality analyses showed the study means to be lower than their corresponding design
values by 10-20%; and (4) these analyses must be considered in light of uncertainties inherent in
the epidemiologic evidence. When taken together, the Administrator judged that, even if it were
appropriate to place more weight on the epidemiologic evidence, this information did not call
into question the adequacy of the current standards (85 FR 82716-82717, December 18, 2020).

In addition to the evidence, the Administrator also considered the potential implications
of the risk assessment. He noted that all risk assessments have limitations and that he remained
concerned about the uncertainties in the underlying epidemiologic data used in the risk
assessment. The Administrator also noted that in previous reviews, these uncertainties and
limitations have often resulted in less weight being placed on quantitative estimates of risk than
on the underlying scientific evidence itself (e.g., 78 FR 3086, 3098-99, January 15, 2013). These
uncertainties and limitations included uncertainty in the shapes of concentration-response
functions, particularly at low concentrations; uncertainties in the methods used to adjust air
quality; and uncertainty in estimating risks for populations, locations and air quality distributions
different from those examined in the underlying epidemiologic study (U.S. EPA, 2020, section
3.3.2.4). Additionally, the Administrator noted similar concern expressed by some members of
the CASAC who support retaining the existing standards; they highlighted similar uncertainties
and limitations in the risk assessment (Cox, 2019). In light of all of this, the Administrator
judged it appropriate to place little weight on quantitative estimates of PIVh.s-associated mortality
risk in reaching conclusions about the level of the primary PM2.5 standards (85 FR 82717,
December 18, 2020).

2 The median of the study-reported mean (or median) PM2 5 concentrations is 13.3 |ig/ml which was also above the
level of the existing standard.

3-9

-------
The Administrator additionally considered an emerging body of evidence from
accountability studies that examined past reductions in ambient PM2.5 and the degree to which
those reductions resulted in public health improvements. While the Administrator agreed with
public commenters that well-designed and conducted accountability studies can be informative,
he viewed that interpreting such studies in the context of the primary PM2.5 standards was
complicated by the fact that some of the available studies had not evaluated PM2.5 specifically
(e.g., as opposed to PM10 or total suspended particulates), did not show changes in PM2.5 air
quality, or had not been able to disentangle health impacts of the interventions from background
trends in health (U.S. EPA, 2020, section 3.5.1). He further recognized that the small number of
available studies that did report public health improvements following past declines in ambient
PM2.5 had not examined air quality meeting the existing standards (U.S. EPA, 2020, Table 3-3).
This included U.S. studies that reported increased life expectancy, decreased mortality, and
decreased respiratory effects following past declines in ambient PM2.5 concentrations. Such
studies examined "starting" annual average PM2.5 concentrations (i.e., prior to the reductions
being evaluated) ranging from about 13.2 to > 20 |ig/m3 (i.e., U.S. EPA, 2020, Table 3-3). Given
the lack of available accountability studies reporting public health improvements attributable to
reductions in ambient PM2.5 in locations meeting the existing standards, together with his broader
concerns regarding the lack of experimental studies examining PM2.5 exposures typical of areas
meeting the existing standards, the Administrator judged that there was considerable uncertainty
in the potential for increased public health protection from further reductions in ambient PM2.5
concentrations beyond those achieved under the existing primary PM2.5 standards (85 FR 82717,
December 18, 2020).

When the above considerations were taken together, the Administrator concluded that the
scientific evidence assessed in the 2019 ISA, together with the analyses in the 2020 PA based on
that evidence and consideration of CASAC advice and public comments, did not call into
question the adequacy of the public health protection provided by the existing annual and 24-
hour PM2.5 standards. In particular, the Administrator judged that there was considerable
uncertainty in the potential for additional public health improvements from reducing ambient
PM2.5 concentrations below the concentrations achieved under the existing primary standards and
that, therefore, standards more stringent than the existing standards (e.g., with lower levels) were
not supported. That is, he judged that such standards would be more than requisite to protect the
public health with an adequate margin of safety. This judgment reflected the Administrator's
consideration of the uncertainties in the potential implications of the lower end of the air quality
distributions from the epidemiologic studies due in part to the lack of supporting evidence from
experimental studies and retrospective accountability studies conducted at PM2.5 concentrations
meeting the existing standards (85 FR 82717, December 18, 2020).

3-10

-------
In reaching this conclusion, the Administrator judged that the existing standards provided
an adequate margin of safety. With respect to the annual standard, the level of 12 |ig/m3 was
below the lowest "starting" concentration (i.e., 13.2 |ig/m3) in the available accountability
studies that showed public health improvements attributable to reductions in ambient PM2.5. In
addition, while the Administrator placed less weight on the epidemiologic evidence for selecting
a standard, he noted that the level of the annual standard was below the reported mean (and
median) concentrations in the majority of the key U.S. epidemiologic studies using ground-based
monitoring data (noting that these means tend to be 10-20% lower than their corresponding area
design values which is the more relevant metric when considering the level of the standard) and
below the mean of the reported means (or medians) of these studies (i.e., 13.5 |ig/m3). In
addition, the Administrator recognized that concentrations in areas meeting the existing 24-hour
and annual standards remained well-below the PM2.5 exposure concentrations consistently shown
to elicit effects in human exposure studies (85 FR 82717-82718, December 18, 2020).

In addition, based on the Administrator's review of the science, including controlled
human exposure studies examining effects following short-term PM2.5 exposures, the
epidemiologic studies, and accountability studies conducted at levels just above the existing
annual standard, he judged that the degree of public health protection provided by the existing
annual standard is not greater than warranted. This judgment, together with the fact that no
CASAC member expressed support for a less stringent standard, led the Administrator to
conclude that standards less stringent than the existing standards (e.g., with higher levels) were
also not supported (85 FR 82718, December 18, 2020).

In reaching his final decision, the Administrator concluded that the scientific evidence
and technical information continued to support the existing annual and 24-hour PM2.5 standards.
This conclusion reflected the Administrator's view that there were important limitations and
uncertainties that remained in the evidence. The Administrator concluded that these limitations
contributed to considerable uncertainty regarding the potential public health implications of
revising the existing primary PM2.5 standards. Given this uncertainty, and noting the advice from
some CASAC members, he concluded that the primary PM2.5 standards, including the indicators
(PM2.5), averaging times (annual and 24-hour), forms (arithmetic mean and 98th percentile,
averaged over three years) and levels (12.0 ng/m3, 35 |j,g/m3), when taken together, remained
requisite to protect the public health. Therefore, in the 2020 review, the Administrator reached
the conclusion that the primary 24-hour and annual PM2.5 standards, together, were requisite to
protect public health from fine particles with an adequate margin of safety, including the health
of at-risk populations, and retained the standards, without revision (85 FR 82718, December 18,
2020).

3-11

-------
3.2 GENERAL APPROACH AND KEY ISSUES IN THIS
RECONSIDERATION OF THE 2020 FINAL DECISION

This reconsideration of the 2020 final decision on the primary PM2.5 standards is most
fundamentally based on the Agency's assessment of the scientific evidence and associated
quantitative analyses to inform the Administrator's judgments regarding primary standards that
are requisite to protect public health with an adequate margin of safety. This PA is intended to
help bridge the gap between the scientific evidence and information assessed in the 2019 ISA
and the ISA Supplement and the judgments required of the Administrator in determining whether
it is appropriate to retain or revise the primary PM2.5 NAAQS. The approach for this
reconsideration builds on the substantial assessments and evaluations performed over the course
of the prior reviews (U.S. EPA, 2011; U.S. EPA, 2020), taking into account the more recent
scientific information and air quality data now available to inform our understanding of the key
policy issues relevant in this reconsideration.

The evaluations in this PA of the scientific assessments in the 2019 ISA and the ISA
Supplement,3 augmented by the quantitative risk analyses, are intended to inform the
Administrator's public health policy judgments and conclusions, including his decisions as to
whether to retain or revise the primary PM2.5 standards. The PA evaluations consider the
potential implications of various aspects of the scientific evidence, the risk-based information,
and the associated uncertainties and limitations. In so doing, the approach for this PA involves
evaluating the scientific and technical information to address a series of key policy-relevant
questions using both evidence- and risk-based considerations. Together, consideration of the full
set of evidence and information available in this reconsideration will inform the answer to the
following initial overarching question for the reconsideration:

• Does the scientific evidence, air quality and quantitative risk information support or
call into question the adequacy of the public health protection afforded by the
current primary annual and 24-hour PM2.5 standards?

In reflecting on this question, we will consider the body of scientific evidence, assessed
in the 2019 ISA and the ISA Supplement and used as a basis for developing or interpreting risk
analyses, including whether it supports or calls into question the scientific conclusions reached in

3 As described in detail in section 1.4.2, the ISA Supplement focuses on a thorough evaluation of some studies that
became available after the literature cutoff date of the 2019 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 ISA (U.S. EPA, 2022). The selection of the health effects to evaluate within the ISA Supplement were based
on the causality determinations reported in the 2019 ISA and the subsequent use of scientific evidence in the 2020
PA. Specifically, for PM2 5-related health effects, the focus within the ISA Supplement is on mortality and
cardiovascular effects. The ISA Supplement does not include an evaluation of studies for other PM2 5-related
health effects (U.S. EPA, 2022).

3-12

-------
the 2020 review regarding health effects related to exposure to PM2.5 in ambient air. Information
available in this reconsideration that may be informative to public health judgments regarding
significance or adversity of key effects will also be considered. Additionally, the available risk
information, whether newly developed for this reconsideration or predominantly developed in
the past and interpreted in light of recent information, will be considered, including consideration
of the extent to which it may continue to support judgments made in the 2020 review. Further, in
considering this question with regard to the primary PM2.5 standards, as in all NAAQS reviews,
we give particular attention to exposures and health risks to at-risk populations (including at-risk
lifestages).4

If the information available in this reconsideration suggests that revision of the current
primary standards would be appropriate to consider, the PA will also evaluate how the standards
might be revised based on the scientific information, air quality assessments, and risk
information, and also considering what the information indicates as to the health protection
expected to be afforded by the current or potential alternative standards. Such an evaluation may
consider the effect of revising one or more elements of the standard (indicator, averaging time,
level, and form), with the impact evaluated being on the resulting potential standard and all of its
elements collectively. Based on such evaluations, the PA would then identify potential
alternative standards (specified in terms of indicator, averaging time, level, and form) intended to
reflect a range of alternative policy judgments as to the degree of protection that is requisite to
protect public health with an adequate margin of safety, and options for standards to achieve it.
The initial overarching policy-relevant question that frames such an evaluation of what revision
of the standard might be appropriate to consider is:

• What range of potential alternative standards could be supported by the available

scientific evidence, air quality and risk information?

The approach to reaching conclusions on the current primary PM2.5 standards and, as
appropriate, on potential alternative standards is summarized in general terms in Figure 3-1.

4 As used here and similarly throughout this document, the term population refers to persons having a quality or
characteristic in common, such as a specific pre-existing illness or a specific age or lifestage. Identifying at-risk
populations involves consideration of susceptibility and vulnerability. Susceptibility refers to innate (e.g., genetic
or developmental aspects) or acquired (e.g., disease or smoking status) sensitivity that increases the risk of health
effects occurring with exposure to PM2 5. Vulnerability refers to an increased risk of PM2 5-related health effects
due to factors such as those related to socioeconomic status, reduced access to health care or exposure.

3-13

-------
Adequacy of Existing Primary PM2 5 Standards

Evidence-Based Considerations

^Degree to which available evidence strengthens support for, or calls
into question, health effects attributable to fine particle exposures

> Evidence for at-risk populations

Degree to which uncertainties have been addressed or new
uncertainties identified

Support for adverse effects at ambient concentrations meeting current
PM2 5 standards

Risk-Based Considerations

Nature, magnitude, and importance of
estimated risks associated with current
primary PM2 5 standards
> Uncertainties in the risk estimates

Appropriate to
consider retaining
current standards

Further evaluate the scientific evidence and risk assessmentto
inform Identification of potential alternatives

Figure 3-1. Overview of general approach for the reconsideration of the 2020 final decision
on the primary P1VI2.5 standards.

3-14

-------
The Agency's approach in reconsidering the primary standards is consistent with
requirements of the provisions of the CAA related to the review of the NAAQS and with how the
EPA and the courts have historically interpreted the CAA. As discussed in section 1.1 above,
these provisions require the Administrator to establish primary standards that, in the
Administrator's judgment, are requisite (i.e., neither more nor less stringent than necessary) to
protect public health with an adequate margin of safety. Consistent with the Agency's approach
across all NAAQS reviews, the approach of this PA to informing these judgments is based on a
recognition that the available health effects evidence generally reflects continuums that include
ambient air exposures for which scientists generally agree that health effects are likely to occur
through lower levels at which the likelihood and magnitude of response become increasingly
uncertain. The CAA does not require the Administrator to establish a primary standard at a zero-
risk level or at background concentration levels, but rather at a level that reduces risk sufficiently
so as to protect public health, including the health of sensitive groups,5 with an adequate margin
of safety.

The decisions on the adequacy of the current primary PM2.5 standards and on any
alternative standards considered in a reconsideration are largely public health policy judgments
made by the Administrator. The four basic elements of the NAAQS (i.e., indicator, averaging
time, form, and level) are generally considered collectively in evaluating the health protection
afforded by the current standards, and by any alternatives considered. The Administrator's final
decisions draw upon the scientific evidence for health effects, quantitative analyses of population
exposures and/or health risks, as available, and judgments about how to consider the
uncertainties and limitations that are inherent in the scientific evidence and quantitative analyses.

3.3 HEALTH EFFECTS EVIDENCE

In this section, we draw from the EPA's synthesis and assessment of the scientific
evidence presented in the 2019 ISA (U.S. EPA, 2019) and the ISA Supplement (U.S. EPA, 2022)
to consider the following policy-relevant question:

5 More than one population group may be identified as sensitive or at-risk in a NAAQS review. Decisions on
NAAQS reflect consideration of the degree to which protection is provided for these sensitive population groups.
To the extent that any particular population group is not among the identified sensitive groups, a decision that
provides protection for the sensitive groups would be expected to also provide protection for other population
groups.

3-15

-------
• To what extent does the currently available scientific evidence, as assessed in the
2019 ISA and the ISA Supplement, support or call into question the public health
protection afforded by the current suite of primary PM2.5 standards?

The 2019 ISA uses a weight-of-evidence framework for characterizing the strength of the
available scientific evidence for health effects attributable to PM exposures (U.S. EPA, 2015a,
Preamble, section 5). This framework provides the basis for robust, consistent, and transparent
evaluation of the scientific evidence, including its uncertainties, and for drawing conclusions on
PM-related health effects. As in previous reviews, the 2019 ISA adopts a five-level hierarchy to
classify the overall weight of evidence into one of the following categories: causal relationship;
likely to be a causal relationship; suggestive of, but not sufficient to infer, a causal relationship;
inadequate to infer a causal relationship; and not likely to be a causal relationship (U.S. EPA,
2015a, Preamble Table II). In using the weight-of-evidence approach to inform judgments about
the causal nature of relationships between PM exposure and health effects, evidence is evaluated
for major outcome categories or groups of related outcomes (e.g., respiratory effects), integrating
evidence from across disciplines, including epidemiologic, controlled human exposure, and
animal toxicological studies and evaluating the coherence of evidence across a spectrum of
related endpoints (U.S. EPA, 2015a, Preamble, section 5.c.). In this PA, we consider the full
body of health evidence, including evidence from the 2019 ISA and the ISA Supplement, placing
the greatest emphasis on the health effects for which the evidence has been judged in the 2019
ISA to demonstrate a "causal" or a "likely to be causal" relationship with PM exposures. The
2019 ISA defines these causality determinations as follows (U.S. EPA, 2019, p. p-20; U.S. EPA,
2015a):

• Causal relationship: the pollutant has been shown to result in health effects at relevant
exposures based on studies encompassing multiple lines of evidence and chance,
confounding, and other biases can be ruled out with reasonable confidence.

• Likely to be a causal relationship: there are studies in which results are not explained by
chance, confounding, or other biases, but uncertainties remain in the health effects evidence
overall. For example, the influence of co-occurring pollutants is difficult to address, or
evidence across scientific disciplines may be limited or inconsistent.

While the 2019 ISA provides the broad scientific foundation for this reconsideration, we
recognize that additional literature has become available since the literature cutoff date of the
2019 ISA that expands the body of evidence that can inform the Administrator's judgments on
the adequacy of the current primary PM2.5 standards. As such, the ISA Supplement builds on the
information in the 2019 ISA with a targeted identification and evaluation of new scientific
information (U.S. EPA, 2022, section 1.2). The ISA Supplement focuses on PM2.5 health effects
evidence where the 2019 ISA concludes a "causal relationship," because such health effects are
given the most weight in an Administrator's decisions in a NAAQS review. The ISA Supplement

3-16

-------
evaluates newly available evidence related to short- and long-term PM2.5 exposure and mortality
and cardiovascular effects given the strength of the evidence available in the 2019 ISA and past
ISAs and AQCDs, as well as the clear adversity of these endpoints. Specifically, U.S. and
Canadian epidemiologic studies for mortality and cardiovascular effects along with air quality
analyses related to concentrations evaluated in controlled human exposure studies associated
with cardiovascular effects, were considered to be of greatest utility in informing the
Administrator's conclusions on the adequacy of the current primary PM2.5 standards. While the
ISA Supplement does not include information for health effects other than mortality and
cardiovascular effects, the evidence as it was assessed in the 2019 ISA is considered in this PA in
reaching conclusions as a part of the reconsideration of the 2020 final decision.

The ISA Supplement also assessed accountability studies because these types of
epidemiologic studies were part of the body of evidence that was a focus of the 2020 review.
Accountability studies inform our understanding of the potential for public health improvements
as ambient PM2.5 concentrations have declined over time. Further, the ISA Supplement
considered 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, given that such studies were highlighted by the CASAC and identified in
public comments in the 2020 review. Since the literature cutoff date for the 2019 ISA, multiple
accountability studies and studies that employ alternative methods for confounder control have
become available for consideration in the ISA Supplement and in this reconsideration.

The ISA Supplement also considered recent health effects evidence that addresses key
scientific issues where the literature has expanded since the completion of the 2019 ISA.7 The
2019 ISA evaluated a few controlled human exposure studies that have investigated the effect of
exposure to near-ambient concentrations of PM2.5 (U.S. EPA, 2019, section 6.1.10 and 6.1.13).
The ISA Supplement adds to this limited evidence, including a recent study conducted in young
healthy individuals exposed to near-ambient PM2.5 concentrations (U.S. EPA, 2022, section
3.3.1). Given the importance of identifying the populations at increased risk of PIVh.s-related
effects, the ISA Supplement also included epidemiologic or exposure studies examining

6 As noted in the ISA Supplement (U.S. EPA, 2022, p. 1-3): "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."

7 As with the epidemiologic studies for long- and short-term PM2 5 exposure and mortality and cardiovascular

effects, epidemiologic studies of exposure or risk disparities and SARS-CoV-2 infection and/or COVID-19 death
were limited to those conducted in the U.S. and Canada.

3-17

-------
exposure or risk disparities by race/ethnicity or socioeconomic status. The ISA Supplement
assessed studies that examined the relationship between short-term PM2.5 exposures and SARS-
CoV-2 infection and/or COVID-19 death, as these studies are a new area of research and were
raised by a number of public commenters in the 2020 review. These types of studies provide
additional information related to factors that may increase risk of PIVh.s-related health effects and
provide additional evidence for consideration by the Administrator in reaching conclusions
regarding the adequacy of the current standards.

The evidence presented within the 2019 ISA, along with the targeted identification and
evaluation of new scientific information in the ISA Supplement, provides the scientific basis for
the reconsideration of the 2020 final decision on the primary PM2.5 standards. In the sections
below, we consider the nature of the health effects attributable to long- and short-term PM2.5
exposures (section 3.3.1), the public health implications and populations potentially at increased
risk for PM-related effects (section 3.3.2), and the PM2.5 concentrations at which effects have
been shown to occur (section 3.3.3).

3.3.1 Nature of Effects

In considering the available evidence for health effects attributable to PM2.5 exposures
presented in the 2019 ISA and the ISA Supplement, this section poses the following policy-
relevant questions:

• To what extent does the currently available scientific evidence strengthen, or
otherwise alter, our conclusions regarding health effects attributable to long- or
short-term fine particle exposures? Have previously identified uncertainties been
reduced? What important uncertainties remain and have new uncertainties been
identified?

In answering these questions, as noted above, we consider the full body of evidence assessed in
the 2019 ISA, along with the targeted evaluation of recent evidence in the ISA Supplement (U.S.
EPA, 2019; U.S. EPA, 2022). In so doing, we place particular emphasis on health outcomes for
which the evidence in the 2019 ISA supports either a "causal" or a "likely to be causal"
relationship. While the strongest evidence focuses on PM2.5, the 2019 ISA also assesses the
evidence for the ultrafine fraction of PM2.5 (ultrafine particles or UFP), generally considered as

3-18

-------
particulates with a diameter less than or equal to 0.1 [j,m8 (typically based on physical size,
thermal diffusivity or electrical mobility) (U.S. EPA, 2019, Preface, p. 11). Table 3-1 lists
causality determinations for all of the health effect categories and exposure durations for both
PM2.5 and UFP, which we consider within this chapter (adapted from U.S. EPA, 2019, Table 1-

4).

8 As described in more detail in the 2019 ISA, UFPs have often been defined as particles <0.1 |im. but depending on
the scientific discipline, the methods used, and the particles sizes examined the UFP-health effects relationship
varies. UFP exposures in animal toxicological and controlled human exposure studies typically use a particle
concentrator, which can result in exposures to particles <0.30 |im (U.S. EPA, 2019, section 2.4.3.1). Whereas
toxicological studies typically rely on examining UFP mass, epidemiologic studies examine multiple UFP
metrics, including particle number concentration (NC), mass concentration (MC), and surface area concentration
(SC). However, depending on the monitor used and the metric, the UFP size distribution included with each of
these ranges can vary. Currently, there is no consensus within the scientific community on the metric that best
represents exposure to UFPs. Consequently, the 2019 ISA focuses on evaluating the UFP-health effects
relationship of particles <0.3 |im for MC and SC metrics included in experimental studies, and any size range that
includes particles <0.1 |im for NC (U.S. EPA, 2019, p. P-15).

3-19

-------
Table 3-1. Key causality determinations for PM2.5 and UFP exposures.

Health Outcome

Size
Fraction

Exposure
Duration

2009 ISA

2019 ISA

Mortality

PM2.5

Long-term
Short-term

Causal

Cardiovascular
effects

PM2.5

Long-term
Short-term

Causal

UFP

Short-term

Suggestive of, but not
sufficient to infer

Respiratory
effects

PM2.5

Long-term
Short-term

Likely to be causal

UFP

Short-term

Suggestive of, but not
sufficient to infer

Cancer

PM2.5

Long-term

Suggestive of, but not
sufficient to infer

Likely to be causal

Nervous System
effects

PM2.5

Long-term

—

Likely to be causal

Short-term

Inadequate

Suggestive of, but not
sufficient to infer

UFP

Long-term

...

Suggestive of, but not
sufficient to infer

Short-term

Inadequate

Suggestive of, but not
sufficient to infer

Metabolic effects

PM2.5

Long-term

...

Suggestive of, but not
sufficient to infer

Short-term

...

Suggestive of, but not
sufficient to infer

Reproduction
and Fertility

Pregnancy and
Birth Outcomes

PM2.5

Long-,
Short-term

Suggestive of, but not
sufficient to infer

Table 3-1 lists the health outcomes for which the 2019 ISA concludes the evidence supports either a causal, a likely to
be causal, or a suggestive relationship. For other health outcomes, the 2019 ISA concludes the evidence is inadequate
to infer a causal relationship (U.S. EPA, 2019, Table 1-4).

The 2009 ISA (U.S. EPA, 2009) made causality determinations for the broad category of "Reproductive and
Developmental Effects." Causality determinations for 2009 represent this broad category and not specifically for "Male
and Female Reproduction and Fertility" and "Pregnancy and Birth Outcomes".

For reproductive and developmental effects, the 2019 ISA's causality determinations reflect the combined evidence for
both short- and long-term exposures (U.S. EPA, 2019, Chapter 9).

3-20

-------
Sections 3.3.1.1 to 3.3.1.5 summarize the evidence supporting the 2019 ISA's "causal" and
"likely to be causal" determinations for PM2.5 (italics in Table 3-1) and integrates the recent
evidence assessed in the ISA Supplement, where available. Section 3.3.1.6 briefly summarizes
the evidence supporting the 2019 ISA's "suggestive of, but not sufficient to infer" causality
determinations, as well as emerging evidence related to SARS-CoV-2 infection and COVID-19
death detailed in the ISA Supplement. Each of these sections focuses on addressing the policy-
relevant questions posed above. Section 3.3.1.7 summarizes the evidence in preceding sections
and revisits the policy-relevant questions posed above. Section 3.3.2 describes the public health
implications and at-risk populations. In section 3.3.3, we present the PM2.5 concentrations in key
studies reporting PIVh.s-related health effects, and section 3.3.4 summarizes the key uncertainties
and limitations associated with the health effects evidence.

3,3,1,1 Mortality
Long-term PM2.5 exposures

The 2009 ISA reported that the evidence was "sufficient to conclude that the relationship
between long-term PM2.5 exposures and mortality is causal" (U.S. EPA, 2009, p. 7-96). The
strongest evidence supporting this conclusion was provided by epidemiologic studies,
particularly those examining two seminal cohorts, the American Cancer Society (ACS) and the
Harvard Six Cities cohorts. Analyses of the Harvard Six Cities cohort included demonstrations
that reductions in ambient PM2.5 concentrations are associated with reduced mortality risk
(Laden et al., 2006) and with increases in life expectancy (Pope et al., 2009). Further support was
provided by other cohort studies conducted in North America and Europe that also reported
positive associations between long-term PM2.5 exposures and risk of mortality (U.S. EPA, 2009).

Cohort studies, assessed in the 2019 ISA, continue to provide consistent evidence of
positive associations between long-term PM2.5 exposures and mortality. These studies add
support for associations with total and non-accidental mortality,9 as well as with specific causes
of death, including cardiovascular disease and respiratory disease (U.S. EPA, 2019, section
11.2.2). Many of these studies have extended the follow-up periods originally evaluated in the
ACS and Harvard Six Cities cohorts and continue to observe positive associations between long-
term PM2.5 exposures and mortality (U.S. EPA, 2019, section 11.2.2.1, Figures 11-18 and 11-19).
Adding to the evaluations of the ACS and Six Cities cohorts, studies conducted in other cohorts
also demonstrate consistent, positive associations between long-term PM2.5 exposure and
mortality across various demographic groups (e.g., age, sex, occupation), spatial and temporal
extents, exposure assessment metrics, and statistical techniques (U.S. EPA, 2019, sections

9 The majority of these studies examined non-accidental mortality outcomes, though some Medicare studies lack
cause-specific death information and, therefore, examine total mortality.

3-21

-------
11.2.2.1, 11.2.5; U.S. EPA, 2022, Table 11-8). This includes some of the largest cohort studies
conducted to date, with analyses of the U.S. Medicare cohort that include nearly 61 million
enrollees (Di et al., 2017b) and studies that control for a range of individual and ecological
covariates, such as race, age, socioeconomic status, smoking status, body mass index, and annual
weather variables (e.g., temperature, humidity).

Many recent North American cohort studies evaluated in the ISA Supplement continue to
examine the relationship between long-term PM2.5 exposure and mortality and report positive
and statistically significant associations. Recent studies continue to utilize large and
demographically diverse cohorts that are generally representative of the national populations in
both the U.S. and Canada, as well as focus on occupation-based specific cohorts. These "studies
published since the 2019 ISA support and extend the evidence base that contributed to the
conclusion of a causal relationship between long-term PM2.5 exposure and mortality" (U.S.
EPA, 2022, section 3.2.2.2.1, Figure 3-19, Figure 3-20)

Furthermore, studies in the 2019 ISA and the ISA Supplement evaluating cause-specific
mortality build on previous research that found consistent, positive associations between
cardiovascular and respiratory mortality, as well as other mortality outcomes. For
cardiovascular-related mortality, the evidence assessed in the ISA Supplement is consistent with
the evidence assessed in the 2019 ISA with recent studies reporting positive associations with
long-term PM2.5 exposure. When evaluating cause-specific cardiovascular mortality, recent
studies report positive associations for a number of outcomes including ischemic heart disease
(IHD) and stroke mortality (U.S. EPA, 2022, Figure 3-23). Recent studies also provide some
initial evidence that people with pre-existing health issues (such as heart failure and diabetes) are
at an increased risk of PM2.5-related effects (U.S. EPA, 2022, section 3.2.2.4) and suggest that
these individuals have a higher risk of mortality overall, which was previously only examined in
studies that used stratified analyses rather than a cohort of people with an underlying health
condition (U.S. EPA, 2022, section 3.2.2.4). With regard to respiratory mortality, epidemiologic
studies assessed in the 2019 ISA and ISA Supplement provide continued support for associations
between long-term PM2.5 exposure and respiratory mortality (U.S. EPA, 2019, section 5.2.10;
U.S. EPA, 2022, Table 3-2)

A series of epidemiologic studies evaluated in the 2019 ISA tested the hypothesis that
past reductions in ambient PM2.5 concentrations have been associated with increased life
expectancy or a decreased mortality rate (U.S. EPA, 2019, section 11.2.2.5). In their original
study, Pope et al. (2009) used air quality data in a cross-sectional analysis from 51 metropolitan
areas across the U.S., beginning in the 1970s through the early 2000s, to demonstrate that a
10 |ig/m3 decrease in long-term PM2.5 concentration was associated with a 0.61year increase in
life expectancy. In a subsequent analysis, these authors extended the period of analysis to include

3-22

-------
2000 to 2007 (Correia et al., 2013), a time period with lower ambient PM2.5 concentrations. In
this follow-up study, a decrease in long-term PM2.5 concentration continued to be associated with
an increase in life expectancy, though the magnitude of the increase was smaller than during the
earlier time period (i.e., a 10 |ig/m3 decrease in long-term PM2.5 concentration was associated
with a 0.35 year increase in life expectancy). Additional studies conducted in the U.S. or Europe
similarly report that reductions in ambient PM2.5 are associated with improvements in longevity
(U.S. EPA, 2019, section 11.2.2.5). Since the literature cutoff date for the 2019 ISA, a few
epidemiologic studies were published that examined the relationship between long-term PM2.5
exposure and life expectancy (U.S. EPA, 2022, section 3.2.1.3) and report results that are
consistent with and expand upon the body of evidence from the 2019 ISA. For example, Bennett
et al. (2019) reported that PM2.5 concentrations above the lowest observed concentration (2.8
|ig/m3) were associated with a 0.15 year decrease in national life expectancy for women and 0.13
year decrease in national life expectancy for men (U.S. EPA, 2022, section 3.2.2.2.4, Figure 3-
25). Another study compared participants living in areas with PM2.5 concentrations >12 |ig/m3 to
participants living in areas with PM2.5 concentrations <12 |ig/m3 and reported that the number of
years of life lost due to living in areas with higher PM2.5 concentrations was 0.84 years over a 5-
year period (Ward-Caviness et al., 2020; U.S. EPA, 2022, section 3.2.2.2.4).

Additionally, a number of accountability studies, which are epidemiologic studies that
evaluate whether an environmental policy or air quality intervention has resulted in reductions in
ambient air pollution concentrations and subsequent reductions in mortality, have emerged and
were evaluated in the ISA Supplement (U.S. EPA, 2022, section 3.2.2.3). For example, Sanders
et al. (2020) examined whether policy actions (i.e., the first annual PM2.5 NAAQS
implementation rule in 2005 for the 1997 annual PM2.5 standard with a 3-year annual average of
15 (j,g/m3) reduced PM2.5 concentrations and mortality rates in Medicare beneficiaries between
2000-2013 and found that following implementation of the annual PM2.5 NAAQS, annual PM2.5
concentrations decreased by 1.59 [j,g/m3 (95% CI: 1.39, 1.80) which corresponded to a reduction
in mortality rates among individuals 65 years and older (0.93% [95% CI: 0.10%, 1.77%]) in non-
attainment counties relative to attainment counties.

The 2019 ISA evaluated a small number of studies that used alternative methods for
confounder control to further assess relationship between long-term PM2.5 exposure and
mortality (U.S. EPA, 2019, section 11.2.2.4). Multiple epidemiologic studies that implemented
alternative methods for confounder control and were published since the literature cutoff date of
the 2019 ISA were evaluated in the ISA Supplement (U.S. EPA, 2022, section 3.2.2.3). These
studies use a variety of statistical methods including generalized propensity score (GPS), inverse
probability weighting (IPW), and difference-in-difference (DID) to reduce uncertainties related
to confounding bias in the association between long-term PM2.5 exposure and mortality. Studies

3-23

-------
that employed these alternative methods for confounder control reported consistent positive
associations that further inform the relationship between long-term PM2.5 exposure and total
mortality (U.S. EPA, 2022, section 3.2.2.3). These studies provide further support of associations
seen in cohort studies and referenced just above.

The 2019 ISA and ISA Supplement also evaluate the degree to which recent studies that
examine the relationship between long-term PM2.5 exposure and mortality have addressed key
policy-relevant issues and/or previously identified data gaps in the scientific evidence, including
methods to estimate exposure, methods to control for confounding, like copollutant confounding,
and the shape of the C-R curve. For example, with respect to exposure assessment, based on its
assessment of the evidence, the 2019 ISA concludes that positive associations between long-term
PM2.5 exposures and mortality are robust across recent analyses using various approaches to
estimate PM2.5 exposures (e.g., based on monitors, modeling, satellites, or hybrid methods that
combine information from multiple sources) (U.S. EPA, 2019, section 11.2.5.1). This includes a
study by Hart et al. (2015) reporting that correction for bias due to exposure measurement error
increases the magnitude of the hazard ratios (confidence intervals widen but the association
remains statistically significant), suggesting that failure to correct for exposure measurement
error could result in attenuation or underestimation of risk estimates.

The 2019 ISA additionally concludes that positive associations between long-term PM2.5
exposures and mortality are robust across statistical models that use different approaches to
control for confounders or different sets of confounders (U.S. EPA, 2019, sections 11.2.3 and
11.2.5), across diverse geographic regions and populations, and across a range of temporal
periods including the periods of declining PM concentrations (U.S. EPA, 2019, sections 11.2.2.5
and 11.2.5.3). Additional evidence further demonstrates that associations with mortality remain
robust in copollutants analyses (U.S. EPA, 2019, section 11.2.3), and that associations persist in
analyses restricted to long-term exposures (annual average PM2.5 concentrations) below 12
Hg/m3 (Di et al., 2017b) or 10 |j,g/m3 (Shi et al., 2016) (i.e., indicating that risks are not
disproportionately driven by the upper portions of the air quality distribution). Recent studies
further assess potential copollutant confounding as reflected in the studies evaluated in the ISA
Supplement that indicate while there is some evidence of potential confounding of the PM2.5-
mortality association by copollutants in some of the studies (i.e., those studies of the Mortality
Air Pollution Associations in Low Exposure Environments (MAPLE) cohort). This result is
inconsistent with other recent studies evaluated in the 2019 ISA that were conducted in the U.S.
and Canada that found associations in both single and copollutant models (U.S. EPA, 2019; U.S.
EPA, 2022, section 3.2.2.4 and 3.1.2.2.8). Additionally, a few studies use statistical techniques to
reduce uncertainties related to potential unmeasured confounders in order to further inform
conclusions on causality for long-term PM2.5 exposure and mortality. For example, studies by

3-24

-------
Greven et al. (2011), Pun et al. (2017), and Eum et al. (2018) decompose ambient PM2.5 into
"spatial" and "spatiotemporal" components in order to evaluate the potential for bias due to
unmeasured confounding. Eum et al. (2018) and Wu et al. (2020) also attempted to address long-
term trends and meteorological variables as potential confounders and found that not adjusting
for temporal trends could overestimate the association, while effect estimates in analyses that
excluded meteorological variables remained unchanged compared to main analyses. The results
of these analyses suggest the presence of unmeasured confounding, though they do not indicate
the direction or magnitude of the bias that could result.10

An additional important consideration in characterizing the public health impacts
associated with PM2.5 exposure is whether concentration-response relationships are linear across
the range of concentrations or if nonlinear relationships exist along any part of this range. Studies
evaluated in the 2019 ISA and the ISA Supplement examine this issue, and continue to provide
evidence of linear, no-threshold relationships between long-term PM2.5 exposures and all-cause
and cause-specific mortality (U.S. EPA, 2019, section 11.2.4; (U.S. EPA, 2022, section 3.2.2.2.7,
Table 3-6). Across the studies evaluated in the 2019 ISA and ISA Supplement, a variety of
statistical methods have been used to assess whether there is evidence of deviations in linearity
(U.S. EPA, 2019, Table 11-7; U.S. EPA, 2022, section 2.2.3.2). Studies have also conducted cut-
point analyses that focus on examining risk at specific ambient PM2.5 concentrations. Generally,
the evidence remains consistent in supporting a no-threshold relationship, and in supporting a
linear relationship for PM2.5 concentrations > 8 [j,g/m3. However, uncertainties remain about the
shape of the C-R curve at PM2.5 concentrations < 8 (J,g/m3, with some recent studies providing
evidence for either a sublinear, linear, or supralinear relationship at these lower concentrations
(U.S. EPA, 2019, section 11.2.4; U.S. EPA, 2022, section 2.2.3.2). There was also some limited
evidence indicating that the slope of the C-R function may be steeper (supralinear) at lower
concentrations for cardiovascular mortality (U.S. EPA, 2022, section 3.1.1.2.6).

The biological plausibility of PIVh.s-attributable mortality is supported by the coherence
of effects across scientific disciplines (i.e., animal toxicological, controlled human exposure
studies, and epidemiologic) when evaluating respiratory and cardiovascular morbidity effects,
which are some of the largest contributors to total (nonaccidental) mortality. The 2019 ISA
outlines the available evidence for biologically plausible pathways by which inhalation exposure
to PM2.5 could progress from initial events (e.g., pulmonary inflammation, autonomic nervous

10 In public comments on the 2019 draft PA, the authors of the Pun et al. (2017) study further note that "the presence
of unmeasured confounding.. .was expected given that we did not control for several potential confounders that
may impact PM2 5-mortality associations, such as smoking, socio-economic status (SES), gaseous pollutants,
PM2 5 components, and long-term time trends in PM2 5" and that "spatial confounding may bias mortality risks
both towards and away from the null" (Docket ID EPA-HQ-OAR-2015-0072-0065; accessible in
https://www.regulations.gov/)

3-25

-------
system activation) to endpoints relevant to population outcomes, particularly those related to
cardiovascular diseases such as coronary heart disease (CHD), stroke and atherosclerosis (U.S.
EPA, 2019, section 6.2.1, Table 11-8), and metabolic effects, including diabetes (U.S. EPA,
2019, section 7.3.1). The 2019 ISA notes "more limited evidence from respiratory morbidity"
(U.S. EPA, 2019, p. 11-101) such as development of chronic obstructive pulmonary disease
(COPD) (U.S. EPA, 2019, section 5.2.1) to support the biological plausibility of mortality due to
long-term PM2.5 exposures (U.S. EPA, 2019, section 11.2.1).

Taken together, recent studies, i.e., those evaluated in the 2019 ISA and in the ISA
Supplement, reaffirm and further strengthen the body of evidence from the 2009 ISA for the
relationship between long-term PM2.5 exposure and mortality. Epidemiologic studies evaluated
in the 2019 ISA, including recent studies evaluated in the ISA Supplement, consistently report
positive associations between long-term PM2.5 exposure and mortality across different
geographic locations, populations, and analytic approaches (U.S. EPA, 2019; U.S. EPA, 2022,
section 3.2.2.4) .

As such, these studies reduce key uncertainties identified in the previous review,
including those related to potential copollutant confounding, and provide additional information
on the shape of the C-R curve. As assessed in the 2019 ISA, experimental and epidemiologic
evidence for cardiovascular effects, and respiratory effects to a more limited degree, supports the
plausibility of mortality due to long-term PM2.5 exposures. The 2019 ISA concludes that,
"collectively, this body of evidence is sufficient to conclude that a causal relationship exists
between long-term PM2.5 exposure and total mortality" (U.S. EPA, 2019, section 11.2.7; p. 11-
102) which is supported and extended by recent evidence evaluated in the ISA Supplement (U.S.
EPA, 2022, section 3.2.2.4).

Short-term PM2.5 exposures

The 2009 ISA concluded that "a causal relationship exists between short-term exposure
to PM2.5 and mortality" (U.S. EPA, 2009). This conclusion was based on the evaluation of both
multi- and single-city epidemiologic studies that consistently reported positive associations
between short-term PM2.5 exposure and non-accidental mortality. These associations were
strongest, in terms of magnitude and precision, primarily at lags of 0 to 1 days. Examination of
the potential confounding effects of gaseous copollutants was limited, though evidence from
single-city studies indicated that gaseous copollutants have minimal effect on the PM2.5-mortality
relationship (i.e., associations remain robust to inclusion of other pollutants in copollutant
models). The evaluation of cause-specific mortality found that effect estimates were larger in
magnitude, but also had larger confidence intervals, for respiratory mortality compared to
cardiovascular mortality. Although the largest mortality risk estimates were for respiratory
mortality, the interpretation of the results was complicated by the limited coherence from studies

3-26

-------
of respiratory morbidity. However, the evidence from studies of cardiovascular morbidity
provided both coherence and biological plausibility for the relationship between short-term PM2.5
exposure and cardiovascular mortality.

Multicity studies evaluated in the 2019 ISA and ISA Supplement provide evidence of
primarily positive associations between daily PM2.5 exposures and mortality, with percent
increases in total mortality ranging from 0.19% (Lippmann et al., 2013) to 2.80% (Kloog et al.,
2013)11 at lags of 0 to 1 days in single pollutant-models. Whereas many studies assign exposures
using data from ambient monitors, other studies employ hybrid modeling approaches, which
estimate PM2.5 concentrations using data from a variety of sources (i.e., satellites, land use
information, air quality models, and PM2.5 monitors) and enable the inclusion of less urban and
more rural locations in analyses (Kloog et al., 2013, Lee et al., 2015, Shi et al., 2016; Lee et al.,
2015). Consistent with the evidence assessed in previous IS As, recent studies evaluated in the
ISA Supplement report more variable results with wider confidence intervals for respiratory
mortality (Lavigne et al., 2018; Shin et al., 2021).

Some studies have expanded the examination of potential confounders, including long-
term temporal trends, weather, and co-occurring pollutants. Mortality associations were found to
remain positive, although in some cases were attenuated, when using different approaches to
account for temporal trends or weather covariates (U.S. EPA, 2019, section 11.1.5.1). For
example, Sacks et al. (2012) examined the influence of model specification using the approaches
for confounder adjustment from models employed in several multicity studies within the context
of a common data set (U.S. EPA, 2019, section 11.1.5.1). These models use different approaches
to control for long-term temporal trends and the potential confounding effects of weather. The
authors report that associations between daily PM2.5 and cardiovascular mortality were similar
across models, with the percent increase in mortality ranging from 1.5-2.0%) (U.S. EPA, 2019,
Figure 11-4). Thus, alternative approaches to controlling for long-term temporal trends and for
the potential confounding effects of weather may influence the magnitude of the association
between PM2.5 exposures and mortality but have not been found to influence the direction of the
observed association (U.S. EPA, 2019, section 11.1.5.1). Taken together, the 2019 ISA and the
ISA Supplement conclude that recent multicity studies conducted in the U.S., Canada, Europe,
and Asia continue to provide consistent evidence of positive associations between short-term
PM2.5 exposures and total mortality across studies that use different approaches to control for the
potential confounding effects of weather (e.g., temperature) (U.S. EPA, 2019, section 1.4.1.5.1;
U.S. EPA, 2022, section 3.2.1.2)

11 As detailed in the Preface to the ISA, risk estimates are for a 10 |ig/m3 increase in 24-hour avg PM2 5
concentrations, unless otherwise noted (U.S. EPA, 2019).

3-27

-------
With regard to copollutants, studies evaluated in the 2019 ISA provide additional
evidence that associations between short-term PM2.5 exposures and mortality remain positive and
relatively unchanged in copollutant models with both gaseous pollutants and PM10-2.5 (U.S. EPA,
2019, section 11.1.4). Additionally, the low (r < 0.4) to moderate correlations (r = 0.4-0.7)
between PM2.5 and gaseous pollutants and PM10-2.5 increase the confidence in PM2.5 having an
independent effect on mortality (U.S. EPA, 2019, section 11.1.4). Consistent with the studies
evaluated in the 2019 ISA, studies evaluated in the ISA Supplement that used data from more
recent years also indicate that associations between short-term PM2.5 exposure and mortality
remain unchanged in copollutant models. However, the evidence indicates that the association
could be larger in magnitude in the presence of some co-occuring pollutants such as oxidant
gases (Lavigne et al., 2018; Shin et al., 2021).

The generally positive associations reported with mortality are supported by a small group
of studies employing alternative methods for confounder control or quasi-experimental statistical
approaches (U.S. EPA, 2019, section 11.1.2.1). For example, two studies by Schwartz et al.
(Schwartz et al., 2015; Schwartz et al., 2017) report associations between PM2.5 instrumental
variables and mortality (U.S. EPA, 2019, Table 11-2), including in an analysis limited to days
with 24-hour average PM2.5 concentrations <30 [j,g/m3 (Schwartz et al., 2017). In addition to the
main analyses, these studies conducted Granger-like causality tests as sensitivity analyses to
examine whether there was evidence of an association between mortality and PM2.5 after the day
of death, which would support the possibility that unmeasured confounders were not accounted
for in the statistical model. Neither study reports evidence of an association with PM2.5 after
death (i.e., they do not indicate unmeasured confounding). A quasi-experimental- study
examines whether a specific regulatory action in Tokyo, Japan (i.e., a diesel emission control
ordinance) resulted in a subsequent reduction in daily mortality (Yorifuji et al., 2016). The
authors report a reduction in mortality in Tokyo due to the ordinance, compared to Osaka, which
did not have a similar diesel emission control ordinance in place. In another study, Schwartz et
al. (2018b) utilized three statistical methods including instrumental variable analysis, a negative
exposure control, and marginal structural models to estimate the association between PM2.5 and
daily mortality (Schwartz et al., 2018b). Results from this study continue to support a
relationship between short-term PM2.5 exposure and mortality. Additional epidemiologic studies
evaluated in the ISA Supplement that employed alternative methods for confounder control to
examine the association between short-term PM2.5 exposure and mortality also report consistent
positive associations in studies that examine effects across multiple cities in the U.S. (U.S. EPA,
2022).

The positive associations for total mortality reported across the majority of studies
evaluated are further supported by analyses reporting generally consistent, positive associations

3-28

-------
with both cardiovascular and respiratory mortality (U.S. EPA, 2019, section 11.1.3). Recent
multicity studies evaluated in the ISA Supplement add to the body of evidence indicating a
relationship between short-term PM2.5 exposure and cause-specific mortality, with more
variability in the magnitude and precision of associations for respiratory mortality (U.S. EPA,
2022, Figure 3-14). For both cardiovascular and respiratory mortality, there has been a limited
assessment of potential copollutant confounding, though initial evidence indicates that
associations remain positive and relatively unchanged in models with gaseous pollutants and
PM10-2.5. This evidence further supports the copollutant analyses conducted for total mortality.
The strong evidence for ischemic events and heart failure, as detailed in the assessment of
cardiovascular morbidity (U.S. EPA, 2019, chapter 6), provides biological plausibility for PM2.5-
related cardiovascular mortality, which comprises the largest percentage of total mortality
(i.e., -33%) (NHLBI, 2017). Although there is evidence for exacerbations of COPD and asthma,
the collective body of respiratory morbidity evidence provides limited biological plausibility for
PM2.5-related respiratory mortality (U.S. EPA, 2019, chapter 5).

In the 2009 ISA, one of the main uncertainties identified was the regional and city-to-city
heterogeneity in PM2.5 mortality associations. Recent studies examine both city-specific as well
as regional characteristics to identify the underlying contextual factors that could contribute to
this heterogeneity (U.S. EPA, 2019, section 11.1.6.3). Analyses focusing on effect modification
of the short-term PM2.5 exposure and mortality relationship by PM2.5 components, regional
patterns in PM2.5 components and city specific differences in composition and sources indicate
some differences in the PM2.5 composition and sources across cities and regions, but these
differences do not fully explain the observed heterogeneity. Additional studies find that factors
related to potential exposure differences, such housing stock and commuting, as well as city
specific factors (e.g., land use, port volume, and traffic information), may explain some of the
observed heterogeneity (U.S. EPA, 2019, section 11.1.6.3). Collectively, studies evaluated in the
2019 ISA and the ISA Supplement indicate that the heterogeneity in PM2.5 mortality risk
estimates cannot be attributed to one factor, but instead a combination of factors including, but
not limited to, PM composition and sources as well as community characteristics that could
influence exposures (U.S. EPA, 2019, section 11.1.12; U.S. EPA, 2022, section 3.2.1.2.1).

A number of studies conducted systematic evaluations of the lag structure of associations
for the PM2.5-mortality relationship by examining either a series of single-day or multiday lags
and these studies continue to support an immediate effect (i.e., lag 0 to 1 days) of short-term
PM2.5 exposures on mortality (U.S. EPA, 2019, section 11.1.8.1; U.S. EPA, 2022, section
3.2.1.1). Recent studies also conducted analyses comparing the traditional 24-hour average
exposure metric with a sub-daily metric (i.e., 1-hour max). These initial studies provide evidence

3-29

-------
of a similar pattern of associations for both the 24-hour average and 1-hour max metric, with the
association larger in magnitude for the 24-hour average metric.

Multicity studies indicate that positive and statistically significant associations with
mortality persist in analyses restricted to short-term (24-hour average PM2.5 concentrations)
PM2.5 exposures below 35 |j,g/m3 (Lee et al., 2015),12 below 30 |j,g/m3 (Shi et al., 2016), and
below 25 |j,g/m3 (Di et al., 2017a), indicating that risks associated with short-term PM2.5
exposures are not disproportionately driven by the peaks of the air quality distribution.

Additional studies examine the shape of the C-R relationship and whether a threshold exists
specifically for PM2.5 (U.S. EPA, 2019, section 11.1.10). These studies have used various
statistical approaches and consistently demonstrate a linear relationship with no evidence of a
threshold. Moreover, recent studies evaluated in the ISA Supplement provide additional support
for a linear, no-threshold C-R relationship between short-term PM2.5 exposure and mortality,
with confidence in the shape decreasing at concentrations below 5 |ig/m3 (Liu et al., 2019;
Lavigne et al., 2018). Recent analyses provide initial evidence indicating that PM2.5-mortality
associations persist and may be larger in magnitude (i.e., a steeper slope) at lower concentrations
(e.g., Di et al., 2017a; Figure 11-12 in U.S. EPA, 2019). However, given the limited data
available at the lower end of the distribution of ambient PM2.5 concentrations, the shape of the C-
R curve remains uncertain at these low concentrations. Although difficulties remain in assessing
the shape of the PM2.5-mortality C-R relationship, to date, studies have not conducted systematic
evaluations of alternatives to linearity, and recent studies continue to provide evidence of a no-
threshold linear relationship, with less confidence at concentrations lower than 5 |ig/m3.

Overall, recent epidemiologic studies build upon and extend the conclusions of the 2009
ISA for the relationship between short-term PM2.5 exposures and total mortality. Supporting
evidence for PM2.5 related cardiovascular morbidity, and more limited evidence from respiratory
morbidity, provides biological plausibility for mortality due to short-term PM2.5 exposures. The
primarily positive associations observed across studies conducted in diverse geographic locations
is further supported by the results from copollutant analyses indicating robust associations, along
with evidence from analyses of the concentration-response relationship. The 2019 ISA states
that, collectively, "this body of evidence is sufficient to conclude that a causal relationship exists
between short-term PM2.5 exposure and total mortality" (U.S. EPA, 2019, pp. 11-58). Recent
evidence evaluated in the ISA Supplement provides "additional support to the evidence base that
contributed to the conclusion of a causal relationship between short-term PM2.5 exposure and
mortality" (U.S. EPA, 2022, section 3.2.1.4, p. 3-69).

12 Lee et al. (2015) also report that positive and statistically significant associations between short-term PM2 5
exposures and mortality persist in analyses restricted to areas with long-term concentrations below 12 |.ig/m\

3-30

-------
3.3,1,2 Cardiovascular Effects

Long-term PM2.5 exposures

The scientific evidence reviewed in the 2009 ISA was "sufficient to infer a causal
relationship between long-term PM2.5 exposure and cardiovascular effects" (U.S. EPA, 2009).
The strongest line of evidence comprised findings from several large epidemiologic studies of
U.S. and Canadian cohorts that consistently showed positive associations between long-term
PM2.5 exposure and cardiovascular mortality (Krewski et al., 2009, Laden et al., 2006, Miller et
al., 2007, Pope et al., 2004). Studies of long-term PM2.5 exposure and cardiovascular morbidity
were limited in number. Biological plausibility and coherence with the epidemiologic findings
were provided by studies using genetic mouse models of atherosclerosis demonstrating enhanced
atherosclerotic plaque development and inflammation, as well as changes in measures of
impaired heart function, following 4- to 6-month exposures to PM2.5 concentrated ambient
particles (CAPs), and by a limited number of studies reporting CAPs-induced effects on
coagulation factors, vascular reactivity, and worsening of experimentally induced hypertension
in mice (U.S. EPA, 2009).

Consistent with the evidence assessed in the 2009 ISA, the 2019 ISA concludes that
recent studies, together with the evidence available in previous reviews, supports a causal
relationship between long-term exposure to PM2.5 and cardiovascular effects. Additionally,
recent epidemiologic studies published since the completion of the 2019 ISA and evaluated in
the ISA Supplement expands the body of evidence and further supports such a conclusion (U.S.
EPA, 2022). As discussed above (section 3.3.1.1), results from U.S. and Canadian cohort studies
evaluated in the 2019 ISA consistently report positive associations between long-term PM2.5
exposure and cardiovascular mortality (U.S. EPA, 2019, Figure 6-19) in evaluations conducted at
varying spatial scales and employing a variety of exposure assessment and statistical methods
(U.S. EPA, 2019, section 6.2.10). Positive associations between long-term PM2.5 exposures and
cardiovascular mortality are generally robust in copollutant models adjusted for ozone, NO2,
PM10-2.5, or SO2. In addition, most of the results from analyses examining the shape of the C-R
relationship for cardiovascular mortality support a linear relationship with long-term PM2.5
exposures and do not identify a threshold below which effects do not occur (U.S. EPA, 2019,
section 6.2.16; Table 6-52).

The body of literature examining the relationship between long-term PM2.5 exposure and
cardiovascular morbidity has greatly expanded since the 2009 ISA, with positive associations
reported in several cohorts in studies evaluated in the 2019 ISA (U.S. EPA, 2019, section 6.2).
Though results for cardiovascular morbidity are less consistent than those for cardiovascular
mortality (U.S. EPA, 2019, section 6.2), studies in the 2019 ISA and the ISA Supplement
provides some evidence for associations between long-term PM2.5 exposures and the progression

3-31

-------
of cardiovascular disease. Positive associations with cardiovascular morbidity (e.g., coronary
heart disease, stroke, arrhythmias, myocardial infarction (MI), and atherosclerosis progression)
are observed in several epidemiologic studies (U.S. EPA, 2019, sections 6.2.2. to 6.2.9; U.S.
EPA, 2022, section 3.1.2.2).

Recent studies published since the literature cutoff date of the 2019 PM ISA further
assessed the relationship between long-term PM2.5 exposure and cardiovascular effects by
conducting accountability analyses or by using alternative methods for confounder control in
evaluating the association between long-term PM2.5 exposure and cardiovascular hospital
admissions (U.S. EPA, 2022, section 3.1.2.3). Increased confidence in the relationship between
long-term PM2.5 exposure and cardiovascular effects is provided by additional studies that use
methods that reduce uncertainties related to potential confounding with statistical methods and/or
study design approaches. For example, to control for potential confounding Yazdi et al. (2021)
used a doubly robust additive model (DRAM) and found an association between long-term
exposure to PM2.5 and cardiovascular effects, including myocardial infarction, stoke, and atrial
fibrillation, among the Medicare population. Additionally, an accountability study by Henneman
et al. (2019b) utilized a difference-in-difference (DID) approach to determine the relationship
between coal-fueled power plant emissions and cardiovascular effects and found that reductions
in PM2.5 concentrations resulted in reductions of cardiovascular-related hospital admissions.
Overall, these studies report consistent findings that long-term PM2.5 exposure is related to
increased hospital admissions for a variety of cardiovascular disease outcomes among large
nationally representative cohorts and provide additional support for a relationship between long-
term PM2.5 exposure and cardiovascular effects.

The positive associations reported in epidemiologic studies are supported by
toxicological evidence for increased plaque progression in mice following long-term exposure to
PM2.5 collected from multiple locations across the U.S. (U.S. EPA, 2019, section 6.2.4.2). A
small number of epidemiologic studies also report positive associations between long-term PM2.5
exposure and heart failure, changes in blood pressure, and hypertension (U.S. EPA, 2019,
sections 6.2.5 and 6.2.7). Associations with heart failure are supported by animal toxicological
studies demonstrating decreased cardiac contractility and function, and increased coronary artery
wall thickness following long-term PM2.5 exposure (U.S. EPA, 2019, section 6.2.5.2). Similarly,
a limited number of animal toxicological studies demonstrating a relationship between long-term
exposure to PM2.5 and consistent increases in blood pressure in rats and mice are coherent with
epidemiologic studies reporting positive associations between long-term exposure to PM2.5 and
hypertension. Moreover, a number of studies assessed in the ISA Supplement focusing on
morbidity outcomes, including those that focused on incidence of MI, atrial fibrillation (AF),
stroke, and congestive heart failure (CHF), expand the evidence pertaining to the shape of the C-

3-32

-------
R relationship between long-term PM2.5 exposure and cardiovascular effects. Additionally,
studies evaluated in the ISA Supplement report positive associations among those with pre-
existing conditions, among patients followed after a cardiac event procedure, and among those
with a first hospital admission for heart attacks among older adults enrolled in Medicare (U.S.
EPA, 2022, sections 3.1.1 and 3.1.2). A number of these studies use statistical techniques that
allow for departures from linearity (U.S. EPA, 2022, Table 3-3), and generally support the
evidence characterized in the 2019 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 (U.S. EPA, 2022, section 3.1.2.2.9).13 Moreover, several recent
epidemiologic studies evaluated in the ISA Supplement reported that the association between
long-term PM2.5 exposure with stroke persisted after adjustment for NO2 but was attenuated in
the model with O3 and oxidant gases represented by the redox weighted average of NO2 and O3
(U.S. EPA, 2022, section 3.1.2.2.8).

Longitudinal epidemiologic analyses also report positive associations with markers of
systemic inflammation (U.S. EPA, 2019, section 6.2.11), coagulation (U.S. EPA, 2019, section
6.2.12), and endothelial dysfunction (U.S. EPA, 2019, section 6.2.13). These results are coherent
with animal toxicological studies generally reporting increased markers of systemic
inflammation, oxidative stress, and endothelial dysfunction (U.S. EPA, 2019, section 6.2.12.2
and 6.2.14).

The 2019 ISA concludes that there is consistent evidence from multiple epidemiologic
studies illustrating that long-term exposure to PM2.5 is associated with mortality from
cardiovascular causes. Epidemiologic studies in the ISA Supplement support and extend the
findings characterized in the 2019 ISA, providing additional evidence of positive associations
between long-term PM2.5 exposure and cardiovascular morbidity (U.S. EPA, 2022, section
3.1.2.2.) 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 support the independence of the PM2.5
associations (U.S. EPA, 2019, Table 3-2; U.S. EPA, 2022). Additional evidence of the
independent effect of PM2.5 on the cardiovascular system is provided by experimental studies in
animals, which 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. The combination of epidemiologic and experimental evidence results in
the 2019 ISA conclusion that "a causal relationship exists between long-term exposure to PM2.5

13 As noted above for mortality, uncertainty in the shape of the C-R relationship increases near the upper and lower
ends of the distribution due to limited data.

3-33

-------
and cardiovascular effects" (U.S. EPA, 2019, section 6.2.18). Studies evaluated in the ISA
Supplement support and extend the evidence that contributed to the conclusion of a causal
relationship between long-term PM2.5 exposure and cardiovascular effects (U.S. EPA, 2022,
section 3.1.2.2).

Short-term PM2.5 exposures

The 2009 ISA concluded that "a causal relationship exists between short-term exposure
to PM2.5 and cardiovascular effects" (U.S. EPA, 2009). The strongest evidence in the 2009 ISA
was from epidemiologic studies of emergency department (ED) visits and hospital admissions
for IHD and HF, with supporting evidence from epidemiologic studies of cardiovascular
mortality (U.S. EPA, 2009). Animal toxicological studies provided coherence and biological
plausibility for the positive associations reported with myocardial ischemia ED visit and hospital
admissions. These included studies reporting reduced myocardial blood flow during ischemia
and studies indicating altered vascular reactivity. In addition, effects of PM2.5 exposure on a
potential indicator of ischemia (i.e., ST segment depression on an electrocardiogram) were
reported in both animal toxicological and epidemiologic panel studies.14 Key uncertainties from
the 2009 ISA resulted from inconsistent results across disciplines with respect to the relationship
between short-term exposure to PM2.5 and changes in blood pressure, blood coagulation markers,
and markers of systemic inflammation. In addition, while the 2009 ISA identified a growing
body of evidence from controlled human exposure and animal toxicological studies, uncertainties
remained with respect to biological plausibility.

Recent evidence assessed in the 2019 ISA and the ISA Supplement supports and extends
the evidence from the 2009 ISA indicating that there is a causal relationship between short-term
PM2.5 exposure and cardiovascular effects. This includes generally positive associations
observed in multicity epidemiologic studies of emergency department visits and hospital
admissions for IHD, heart failure (HF), and combined cardiovascular-related endpoints. In
particular, nationwide studies of older adults (65 years and older) using Medicare records report
positive associations between PM2.5 exposures and hospital admissions for HF (U.S. EPA, 2019,
section 6.1.3.1). Moreover, recent multicity studies, published after the literature cutoff date of
the 2019 ISA, are coherent with studies evaluated in the 2019 ISA that report positive association
between short-term PM2.5 exposure and ED visits and hospital admission for IHD, heart attacks,
and HF (U.S. EPA, 2022, section 3.1). Epidemiologic studies conducted in single cities
contribute some support, though associations reported in single-city studies are less consistently
positive than in multicity studies, and include a number of studies reporting null associations

14 Some animal studies included in the 2009 ISA examined exposures to mixtures, such as motor vehicle exhaust or
woodsmoke. In these studies, it was unclear if the resulting cardiovascular effects could be attributed specifically
to the fine particle component of the mixture.

3-34

-------
(U.S. EPA, 2019, sections 6.1.2 and 6.1.3). When considered as a whole; however, the recent
body of IHD and HF epidemiologic evidence supports the evidence from previous IS As
reporting mainly positive associations between short-term PM2.5 concentrations and emergency
department visits and hospital admissions.

The ISA Supplement also includes a small number of epidemiologic studies, published
since the literature cutoff date for the 2019 ISA, including accountability analyses and
epidemiologic studies that employ alternative methods for confounder control to evaluate the
association between short-term PM2.5 exposure and cardiovascular-related effects (U.S. EPA,
2022, section 3.1.1.3). These studies report positive associations across a number of statistical
approaches, providing additional support for a relationship between short-term PM2.5 exposure
and cardiovascular effects, while also reducing uncertainties related to potential confounder bias.

Consistent with the evidence assessed in the 2019 ISA, some studies evaluated in the ISA
Supplement report no evidence of an association with stroke, regardless of stroke subtype.
Additionally, as in the 2019 ISA, evidence evaluated in the ISA Supplement 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 (U.S. EPA, 2022, section 3.1.1.2).

A number of controlled human exposure, animal toxicological, and epidemiologic panel
studies provide evidence that PM2.5 exposure could plausibly result in IHD or HF through
pathways that include endothelial dysfunction, arterial thrombosis, and arrhythmia (U.S. EPA,
2019, section 6.1.1). The most consistent evidence from recent controlled human exposure
studies is for endothelial dysfunction, as measured by changes in brachial artery diameter or flow
mediated dilation. All but one of the available controlled human exposure studies examining the
potential for endothelial dysfunction report an effect of PM2.5 exposure on measures of blood
flow (U.S. EPA, 2019, section 6.1.13.2). These studies report variable results regarding the
timing of the effect and the mechanism by which reduced blood flow occurs (i.e., availability vs
sensitivity to nitric oxide). Some controlled human exposure studies using CAPs report evidence
for small increases in blood pressure (U.S. EPA, 2019, section 6.1.6.3). In addition, although not
entirely consistent, there is also some evidence across controlled human exposure studies for
conduction abnormalities/arrhythmia (U.S. EPA, 2019, section 6.1.4.3), changes in heart rate
variability (HRV) (U.S. EPA, 2019, section 6.1.10.2), changes in hemostasis that could promote
clot formation (U.S. EPA, 2019, section 6.1.12.2), and increases in inflammatory cells and
markers (U.S. EPA, 2019, section 6.1.11.2). A recent study by Wyatt et al. (2020a) adds to the
limited evidence base of controlled human exposure studies conducted at near ambient PM2.5
concentrations. The study, completed in healthy young adults subject to intermittent exercise,
reported evidence of significant results for some cardiovascular effects (e.g., systematic

3-35

-------
inflammation markers, including C-reactive protein (CRP), and cardiac repolarization). Thus,
when taken as a whole, controlled human exposure studies are coherent with epidemiologic
studies in that they demonstrate short-term exposures to PM2.5 may result in the types of
cardiovascular endpoints that could lead to emergency department visits and hospital admissions
in some people.

Animal toxicological studies published since the 2009 ISA also support a relationship
between short-term PM2.5 exposure and cardiovascular effects. A study demonstrating decreased
cardiac contractility and left ventricular pressure in mice is coherent with the results of
epidemiologic studies reporting associations between short-term PM2.5 exposure and heart failure
(U.S. EPA, 2019, section 6.1.3.3). In addition, and as with controlled human exposure studies,
there is generally consistent evidence in animal toxicological studies for indicators of endothelial
dysfunction (U.S. EPA, 2019, section 6.1.13.3). Studies in animals also provide evidence for
changes in a number of other cardiovascular endpoints following short-term PM2.5 exposure.
Although not entirely consistent, these studies provide some evidence of conduction
abnormalities and arrhythmia (U.S. EPA, 2019, section 6.1.4.4), changes in HRV (U.S. EPA,
2019, section 6.1.10.3), changes in blood pressure (U.S. EPA, 2019, section 6.1.6.4), and
evidence for systemic inflammation and oxidative stress (U.S. EPA, 2019, section 6.1.11.3).

In summary, recent evidence evaluated in the 2019 ISA and the ISA Supplement further
supports and extends the conclusions of the evidence base reported in the 2009 ISA. In support
of epidemiologic studies reporting robust associations in copollutant models, direct evidence for
an independent effect of PM2.5 on cardiovascular effects can be found in a number of controlled
human exposure and animal toxicological studies. Coherent with these results are epidemiologic
panel studies reporting that PM2.5 exposure is associated with some of the same cardiovascular
endpoints reported in experimental studies. For these effects, there are inconsistencies in results
across some animal toxicological, controlled human exposure, and epidemiologic panel studies,
though this may be due to substantial differences in study design and/or study populations.
Overall, the results from 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 changes in blood
pressure provide coherence and biological plausibility for the consistent results from
epidemiologic studies observing positive associations between short-term PM2.5 concentrations
and IHD and HF, and ultimately cardiovascular mortality. The 2019 ISA concludes that, overall,
"there continues to be sufficient evidence to conclude that a causal relationship exists between
short-term PM2.5 exposure and cardiovascular effects" (U.S. EPA, 2019, p. 6-138), which is
further supported by recent studies evaluated in the ISA Supplement (U.S. EPA, 2022, section
3.1.1.4).

3-36

-------
3,3,1,3 Respiratory Effects

Long-term PM2.5 exposures

The 2009 ISA concluded that "a causal relationship is likely to exist between long-term
PM2.5 exposure and respiratory effects" (U.S. EPA, 2009). This conclusion was based mainly on
epidemiologic evidence demonstrating associations between long-term PM2.5 exposure and
changes in lung function or lung function growth in children. Biological plausibility was
provided by a single animal toxicological study examining pre- and post-natal exposure to PM2.5
CAPs, which found impaired lung development. Epidemiologic evidence for associations
between long-term PM2.5 exposure and other respiratory outcomes, such as the development of
asthma, allergic disease, and COPD; respiratory infection; and the severity of disease was
limited, both in the number of studies available and the consistency of the results. Experimental
evidence for other outcomes was also limited, with one animal toxicological study reporting that
long-term exposure to PM2.5 CAPs results in morphological changes in nasal airways of healthy
animals. Other animal studies examined exposure to mixtures, such as motor vehicle exhaust and
woodsmoke, and effects were not attributed specifically to the particulate components of the
mixture.

Cohort studies evaluated in the 2019 ISA provided additional support for the relationship
between long-term PM2.5 exposure and decrements in lung function growth (as a measure of lung
development), indicating a robust and consistent association across study locations, exposure
assessment methods, and time periods (U.S. EPA, 2019, section 5.2.13). This relationship was
further supported by a retrospective study that reports an association between declining PM2.5
concentrations and improvements in lung function growth in children (U.S. EPA, 2019,
section 5.2.11). Epidemiologic studies also examine asthma development in children (U.S. EPA,
2019, section 5.2.3), with prospective cohort studies reporting generally positive associations,
though several are imprecise (i.e., they report wide confidence intervals). Supporting evidence is
provided by studies reporting associations with asthma prevalence in children, with childhood
wheeze, and with exhaled nitric oxide, a marker of pulmonary inflammation (U.S. EPA, 2019,
section 5.2.13). Additionally, animal toxicological study showing the development of an allergic
phenotype and an increase in a marker of airway responsiveness provides biological plausibility
for allergic asthma (U.S. EPA, 2019, section 5.2.13). Other epidemiologic studies report a
PM2.5-related acceleration of lung function decline in adults, while improvement in lung function
was observed with declining PM2.5 concentrations (U.S. EPA, 2019, section 5.2.11). A
longitudinal study found declining PM2.5 concentrations are also associated with an improvement
in chronic bronchitis symptoms in children, strengthening evidence reported in the 2009 ISA for
a relationship between increased chronic bronchitis symptoms and long-term PM2.5 exposure
(U.S. EPA, 2019, section 5.2.11). A common uncertainty across the epidemiologic evidence is

3-37

-------
the lack of examination of copollutants to assess the potential for confounding. While there is
some evidence that associations remain robust in models with gaseous pollutants, a number of
these studies examining copollutant confounding were conducted in Asia, and thus have limited
generalizability due to high annual pollutant concentrations.

When taken together, the 2019 ISA concludes that the "epidemiologic evidence strongly
supports a relationship with decrements in lung function growth in children" and "with asthma
development in children, with increased bronchitis symptoms in children with asthma, with an
acceleration of lung function decline in adults, and with respiratory mortality and cause-specific
respiratory mortality for COPD and respiratory infection" (U.S. EPA, 2019, p. 1-34). In support
of the biological plausibility of such associations reported in epidemiologic studies of respiratory
health effects, animal toxicological studies continue to provide direct evidence that long-term
exposure to PM2.5 results in a variety of respiratory effects. Animal studies in the 2019 ISA show
pulmonary oxidative stress, inflammation, and morphologic changes in the upper (nasal) and
lower airways. Other results show that changes are consistent with the development of allergy
and asthma, and with impaired lung development. Overall, the 2019 ISA concludes that "the
collective evidence is sufficient to conclude that a causal relationship is likely to exist between
long-term PM2.5 exposure and respiratory effects" (U.S. EPA, 2019, section 5.2.13).

Short-term PM2.5 exposures

The 2009 ISA (U.S. EPA, 2009) concluded that a "causal relationship is likely to exist"
between short-term PM2.5 exposure and respiratory effects. This conclusion was based mainly on
the epidemiologic evidence demonstrating positive associations with various respiratory effects.
Specifically, the 2009 ISA described epidemiologic evidence as consistently showing
PM2.5-associated increases in hospital admissions and emergency department visits for chronic
obstructive pulmonary disease (COPD) and respiratory infection among adults or people of all
ages, as well as increases in respiratory mortality. These results were supported by studies
reporting associations with increased respiratory symptoms and decreases in lung function in
children with asthma, though the epidemiologic evidence was inconsistent for hospital
admissions or emergency department visits for asthma. Studies examining copollutants models
showed that PM2.5 associations with respiratory effects were robust to inclusion of CO or SO2 in
the model, but often were attenuated (though still positive) with inclusion of O3 or NO2. In
addition to the copollutants models, evidence supporting an independent effect of PM2.5 exposure
on the respiratory system was provided by animal toxicological studies of PM2.5 CAPs
demonstrating changes in some pulmonary function parameters, as well as inflammation,
oxidative stress, injury, enhanced allergic responses, and reduced host defenses. Many of these
effects have been implicated in the pathophysiology for asthma exacerbation, COPD
exacerbation, or respiratory infection. In the few controlled human exposure studies conducted in

3-38

-------
individuals with asthma or COPD, PM2.5 exposure mostly had no effect on respiratory
symptoms, lung function, or pulmonary inflammation. Available studies in healthy people also
did not clearly demonstrate respiratory effects following short-term PM2.5 exposures.

Epidemiologic studies evaluated in the 2019 ISA continue to provide strong evidence for
a relationship between short-term PM2.5 exposure and several respiratory-related endpoints,
including asthma exacerbation (U.S. EPA, 2019, section 5.1.2.1), COPD exacerbation (U.S.
EPA, 2019, section 5.1.4.1), and combined respiratory-related diseases (U.S. EPA, 2019, section
5.1.6), particularly from studies examining emergency department visits and hospital admissions.
The generally positive associations between short-term PM2.5 exposure and asthma and COPD
emergency department visits and hospital admissions 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 (U.S. EPA, 2019, sections
5.1.2.2 and 5.4.1.2). The collective body of epidemiologic evidence for asthma exacerbation is
more consistent in children than in adults. Additionally, epidemiologic studies examining the
relationship between short-term PM2.5 exposure and respiratory mortality provide evidence of
consistent positive associations, demonstrating a continuum of effects (U.S. EPA, 2019, section
5.1.9).

Building off the studies evaluated in the 2009 and 2019 ISA, epidemiologic studies
expand the assessment of potential copollutant confounding. There is 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) (U.S. EPA,
2019, section 5.1.10.1).

In the 2019 ISA, the uncertainty related to whether there is an independent effect of PM2.5
on respiratory health is also partially addressed by findings from animal toxicological studies.
Specifically, short-term exposure to PM2.5 enhanced asthma-related responses in an animal
model of allergic airways disease and enhanced lung injury and inflammation in an animal model
of COPD (U.S. EPA, 2019, sections 5.1.2.4.4 and 5.1.4.4.3). The experimental evidence
provides biological plausibility for some respiratory-related endpoints, including limited
evidence of altered host defense and greater susceptibility to bacterial infection as well as
consistent evidence of respiratory irritant effects. Animal toxicological evidence for other
respiratory effects is inconsistent. A recent study by Wyatt et al. (2020a) was conducted at near
ambient PM2.5 concentrations and adds to the limited evidence base of controlled human
exposure studies. The study, completed in healthy young adults subject to intermittent exercise,
reported evidence for some significant respiratory effects (e.g., decrease in lung function),

3-39

-------
although the findings of this study are inconsistent with the controlled human exposure studies
evaluated in the 2019 ISA (U.S. EPA 2019, section 5.1.7.2, 5.1.2.3., and 6.1.11.2.1).

The 2019 ISA concludes that "[t]he strongest evidence of an effect of short-term PM2.5
exposure on respiratory effects is provided by epidemiologic studies of asthma and COPD
exacerbation. While animal toxicological studies provide biological plausibility for these
findings, some uncertainty remains with respect to the independence of PM2.5 effects" (U.S.
EPA, 2019, p. 5-155). When taken together, the 2019 ISA concludes that this evidence "is
sufficient to conclude that a causal relationship is likely to exist between short-term PM2.5
exposure and respiratory effects" (U.S. EPA, 2019, p. 5-155).

3.3.1.4 Cancer - Long-term PM2.5 Exposures

The 2009 ISA concluded that the overall body of evidence was "suggestive of a causal
relationship between relevant PM2.5 exposures and cancer" (U.S. EPA, 2009). This conclusion
was based primarily on positive associations observed in a limited number of epidemiologic
studies of lung cancer mortality. The few epidemiologic studies that had evaluated PM2.5
exposure and lung cancer incidence or cancers of other organs and systems generally did not
show evidence of an association. Toxicological studies did not focus on exposures to specific
PM size fractions, but rather investigated the effects of exposures to total ambient PM, or other
source-based PM such as wood smoke. Collectively, results of in vitro studies were consistent
with the larger body of evidence demonstrating that ambient PM and PM from specific
combustion sources are mutagenic and genotoxic. However, animal inhalation studies found
little evidence of tumor formation in response to chronic exposures. A small number of studies
provided preliminary evidence that PM exposure can lead to changes in methylation of DNA,
which may contribute to biological events related to cancer.

Since the 2009 ISA, additional cohort studies provide evidence that long-term PM2.5
exposure is positively associated with lung cancer mortality and with lung cancer incidence, and
provide initial evidence for an association with reduced cancer survival (U.S. EPA, 2019, section
10.2.5). Re-analyses of the ACS cohort using different years of PM2.5 data and follow up, along
with various exposure assignment approaches, provide consistent evidence of positive
associations between long-term PM2.5 exposure and lung cancer mortality (U.S. EPA, 2019,
Figure 10-3). Additional support for positive associations with lung cancer mortality is provided
by epidemiologic studies using individual-level data to control for smoking status, by studies of
people who have never smoked (though such studies generally report wide confidence intervals
due to the small number of lung cancer mortality cases within this population), and in analyses of
cohorts that relied upon proxy measures to account for smoking status (U.S. EPA, 2019, section
10.2.5.1.1). Although studies that have evaluated lung cancer incidence, including studies of

3-40

-------
people who have never smoked, are limited in number, studies in the 2019 ISA generally report
positive associations with long-term PM2.5 exposures (U.S. EPA, 2019, section 10.2.5.1.2). A
subset of the studies focusing on lung cancer incidence also examined histological subtype,
providing some evidence of positive associations for adenocarcinomas, the predominate subtype
of lung cancer observed in people who have never smoked (U.S. EPA, 2019, section 10.2.5.1.2).
Associations between long-term PM2.5 exposure and lung cancer incidence were found to remain
relatively unchanged, though in some cases confidence intervals widened, in analyses that
attempted to reduce exposure measurement error by accounting for length of time at residential
address or by examining different exposure assignment approaches (U.S. EPA, 2019, section

10.2.5.1.2).

The 2019 ISA evaluates the degree to which epidemiologic studies have addressed the
potential for confounding by copollutants and the shape of the concentration-response
relationship. To date, relatively few studies have evaluated the potential for copollutant
confounding of the relationship between long-term PM2.5 exposure and lung cancer mortality or
incidence. The small number of such studies have generally focused on O3 and report that PM2.5
associations remain relatively unchanged in copollutant models (U.S. EPA, 2019, section

10.2.5.1.3). However, available studies have not systematically evaluated the potential for
copollutant confounding by other gaseous pollutants or by other particle size fractions (U.S.
EPA, 2019, section 10.2.5.1.3). Compared to total (non-accidental) mortality (U.S. EPA, 2019,
section 10.2.4.1.4), fewer studies have examined the shape of the concentration-response curve
for cause-specific mortality outcomes, including lung cancer. Several studies of lung cancer
mortality and incidence have reported no evidence of deviations from linearity in the shape of
the concentration-response relationship (Lepeule et al., 2012; Raaschou-Nielsen et al., 2013;
Puett et al., 2014), though authors provided only limited discussions of results (U.S. EPA, 2019,
section 10.2.5.1.4).

In support of the biological plausibility of an independent effect of PM2.5 on lung cancer,
the 2019 ISA notes evidence from recent experimental and epidemiologic studies demonstrating
the potential role of PM2.5 exposure in genotoxicity (U.S. EPA, 2019, section 10.2.7). For
example, both in vitro and in vivo toxicological studies have shown that PM2.5 exposure can
result in DNA damage (U.S. EPA, 2019, section 10.2.2). Although such effects do not
necessarily equate to carcinogenicity, the evidence that PM exposure can damage DNA, and
elicit mutations, provides support for the plausibility of epidemiologic associations with lung
cancer mortality and incidence. Additional supporting studies indicate the occurrence of
micronuclei formation and chromosomal abnormalities (U.S. EPA, 2019, section 10.2.2.3), and
differential expression of genes that may be relevant to cancer pathogenesis, following PM
exposures. Experimental and epidemiologic studies that examine epigenetic effects indicate

3-41

-------
changes in DNA methylation, providing some support for PM2.5 exposure contributing to
genomic instability (U.S. EPA, 2019, section 10.2.3). Overall, there is limited evidence that
long-term PM2.5 exposure is associated with cancers in other organ systems, but there is some
evidence that PM2.5 exposure may reduce survival in individuals with cancer (U.S. EPA, 2019
section 10.2.7; U.S. EPA, 2022, section 2.1.1.4.1).

Epidemiologic evidence for associations between PM2.5 and lung cancer mortality and
incidence, together with evidence supporting the biological plausibility of such associations,
contributes to the 2019 ISA's conclusion that the evidence "is sufficient to conclude that a causal
relationship is likely to exist between long-term PM2.5 exposure and cancer" (U.S. EPA, 2019,
section 10.2.7).

3.3.1.5 Nervous System Effects

Long-term PM2.5 exposures

Reflecting the very limited evidence available in the 2012 review, the 2009 ISA did not
make a causality determination for long-term PM2.5 exposures and nervous system effects (U.S.
EPA, 2009). Since the last review, this body of evidence has grown substantially (U.S. EPA,
2019, section 8.2). Animal toxicology studies assessed in the 2019 ISA report that long-term
PM2.5 exposures can lead to morphologic changes in the hippocampus and to impaired learning
and memory. This evidence is consistent with epidemiologic studies reporting that long-term
PM2.5 exposure is associated with reduced cognitive function (U.S. EPA, 2019, section 8.2.5).
Further, while the evidence is limited, the presence of early markers of Alzheimer's disease
pathology has been demonstrated in rodents following long-term exposure to PM2.5 CAPs. These
findings support reported associations with neurodegenerative changes in the brain
(i.e., decreased brain volume), all-cause dementia, or hospitalization for Alzheimer's disease in a
small number of epidemiologic studies (U.S. EPA, 2019, section 8.2.6). Additionally, loss of
dopaminergic neurons in the substantia nigra, a hallmark of Parkinson disease, has been reported
in mice (U.S. EPA, 2019, section 8.2.4), though epidemiologic studies provide only limited
support for associations with Parkinson's disease (U.S. EPA, 2019, section 8.2.6). Overall, the
lack of consideration of copollutant confounding introduces some uncertainty in the
interpretation of epidemiologic studies of nervous system effects, but this uncertainty is partly
addressed by the evidence for an independent effect of PM2.5 exposures provided by
experimental animal studies.

In addition to the findings described above, which are most relevant to older adults,
several 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) are observed in multiple epidemiologic studies (U.S. EPA, 2019,

3-42

-------
section 8.2.7.2), while studies of cognitive function provide little support for an association (U.S.
EPA, 2019, section 8.2.5.2). Interpretation of these epidemiologic studies is limited due to the
small number of studies, their lack of control for potential confounding by copollutants, and
uncertainty regarding the critical exposure windows. Biological plausibility is provided for the
ASD findings by a study in mice that found inflammatory and morphologic changes in the
corpus collosum and hippocampus, as well as ventriculomegaly (i.e., enlarged lateral ventricles)
in young mice following prenatal exposure to PM2.5 CAPs.

Taken together, the 2019 ISA concludes that studies indicate long-term PM2.5 exposures
can lead to effects on the brain associated with neurodegeneration (i.e., neuroinflammation and
reductions in brain volume), as well as cognitive effects in older adults (U.S. EPA, 2019, Table
1-2). Animal toxicology studies provide evidence for a range of nervous system effects in adult
animals, including neuroinflammation and oxidative stress, neurodegeneration, and cognitive
effects, and effects on neurodevelopment in young animals. The epidemiologic evidence is more
limited, but studies generally support associations between long-term PM2.5 exposure and
changes in brain morphology, cognitive decrements and dementia. There is also initial, and
limited, evidence for neurodevelopmental effects, particularly ASD. The consistency and
coherence of the evidence supports the 2019 ISA's conclusion that "the collective evidence is
sufficient to conclude that a causal relationship is likely to exist between long-term PM2.5
exposure and nervous system effects" (U.S. EPA, 2019, section 8.2.9).

3,3,1.6 Other Effects

Compared to the health outcomes discussed above, the 2019 ISA concludes that there is
greater uncertainty in the evidence linking PM2.5, or UFP, exposures with other health outcomes,
reflected in conclusions that the evidence is "suggestive of, but not sufficient to infer, a causal
relationship." The sections below summarize the 2019 ISA conclusions for these outcomes for
long-term (section 3.3.1.6.1) and short-term (section 3.3.1.6.2) PM2.5 exposures, while section
3.3.1.6.3 summarizes conclusions for long- and short-term UFP exposures. Section 3.3.1.6.4
summarizes information assessed in the ISA Supplement related to the emerging area of SARS-
CoV-2 infection and COVID-19 death.

3.3.1.6.1 Long-term PM2.5 Exposures

As indicated in Table 3-1 above, the 2019 ISA concludes that the evidence is "suggestive
of, but not sufficient to infer, a causal relationship" between long-term PM2.5 exposures and
metabolic effects and reproductive and developmental effects (reproduction and fertility;
pregnancy and birth outcomes). These conclusions reflect evidence that is "generally supportive
but not entirely consistent or is limited overall" where "[cjhance, confounding, and other biases

3-43

-------
cannot be ruled out" (U.S. EPA, 2019, Preface, p. P-20). The basis for these causality
determinations is summarized briefly below.

Metabolic effects

There were no causality determinations for long-term PM2.5 exposure and metabolic
effects in the 2009 ISA (U.S. EPA, 2009). However, the literature pertaining to the effect of
long-term exposure to PM2.5 and metabolic effects has expanded substantially since the 2009
ISA, and consists of both epidemiologic and experimental evidence (U.S. EPA, 2019, section
7.2). Epidemiologic studies report positive associations between long-term PM2.5 exposure and
diabetes-related mortality. In addition, although results were not consistent across cohorts, there
is some evidence from epidemiologic studies for positive associations with incident diabetes,
metabolic syndrome, and alterations in glucose and insulin homeostasis. Consideration of
copollutant confounding was limited. In animal toxicologic studies, there is some support for a
relationship between long-term PM2.5 exposure and metabolic effects from experimental studies
demonstrating increased blood glucose, insulin resistance, and inflammation and visceral
adiposity but the experimental evidence was not entirely consistent. Based on this evidence, the
2019 ISA concludes that, "[ojverall, the collective evidence is suggestive of, but is not sufficient
to infer, a causal relationship between long-term PM2.5 exposure and metabolic effects" (U.S.
EPA, 2019, p. 7-52).

Reproductive and developmental effects

The 2009 ISA determined that the evidence was "suggestive of a causal relationship" for
the association between long-term PM2.5 exposure and reproductive and developmental
outcomes. The body of literature characterizing these relationships has grown since the 2009
ISA, with much of the evidence focusing on reproduction and fertility or pregnancy and birth
outcomes, though important uncertainties persist (U.S. EPA, 2019, sections 9.1.1, 9.1.2, 9.1.5).

Effects of PM2.5 exposure on sperm have been studied in both epidemiology and
toxicology studies and shows the strongest evidence in epidemiologic studies for impaired sperm
motility and in animal toxicological studies for impaired spermiation. Epidemiologic evidence on
sperm morphology have reported inconsistent results. Evidence for effects of PM2.5 exposure on
female reproduction also comes from both epidemiology and toxicology studies. In the
epidemiologic literature, results on human fertility and fecundity are limited, but the evidence on
in vitro fertilization indicates a modest association of PM2.5 exposures with decreased odds of
becoming pregnant. Studies in rodents have shown ovulation and estrus are affected by PM2.5
exposure. Biological plausibility for outcomes related to male and female fertility and
reproduction comes from laboratory animal studies demonstrating genetic and epigenetic
changes in germ cells with PM2.5 exposure. The 2019 ISA concludes that, "[cjollectively, the

3-44

-------
evidence is suggestive of, but not sufficient to infer, a causal relationship between PM2.5
exposure and male and female reproduction and fertility" (U.S. EPA, 2019, p. 9-43).

With regard to pregnancy and birth outcomes, while the collective evidence for many of
the outcomes examined is not consistent, there are some animal toxicology and epidemiologic
studies that indicate an association between PM2.5 exposures and reduced fetal growth, low birth
weight and preterm birth. Most of the epidemiologic studies do not control for copollutant
confounding and do not identify a specific sensitive window of exposure but results from animal
toxicologic studies provide biological plausibility for these outcomes, as well as support for
multiple sensitive windows for PM2.5 exposure-associated outcomes. There is also epidemiologic
evidence for congenital heart defects of different types, as well as biological plausibility to
support this outcome from the animal toxicology literature. However, evidence for a relationship
between PM2.5 exposure and various pregnancy-related pathologies, including gestational
hypertension, pre-eclampsia and gestational diabetes is inconsistent. Biological plausibility for
effects of PM2.5 exposure and various pregnancy and birth outcomes is provided by studies
showing that PM2.5 exposure in laboratory rodents resulted in impaired implantation and vascular
endothelial dysfunction. Coherence with toxicological studies is provided by epidemiologic
studies in humans reporting associations with epigenetic changes to the placenta and impaired
fetal thyroid function. When taken together, the 2019 ISA concludes that the available evidence,
including uncertainties that evidence, is "suggestive of, but not sufficient to infer, a causal
relationship between exposure to PM2.5 and pregnancy and birth outcomes" (U.S. EPA, 2019, p.
9-44).

3.3.1.6.2 Short-term PM2.5 Exposures

As indicated in Table 3-1 above, the 2019 ISA concludes that the evidence is "suggestive
of, but not sufficient to infer, a causal relationship" between short-term PM2.5 exposures and
metabolic effects and nervous system effects. As for the outcomes related to long-term
exposures, discussed above, these conclusions reflect evidence that is "generally supportive but
not entirely consistent or is limited overall" where "[cjhance, confounding, and other biases
cannot be ruled out" (U.S. EPA, 2019, Preface, p.P-20). The basis for these causality
determinations is summarized briefly below.

Metabolic effects

There were no studies of the effect of short-term PM2.5 exposure and metabolic effects
reviewed in the 2009 ISA (U.S. EPA, 2009). New evidence for a relationship between short-term
PM2.5 exposure and metabolic effects is based on a small number of epidemiologic and animal
toxicological studies reporting effects on glucose and insulin homeostasis and other indicators of
metabolic function such as inflammation in the visceral adipose tissue and liver (U.S. EPA,

3-45

-------
2019, section 7.1). The 2019 ISA concludes that, overall, the collective evidence "is suggestive
of, but not sufficient to infer, a causal relationship between short-term PM2.5 exposure and
metabolic effects" (U.S. EPA, 2019, p. 7-11).

Nervous system effects

The evidence reviewed in the 2009 ISA was characterized as "inadequate to infer" a
causal relationship between short-term PM2.5 exposure and nervous system effects (U.S. EPA,
2009), based on a small number of experimental animal studies. Studies assessed in the 2019
ISA provide additional evidence that short-term exposure to PM2.5 can affect the nervous system
(U.S. EPA, 2019, section 8.1). The strongest evidence is provided by experimental studies in
mice that show effects on the brain. These toxicological studies demonstrate changes in
neurotransmitters in the hypothalamus that are linked to sympathetic nervous system and
hypothalamic-pituitary-adrenal (HPA) stress axis activation, as well as upregulation of
inflammation-related genes, changes in cytokine levels, and other changes that are indicative of
brain inflammation. In addition, an association of short-term PM2.5 exposure with hospital
admissions for Parkinson's disease was observed indicating the potential for exacerbation of
neurological diseases. The 2019 ISA concludes that, overall, the collective evidence "is
suggestive of, but not sufficient to infer, a causal relationship between short-term exposure to
PM2.5 and nervous system effects" (U.S. EPA, 2019, p. 8-15).

3.3.1.6.3 Long- and Short-Term UFP Exposures

Recent studies evaluated in the 2019 PM ISA have further explored the relationship
between short- and long-term UFP exposure and health effects as detailed below. Across both
experimental and epidemiologic studies, the same uncertainties and limitations in the evidence
for each exposure duration and health effect category persists as those identified in the 2009 PM
ISA. Of these uncertainties and limitations two are noted prior to summarizing the health effects
evidence because they are inherent across each health effect category evaluated and directly
inform each causality determination: (1) definition of UFPs, and (2) exposure metric.

With respect to the definition of UFPs, the assessment of study results across
experimental and epidemiologic studies is complicated by how UFPs are defined in each study
and the resulting UFP size distribution examined. EPA has defined UFPs as particles <0.1 |im in
aerodynamic diameter. However, experimental studies, both animal toxicological and controlled
human exposure, typically use a particle concentrator, which can result in exposures to particles
< 0.30 |im (U.S. EPA, 2019, Section P.3.1). In contrast, epidemiologic studies often examine
various size ranges under 0.1 |im. However, there are instances where a direct size range is not
specified depending on the exposure metric used, which can result in exposures to particles
larger than 0.1 |im in some studies (U.S. EPA, 2019, Section P.3.1). An additional complication

3-46

-------
when evaluating UFP studies is the nonuniformity in the exposure metric used with studies using
one or multiple metrics including particle number concentration [NC], surface area concentration
[SC], or mass concentration [MC)] (U.S. EPA, 2019, Section P.3.1) to estimate UFP exposures.
The inconsistency in both the definition of UFPs and the exposure metric used across studies
complicates the interpretation of results. Lastly, a lack of UFP monitoring, specifically within the
U.S., to characterize spatial and temporal concentrations leads to uncertainty in characterizing
UFP exposures and the inability to fully evaluate the potential for copollutant confounding in
epidemiologic studies, which limits inference about an independent effect of UFP exposures on
health (U.S. EPA, 2019, section 5.5)

Cardiovascular effects

In the 2009 ISA, the evidence from toxicological studies, many of which examined
exposures to whole diesel exhaust or wood smoke rather than UFP alone, was suggestive of a
causal relationship between short-term UFP exposure and cardiovascular effects. Since the 2009
ISA, there have been additional studies published describing the relationship between short-term
UFP exposure and cardiovascular effects. This includes a small number of epidemiologic panel
studies that have observed positive associations between short-term exposure to UFPs and
measures of HRV (U.S. EPA, 2019, section 6.5.9.1) and markers of coagulation (U.S. EPA,
2019, section 6.5.11.1) although there are also studies that did not report such UFP-related
effects. In addition, there is evidence from a single controlled human exposure study indicating
decreases in the anticoagulant proteins plasminogen and thrombomodulin in individuals with
metabolic syndrome (U.S. EPA, 2019, section 6.5.11.2). There is inconsistent evidence from
controlled human exposure and epidemiologic panel studies for endothelial dysfunction, changes
in blood pressure, and systemic inflammation following short-term exposure to UFPs. Notably,
there is little evidence of an effect when considering short-term UFP exposure on other
cardiovascular endpoints as well as cardiovascular-disease emergency department visits or
hospital admissions. The assessment of study results across experimental and epidemiologic
studies is complicated by differences in the size distributions examined between disciplines and
by the nonuniformity in the exposure metrics examined (e.g., particle number concentration,
surface area concentration, and mass concentration) (U.S. EPA, 2019, section 1.4.3). When
considered as a whole, the 2019 ISA concludes that the evidence is "suggestive of, but not
sufficient to infer, a causal relationship between short-term exposure UFP exposure and
cardiovascular effects" (U.S. EPA, 2019, p. 6-304).

Respiratory effects

A limited number of studies examining short-term exposure to UFPs and respiratory
effects were reported in the 2009 ISA, which concluded that the relationship between short-term
exposure to UFP and respiratory effects is "suggestive of a causal relationship." This conclusion

3-47

-------
was based on epidemiologic evidence indicating associations with combined respiratory-related
diseases, respiratory infection, and asthma exacerbation. In addition, personal exposures to
ambient UFP were associated with lung function decrements in adults with asthma. The few
available experimental studies provided limited coherence with epidemiologic findings for
asthma exacerbation. Studies assessed in the 2019 ISA add to this evidence base and support
epidemiologic evidence for asthma exacerbation and combined respiratory-related diseases but
do not rule out chance, confounding, and other biases (U.S. EPA, 2019, section 5.5). For
example, associations persist in one epidemiologic study with adjustment for NO2, but not in
another. Additional supporting evidence, showing decrements in lung function and enhancement
of allergic inflammation and other allergic responses, is provided by a controlled human
exposure study in adults with asthma and by animal toxicological studies in an animal model of
allergic airway disease. For combined respiratory-related diseases, recent findings add
consistency for hospital admissions and emergency department visits and indicate lung function
changes among adults with asthma or COPD. Uncertainty remains regarding the characterization
of UFP exposures and the potential for copollutant confounding in epidemiologic studies, which
limits inference about an independent effect of UFP exposures (U.S. EPA, 2019, section 5.5).
The 2019 ISA concludes that, overall, the evidence is "suggestive of, but not sufficient to infer, a
causal relationship between short-term UFP exposure and respiratory effects" (U.S. EPA, 2019,
p. 5-303).

Nervous System Effects and Long-term Exposure

The 2009 ISA reported one animal toxicological study examining long-term exposure to
UFP and nervous system effects, with no supporting epidemiologic studies. Animal toxicological
studies evaluated in the 2019 ISA substantially add to this evidence base. Multiple toxicological
studies of long-term UFP exposure conducted in adult mice provide consistent evidence of brain
inflammation and oxidative stress in the whole brain, hippocampus, and cerebral cortex (U.S.
EPA, 2019, section 8.6.3). Studies also found morphologic changes, specifically
neurodegeneration in specific regions of the hippocampus and pathologic changes characteristic
of Alzheimer's disease, and initial evidence of behavioral effects in adult mice (U.S. EPA, 2019,
sections 8.6.4 and 8.6.5). Toxicological studies examining pre- and post-natal UFP exposures
provide extensive evidence for behavioral effects, altered neurotransmitters, neuroinflammation,
and morphologic changes (U.S. EPA, 2019, section 8.6.6.2). Persistent ventriculomegaly was
observed in male, but not female, mice exposed postnatally to UFP (U.S. EPA, 2019,
section 8.6.6). Epidemiologic evidence is limited to a single study of school children that
provides support for the experimental results. This study, which did not consider copollutant
confounding, reports an association between long-term exposure to UFP, which was measured at
the school, and decrements on tests of attention and memory. However, uncertainties remain as a

3-48

-------
result of inadequate assessment of potential copollutant confounding, the spatial variation in UFP
concentrations, and exposure measurement error. Based primarily on the animal toxicological
evidence of neurotoxicity and altered neurodevelopment, the 2019 ISA concludes that the
evidence is "suggestive of, but not sufficient to infer, a causal relationship" between long-term
UFP exposure and nervous system effects (U.S. EPA, 2019, section 8.6.7).

Nervous system effects and Short-term Exposure

The 2009 ISA reported limited animal toxicological evidence of a relationship between
short-term exposure to UFP and nervous system effects, without supporting epidemiologic
studies. Several experimental studies evaluated in the 2019 ISA add to this evidence base. In the
2019 ISA, the strongest evidence for a relationship between short-term UFP exposure and
nervous system effects is provided by animal toxicological studies that show inflammation and
oxidative stress in multiple brain regions following exposure to UFP. There is a lack of evidence
from epidemiologic studies (U.S. EPA, 2019, section 8.5). The 2019 ISA concludes that, overall,
the collective evidence is "suggestive of, but not sufficient to infer, a causal relationship between
short-term UFP exposure and nervous system effects" (U.S. EPA, 2019, p. 8-86).

3.3.1.6.4 SARS-CoV-2 Infection and COVID-19 Death

With the advent of the global COVID-19 pandemic, a number of recent studies evaluated
in the ISA Supplement examined the role of ambient air pollution, specifically PM2.5, on SARS-
CoV-2 infections and COVID-19 deaths, including a few studies within the U.S. and Canada
(U.S. EPA, 2022; section 3.3.2). While there is no exact corollary within the 2019 ISA for these
types of studies, the 2019 ISA presented evidence that evaluates the potential relationship
between short- and long-term PM2.5 exposure and respiratory infection (U.S. EPA, 2019, section
5.1.5 and 5.2.6). Studies assessed in the 2019 ISA report some evidence of positive associations
between short-term PM2.5 and hospital admissions and emergency department visits for
respiratory infections, however the interpretation of these studies is complicated by the
variability in the type of respiratory infection outcome examined (U.S. EPA, 2019, Figure 5-7).
In the 2019 ISA, studies of long-term PM2.5 exposure were limited and while there were some
positive associations reported, there was minimal overlap in respiratory infection outcomes
examined across studies. Exposure to PM2.5 has been shown to impair host defense, specifically
altering macrophage function, providing a biological pathway by which PM2.5 exposure could
lead to respiratory infection (U.S. EPA, 2019, sections 5.1.1 and 5.1.5.) There is some additional
evidence that PM2.5 exposure can lead to decreases in an individual's immune response, which
can subsequently facilitate replication of respiratory viruses (Bourdrel et al., 2021).

As assessed in the ISA Supplement, a number of studies examined whether daily changes
in PM2.5 can influence SARS-CoV-2 infection and COVID-19 death (U.S. EPA, 2022, section

3-49

-------
3.3.2.1). Additionally, several studies assessed in the ISA Supplement evaluated whether long-
term PM2.5 exposure is related to increased susceptibility to SARS-CoV-2 infection and COVID-
19 death in North America (U.S. EPA, 2022, section 3.3.2.2). While there is initial evidence of
positive associations with, SARS-CoV-2 infection and COVID-19 death, uncertainties remain
due to methodological issues that may influence the results, including: (1) the use of ecological
study design; (2) studies were conducted during the ongoing pandemic when the etiology of
COVID-19 was still not well understood (e.g., specifically, there are important differences in
COVID-19-related outcomes by a variety of factors such as race and socioeconomic status); and
(3) 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) (U.S. EPA, 2022, chapter 5). Taken
together, while there is initial evidence of positive associations with SARS-CoV-2 infection and
COVID-19 death, uncertainties remain due to methodological issues.

3,3,1,7 Summary

Based on the evidence assessed in the 2019 ISA and the ISA Supplement (U.S. EPA,
2019, U.S. EPA, 2021a, U.S. EPA, 2022), and summarized in sections 3.3.1.1 to 3.3.1.6 above,
we revisit the policy-relevant questions posed at the beginning of this section:

• To what extent does the scientific evidence strengthen, or otherwise alter, our

conclusions regarding health effects attributable to long- or short-term fine particle
exposures? Have previously identified uncertainties been reduced? What important
uncertainties remain and have new uncertainties been identified?

We consider these questions in the context of the evidence for effects of long- and short-
term PM2.5 exposures. Studies reviewed in the 2019 ISA and the ISA Supplement expand our
understanding of the PIVh.s-related health effects from long- and short- term exposures, as well as
reduced important uncertainties identified in prior reviews. Epidemiologic studies consistently
report positive associations between PM2.5 exposures and a wide range of health outcomes,
including total and cause-specific mortality (e.g., cardiovascular and respiratory mortality),
cardiovascular and respiratory morbidity, lung cancer, and nervous system effects. Such
associations have been reported in analyses employing a variety of study designs, approaches to
estimating PM2.5 exposures, statistical models, and long-term exposure windows (i.e., the
exposure period that is associated with the health outcome). Recent U.S. and Canadian
epidemiologic studies evaluated in the ISA Supplement provide additional support for the
conclusions of the 2019 ISA. Overall, these studies support, and in some instances strengthen,
the evidence presented in the 2019 ISA of long-term PM2.5 exposures and health effects. Cohort
studies assessed in the ISA Supplement add to the large body of evidence exhibiting consistent,
positive associations between long-term PM2.5 exposure and mortality detailed in the 2019 ISA.
While relatively fewer recent U.S. and Canadian epidemiologic studies examined short-term

3-50

-------
PM2.5 exposure and mortality, these studies continue to provide evidence of positive associations
with all-cause and total (nonaccidental) mortality, in addition to cause-specific mortality
outcomes. Further, the 2019 ISA and ISA Supplement include retrospective studies that
demonstrate improvements in health outcomes, including increased life expectancy, decreasing
mortality, or decreasing respiratory effects, as a result of decreases in ambient PM2.5
concentrations over time. Lastly, the biological plausibility of PIVh.s-attributable mortality is
supported by the coherence of effects across scientific disciplines (i.e., animal toxicological,
controlled human exposure studies, and epidemiologic) when evaluating respiratory and
cardiovascular morbidity effects, which are some of the largest contributors to total
(nonaccidental) mortality.

Epidemiologic studies (for short-term and long-term exposure) evaluated in the 2019 ISA
and the ISA Supplement assessed the role potential uncertainties may have on the health-effect
associations, and examined various exposure windows, approaches to adjust for confounding
variables, and exposure assessment methods that used different sources of data and were
conducted at different spatial resolutions. These evaluations increased confidence in the causal
relationship between long-term PM2.5 exposure and mortality. Moreover, this evidence further
informs whether there is evidence of copollutant confounding, and although there were some
differences across studies, generally positive associations persisted in copollutant models. Some
studies reported that associations persisted in analyses that exclude PM2.5 exposures near the
upper end of the air quality distribution. Overall, the assessment of the C-R relationship
continues to generally support a linear, no-threshold relationship with some recent studies
providing evidence for either a sublinear, linear, or supralinear relationship at these lower
concentrations.

Building on the evidence presented in the 2019 ISA, the evidence assessed in the ISA
Supplement provides additional information to address key uncertainties associated with the
health effects evidence. The ISA Supplement examined an expanded body of evidence related to
alternative methods for confounder control, to further evaluate the causal nature of associations
between exposure to PM2.5 and mortality. Consistent with the 2019 ISA, this expanded body of
evidence reduces uncertainties related to confounding and provides robust support for positive
and significant associations seen in cohort studies of long-term exposure to PM2.5. Although
there were fewer more recent multicity studies conducted in the U.S. and Canada examining the
relationship between short-term exposure and mortality than for long-term exposure, the studies
assessed in the ISA Supplement add to the extensive evidence evaluated in the 2019 ISA.
Furthermore, these studies report consistent positive associations across studies that are using
different exposure assessment methods, statistical models, as well as different methods to control
for confounding effects.

3-51

-------
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 evidence
evaluated in the 2019 ISA. Studies examining short-term PM2.5 exposure report consistent
positive associations for cardiovascular-related emergency department visits and hospital
admissions, specifically for ischemic heart disease, myocardial infarction, and heart failure. In
studies evaluating long-term exposures there remains strong evidence for cardiovascular-related
mortality with support from studies of cardiovascular morbidity outcomes, including coronary
heart disease, stroke, and atherosclerosis progression, among individuals with preexisting
diseases or patients followed after a cardiac event or procedure. In addition, the studies provide
evidence of an immediate effect of short-term-related PM2.5 exposure on cardiovascular-related
outcomes, especially during the first few days following exposure.

With respect to long-term PM2.5 exposure, the strongest evidence associated with
cardiovascular mortality is exhibited in studies that report positive associations with ischemic
heart disease and stroke mortality. Furthermore, recent studies examining association between
long-term PM2.5 exposure and cardiovascular morbidity, specifically coronary heart disease,
stroke, and atherosclerosis progression, most consistently report positive associations when
focusing on individuals with pre-existing diseases and patients followed after a cardiac event or
procedure, and not the general population as a whole, supporting and extending the evidence
presented in 2019 ISA. The 2019 ISA also assessed controlled human exposure studies that were
conducted in Europe at near-ambient PM2.5 concentrations and provide initial evidence of
vascular changes and heart rate variability as well as changes in cardiac and lung function as well
as inflammation.

The ISA Supplement also evaluates epidemiologic studies that examine the relationship
between PM2.5 exposure and SARS-CoV-2 infection and COVID-19 death. While these studies
report positive associations, there a number of methodological limitations which include: (1)
employing an ecological study design, (2) conducting research while COVID-19 etiology was
poorly understood, and (3) the lack of accounting for key factors in disease transmission such as
use of mask, stay home orders, and testing capacity.

Thus, when taken together, the evidence available in the ISA Supplement reaffirms, and
in some cases strengthens, the conclusions from the 2019 ISA regarding long- and short-term
PM2.5 exposures and mortality and cardiovascular effects.

3.3.2 Public Health Implications and At-Risk Populations

The public health implications of the evidence regarding PM2.5 health effects, as for other
effects, are dependent on the type and severity of the effects, as well as the size of the population
affected. Such factors are discussed here in the context of our consideration of the health effects

3-52

-------
evidence related to PM2.5 in ambient air. Additionally, we summarize the information on
population groups at risk of the effects of PM2.5 in ambient air.

• Does the evidence alter our understanding of populations that are particularly at
risk from PM2.5 exposures? What are important uncertainties in that evidence?

The information available in this reconsideration has not altered our understanding of
human populations at risk of health effects from PM2.5 exposures. As recognized in the 2020
review, the 2019 ISA cites extensive evidence indicating that "both the general population as
well as specific populations and lifestages are at risk for PM2.5-related health effects" (U.S. EPA,
2019, p. 12-1). Factors that may contribute to increased risk of PIVh.s-related health effects
include lifestage (children and older adults), pre-existing diseases (cardiovascular disease and
respiratory disease), race/ethnicity, and socioeconomic status.15

Children make up a substantial fraction of the U.S. population and often have unique
factors that contribute to their risk of experiencing a health effect due to exposures to ambient air
pollutants because of their continuous growth and development.16 Children may be particularly
at risk for health effects related to ambient PM2.5 exposures compared with adults because they
have (1) a developing respiratory system, (2) increased ventilation rates relative to body mass
compared with adults, and (3) an increased proportion of oral breathing, particularly in boys,
relative to adults (U.S. EPA, 2019, section 12.5.1.1). There is strong evidence that demonstrates
PM2.5 associated health effects in children, particularly from epidemiologic studies of long-term
PM2.5 exposure and impaired lung function growth, decrements in lung function, and asthma
development. However, there is limited evidence from stratified analyses that children are at
increased risk of PM2.5-related health effects compared to adults. Additionally, there is some
evidence that indicates that children receive higher PM2.5 exposures than adults, and dosimetric
differences in children compared to adults can contribute to higher doses (U.S. EPA, 2019,
section 12.5.1.1).

In the U.S., older adults, often defined as adults 65 years of age and older, represent an
increasing portion of the population and often have pre-existing diseases or conditions that may
compromise biological function. While there is limited evidence to indicate that older adults
have higher exposures than younger adults, older adults may receive higher doses of PM2.5 due to
dosimetric differences. There is consistent evidence from studies of older adults demonstrating
generally consistent, positive associations in studies examining health effects from short- and
long-term PM2.5 exposure and cardiovascular or respiratory hospital admissions, emergency

15 As described in the 2019 ISA, other factors that have the potential to contribute to increased risk include obesity,
diabetes, genetic factors, smoking status, sex, diet, and residential location (U.S. EPA, 2019, chapter 12).

16 Children, as used throughout this PA, generally refers to those younger than 18 years old.

3-53

-------
department visits, or mortality (U.S. EPA, 2019, sections 6.1, 6.2, 11.1, 11.2, 12.5.1.2).
Additionally, several animal toxicological, controlled human exposure, and epidemiologic
studies did not stratify results by lifestage, but instead focused the analyses on older individuals,
and can provide coherence and biological plausibility for the occurrence among this lifestage
(U.S. EPA, 2019, section 12.5.1.2).

Individuals with pre-existing disease may be considered at greater risk of an air pollution-
related health effect than those without disease because they are likely in a compromised
biological state that can vary depending on the disease and severity. With regard to
cardiovascular disease, we first note that cardiovascular disease is the leading cause of death in
the U.S., accounting for one in four deaths, and approximately 12% of the adult population in the
U.S. has a cardiovascular disease (U.S. EPA, 2019, section 12.3.1). Strong evidence
demonstrates that there is a causal relationship between cardiovascular effects and long- and
short-term exposures to PM2.5. Some of the evidence supporting this conclusion is from studies
of panels or cohorts with pre-existing cardiovascular disease, which provide supporting evidence
but do not directly demonstrate an increase in risk (U.S. EPA, 2019, section 12.3.1).
Epidemiologic evidence indicates that individuals with pre-existing cardiovascular disease may
be at increased risk for PM2.5-associated health effects compared to those without pre-existing
cardiovascular disease. While the evidence does not consistently support increased risk for all
pre-existing cardiovascular diseases, there is evidence that certain pre-existing cardiovascular
diseases (e.g., hypertension) may be a factor that increases PIVh.s-related risk. Furthermore, there
is strong evidence supporting a causal relationship for long- and short-term PM2.5 exposure and
cardiovascular effects, particularly for IHD (U.S. EPA, 2019, chapter 6, section 12.3.1).

With regard to respiratory disease, we first note that the most chronic respiratory diseases
in the U.S. are asthma and COPD. Asthma affects a substantial fraction of the U.S. population
and is the leading chronic disease among children. COPD primarily affects older adults and
contributes to compromised respiratory function and underlying pulmonary inflammation. The
body of evidence indicates that individuals with pre-existing respiratory diseases, particularly
asthma and COPD, may be at increased risk for PM2.5-related health effects compared to those
without pre-existing respiratory diseases (U.S. EPA, 2019, section 12.3.5). There is strong
evidence indicating PIVh.s-associated respiratory effects among those with asthma, which forms
the primary evidence base for the likely to be causal relationship between short-term exposures
to PM2.5 and respiratory health effects (U.S. EPA, 2019, section 12.3.5). For asthma,
epidemiologic evidence demonstrates associations between short-term PM2.5 exposures and
respiratory effects, particularly evidence for asthma exacerbation, and controlled human
exposure and animal toxicological studies demonstrate biological plausibility for asthma
exacerbation with PM2.5 exposures (U.S. EPA, 2019, section 12.3.5.1). For COPD,

3-54

-------
epidemiologic studies report positive associations between short-term PM2.5 exposures and
hospital admissions and emergency department visits for COPD, with supporting evidence from
panel studies demonstration COPD exacerbation. Epidemiologic evidence is supported by some
experimental evidence of COPD-related effects, which provides support for the biological
plausibility for COPD in response to PM2.5 exposures (U.S. EPA, 2019, section 12.3.5.2).

There is strong evidence for racial and ethnic disparities in PM2.5 exposures and PM2.5-
related health risk, as assessed in the 2019 ISA and with even more evidence available since the
literature cutoff date for the 2019 ISA and evaluated in the ISA Supplement. There is strong
evidence demonstrating that Black and Hispanic populations, in particular, have higher PM2.5
exposures than non-Hispanic White populations (U.S. EPA, 2019, Figure 12-2; U.S. EPA, 2022,
Figure 3-38). Black populations or individuals that live in predominantly Black neighborhoods
experience higher PM2.5 exposures, in comparison to non-Hispanic White populations. There is
also consistent evidence across multiple studies that demonstrate increased risk of PIVh.s-related
health effects, with the strongest evidence for health risk disparities for mortality (U.S. EPA,
2019, section 12.5.4). There is also evidence of health risk disparities for both Hispanic and non-
Hispanic Black populations compared to non-Hispanic White populations for cause-specific
mortality and incident hypertension (U.S. EPA, 2022, section 3.3.3.2).

Socioeconomic status (SES) is a composite measure that includes metrics such as
income, occupation, or education, and can play a role in access to healthy environments as well
as access to healthcare. SES may be a factor that contributes to differential risk from PM2.5-
related health effects. Studies assessed in the 2019 ISA and ISA Supplement provide evidence
that lower SES communities are exposed to higher concentrations of PM2.5 compared to higher
SES communities (U.S. EPA, 2019, section 12.5.3; U.S. EPA, 2022, section 3.3.3.1.1). Studies
using composite measures of neighborhood SES consistently demonstrated a disparity in both
PM2.5 exposure and the risk of PIVh.s-related health outcomes. There is some evidence that
supports associations larger in magnitude between mortality and long-term PM2.5 exposures for
those with low income or living in lower income areas compared to those with higher income or
living in higher income neighborhoods (U.S. EPA, 2019, section 12.5.3; U.S. EPA, 2022, section
3.3.3.1.1). Additionally, evidence supports conclusions that lower SES is associated with cause-
specific mortality and certain health endpoints (i.e., HI and CHF), but less so for all-cause or
total (non-accidental) mortality (U.S. EPA, 2022, section 3.3.3.1).

• What does the available information indicate with regard to the size of at-risk
populations and their distribution in the U.S.?

The magnitude and characterization of a public health impact is dependent upon the size
and characteristics of the populations affected, as well as the type or severity of the effects. As
summarized above, lifestage (children and older adults), race/ethnicity and socioeconomic status

3-55

-------
are factors that increase the risk of PM2.5-related health effects. The American Community
Survey (ACS) for 2019 estimates that approximately 22% and 16% of the U.S. population are
children (age <18) and older adults (age 65+), respectively. For all ages, non-Hispanic Black and
Hispanic populations are approximately 12% and 18% of the overall U.S. population in 2019.
Table 3-2 below considers the currently available information that helps to characterize key
features of these populations.

Table 3-2. National demographic information, 2019.

Characteristic1

Number

Percent of Total

Total

328,239,523

Child (Age <18)

72,967,785

22.2

Adult (Age 18+)

255,271,738

77.8

All Age Groups

0-4 years

19,404,835

5.9

5-14 years

41,113,916

12.5

15-19 years

21,353,524

6.5

20-24 years

21,468,680

6.5

25-34 years

45,578,475

13.9

35-64 years

125,246,065

38.1

65+ years

54,074,028

16.4

Race/Ethnicity

328.239.523

White NH2

196,789,401

Black NH

40,596,040

12.4

American Indian or Alaska Native NH

2,236,348

0.7

Asian NH

18,427,914

5.6

Hispanic, all

60,481,746

18.4

Other NH

9,708,074

Household Income (past 12 months)3

Less than $10,000

5.8

$10,000 to $14,999

4.0

$15,000 to $24,999

8.3

$25,000 to $34,999

8.4

$35,000 to $49,999

11.9

$50,000 to $74,999

17.4

$75,000 to $99,999

12.8

$100,000 to $149,999

15.7

$150,000 to $199,999

7.2

$200,000 or more

8.5

Educational Attainment4

Less than high school

25,618,541

11.4

High school graduate (or equivalent)

60,482,353

26.9

Some college, no degree

44,914,086

3-56

-------
Associate's degree

19,381,937

8.6

Bachelor's degree

45,730,479

20.3

Graduate or professional degree

28,771,172

12.8

1 Numbers within selected characteristics may not sum to total due to rounding

2 NH = non-Hispanic

3 Household income in the past 12 months in 2019 inflation-adjusted dollars.

4 Educational attainment for population aged 25 years and older.

Adapted from the 2019 American Community Survey and Housing Survey. Available at:

DemoaraDhics: httDs://data.census.aov/cedsci/table?a=United%20States&tid=ACSDP1Y2019.DP05

Income: httos://data.census.aov/cedsci/table?a=United%20States&t=lncome%20and%20Povertv&tid=ACSST1Y2019.S1901

Education:

httos://data.census.aov/cedsci/table?a=United%20States&t=Education%3AEducational%20Attainment&tid=ACSST1Y2019.S1501

As noted above, individuals with pre-existing cardiovascular disease and pre-existing
respiratory disease may also be at increased risk of PIVh.s-related health effects. Table 3-3 below
considers the currently available information that helps to characterize key features of
populations with cardiovascular or respiratory diseases or conditions. The National Center for
Health Statistics data for 2018 indicate that, for adult populations, older adults (e.g., those 65
years and older) have a higher prevalence of cardiovascular diseases compared to younger adults
(e.g., those 64 years and younger). For respiratory diseases, older adults also have a higher
prevalence of emphysema than younger adults, and adults 44 years or older have a higher
prevalence of chronic bronchitis. However, the prevalence for asthma is generally similar across
all adult age groups.

With respect to race, American Indians or Alaskan Natives have the highest prevalence of
all heart disease and coronary heart disease, while Blacks have the highest prevalence of
hypertension and stroke. Hypertension has the highest prevalence across all racial groups
compared to other cardiovascular diseases or conditions, ranging from approximately 22% to
32% of each racial group. Overall, the prevalence of cardiovascular diseases or conditions is
lowest for Asians compared to Whites, Blacks, and American Indians or Alaskan Natives.

Asthma prevalence is highest among Black and American Indian or Alaska Native populations,
while prevalence is generally similar across racial groups for chronic bronchitis and emphysema.
Overall, the prevalence for respiratory diseases is lowest for Asians compared to Whites, Blacks,
and American Indians or Alaskan Natives. With regard to ethnicity, cardiovascular and
respiratory disease prevalence across all diseases or conditions is generally similar between
Hispanic and non-Hispanic populations, although non-Hispanics have a slightly higher
prevalence compared to Hispanics.

3-57

-------
Table 3-3. Prevalence of cardiovascular and respiratory diseases among adults by age, race, and ethnicity in the U.S. in 2018.

Adults (18+)

Age (%)1

Race (%)2

Ethnicity (%)3

Chronic
Disease or
Condition

N (in
thousands)

18-44

44-64

65-74

75+

White

Black

American
Indian or
Alaska
Native

Asian

Hispanic

Non-
Hispanic

All (N, in
thousands)

249,456

115,008

83,038

30,809

20,601

193,454

30,813

2,810

15,960

40,749

208,706

Selected Cardiovascular Diseases/Conditions

All heart disease

30,252

4.8

11.8

23.6

37.3

11.5

10.0

14.6

7.7

8.2

11.7

Coronary heart
disease

15,780

1.0

6.0

15.5

23.9

5.7

5.4

8.6

4.4

5.1

5.7

Hypertension

67,856

00
CO

34.4

54.4

61.1

23.9

32.2

27.2

21.9

23.7

25.1

Stroke

7,801

0.6

3.1

6.9

11.8

2.6

3.9

3.0

2.7

2.5

2.9

Selected Respiratory Diseases

Asthma4

19,233

7.2

8.3

8.6

6.7

7.5

9.1

9.5

3.7

6.0

8.1

COPD - chronic
bronchitis

9,003

2.2

4.5

5.1

5.6

3.6

3.4

1.1

2.7

3.6

COPD-
emphysema

3,780

0.2

1.6

4.1

4.5

1.4

1.1

0.4

0.7

1.0

1.4

1 Percentage of individual adults within each age group with disease, based on N (at the top of each age column).

2 Percentage of individual adults within each race group with disease, based on N (at the top of each race column).

3 Percentage of individual adults within each ethnic group with disease, based on N (at the top of each ethnic column).

4 Asthma prevalence is reported for "still has asthma."

* Estimate does not meet NCHS standards of reliability.

Source: (Insert cites); National Center for Health Statistics, Summary Health Statistics, National Health Interview Survey, 2018; Tables A-1 and A-2.

3-58

-------
Taken together, this information indicates that the groups at increased risk of PM2.5-
related health effects represent a substantial portion of the total U.S. population. In evaluating the
primary PM2.5 standards, an important consideration is the potential PM2.5-related public health
impacts in these populations.

3.3.3 PM2.5 Concentrations in Key Studies Reporting Health Effects

To inform conclusions on the adequacy of the public health protection provided by the
current primary PM2.5 standards, this section evaluates the PM2.5 exposures and ambient
concentrations (i.e., used as surrogates for exposures in epidemiologic studies) in studies
reporting PM2.5-related health effects. We specifically consider the following overarching
questions:

• What are the short- or long-term PM2.5 exposures that have been associated with
health effects and to what extent does the evidence support the occurrence of such
effects for air quality meeting the current primary PM2.5 standards?

In addressing these questions, we emphasize health outcomes for which the 2019 ISA concludes
that the evidence supports a "causal" or a "likely to be causal" relationship with PM2.5 exposures.
As discussed above, this includes mortality, cardiovascular effects, and respiratory effects
associated with short- or long-term PM2.5 exposures and cancer and nervous system effects
associated with long-term PM2.5 exposures. While the causality determinations in the 2019 ISA
are informed by studies evaluating a wide range of PM2.5 concentrations, this section considers
the degree to which the evidence in the 2019 ISA and ISA Supplement supports the occurrence
of PM-related effects at concentrations relevant to informing conclusions on the primary PM2.5
standards. Section 3.3.3.1 considers the exposure concentrations that have been evaluated in
experimental studies and section 3.3.3.2 considers the ambient concentrations in locations
evaluated by epidemiologic studies.

3,3.3.1 PM Exposure Concentrations Evaluated in Experimental Studies

As stated in the 2019 ISA, the evidence for a particular PIVh.s-related health outcome is
strengthened when results from experimental studies demonstrate biologically plausible
mechanisms through which adverse human health outcomes could occur (U.S. EPA, 2015a,
Preamble p. 20). Two types of experimental studies are of particular importance in understanding
the effects of PM exposures: controlled human exposure and animal toxicology studies. In such
studies, investigators expose human volunteers or laboratory animals to known concentrations of
air pollutants under carefully regulated environmental conditions and activity levels. Thus,
controlled human exposure and animal toxicology studies can provide information on the health
effects of experimentally administered pollutant exposures under highly controlled laboratory
conditions (U.S. EPA, 2015a, Preamble, p. 11).

3-59

-------
In this section, we consider the PM2.5 exposure concentrations shown to elicit effects in
controlled human exposure studies and in animal toxicology studies. In particular, we consider
the consistency of specific PIVh.s-related effects across studies, the potential adversity of such
effects, and the degree to which exposures shown to cause effects are likely to occur in areas
meeting the current primary standards. To address these issues, we consider the following
question:

• To what extent does the evidence from controlled human exposure or animal

toxicology studies support the potential for adverse cardiovascular, respiratory, or
other effects following PM2.5 exposures likely to occur in areas meeting the current
or alternative primary standards?

Controlled Human Exposure Studies

As discussed in detail in the 2019 ISA (U.S. EPA, 2019, section 6.1), controlled human
exposure studies have reported that PM2.5 exposures lasting from less than one hour up to five
hours can impact cardiovascular function.17 The most consistent evidence from these studies is
for impaired vascular function (U.S. EPA, 2019, section 6.1.13.2). In addition, although less
consistent, the 2019 ISA notes that studies examining PM2.5 exposures also provide evidence for
increased blood pressure (U.S. EPA, 2019, section 6.1.6.3), conduction abnormalities/arrhythmia
(U.S. EPA, 2019, section 6.1.4.3), changes in heart rate variability (U.S. EPA, 2019, section
6.1.10.2), changes in hemostasis that could promote clot formation (U.S. EPA, 2019, section
6.1.12.2), and increases in inflammatory cells and markers (U.S. EPA, 2019, section 6.1.11.2).
The 2019 ISA concludes that, when taken as a whole, controlled human exposure studies
demonstrate that short-term exposure to PM2.5 may impact cardiovascular function in ways that
could lead to more serious outcomes (U.S. EPA, 2019, section 6.1.16). Thus, such studies can
provide insight into the potential for specific PM2.5 exposures to result in physiological changes
that could increase the risk of more serious effects.

Table 3-4 below summarizes information from the 2019 ISA and ISA Supplement on
available controlled human exposure studies that evaluate effects on markers of cardiovascular
function following exposures to PM2.5, either as concentrated ambient particles (CAP) or in
unfiltered versus filtered air.18

17 In contrast, controlled human exposure studies provide little evidence for respiratory effects following short-term
PM2.5 exposures (U.S. EPA, 2019, section 5.1, Table 5-18). Therefore, this section focuses on cardiovascular
effects evaluated in controlled human exposure studies of PM2 5 exposure.

18 Table 3-4 identifies controlled human exposure studies included in the 2019 ISA and ISA Supplement that
examine the potential for PM2 5 exposures to alter markers of cardiovascular function and is ordered by exposure
concentration. Studies that focus on specific components of PM25 (e.g., endotoxin), or studies that evaluated
PM2 5 exposures only in the presence of an intervention (e.g., dietary intervention) or other pollutant (e.g., ozone),
are not included.

3-60

-------
Table 3-4. Summary of information from PM2.5 controlled human exposure studies.

Study

Population

Exposure Details
(average concentration;
duration)

Results

Brauner et al.,
2008

Healthy adults

10.5 |jg/m3 PM2.5
(unfiltered) vs below
detection (filtered); 24 h

No significant effect on markers of vascular
function

Hemmingsen et
al., 2015a,
Hemmingsen et
al., 2015b

Healthy,
overweight
older adults

24 |jg/m3 (unfiltered) vs
3.0 |jg/m3 (filtered)
Copenhagen PM; 5 h

Impaired vascular function and altered heart rate
variability; no significant changes in blood
pressure or markers of inflammation or oxidative
stress

Wyattet al.,
2020a *

Healthy young
adults (18-35)

37.8 pg/m3 CAP vs2.1
|jg/m3 (filtered); 4h

Increased blood inflammatory markers;
Inconsistent changes in HRV

Urch et al., 2010

Non-asthmatic
and mild
asthmatic
adults

64 pg/m3 CAP (lower
exposure); 2 h

No significant change in blood markers of
inflammation or oxidative stress

Huang et al., 2012

Healthy adults

90 pg/m3 CAP; 2 h

No significant changes in heart rate variability

Devlin et al., 2003

Healthy older
adults

99 pg/m3 CAP1; 2 h

Decreased heart rate variability

Hazuchaet al.,
2013

Adult current
and former
smokers

109 pg/m3 CAP; 2 h

No significant changes in markers of
inflammation or coagulation

Ghio et al., 2000

Healthy young
adults

120 pg/m3 CAP; 2 h

Increased fibrinogen (coagulation)

Ghio et al., 2003

Healthy young
adults

120 pg/m3 CAP; 2 h

Increased fibrinogen; no significant effect on
markers of inflammation

Urch et al., 2010

Non-asthmatic
and mild
asthmatic
adults

140 pg/m3 CAP (higher
exposure); 2 h

Increased blood inflammatory markers

Brook et al., 2009

Healthy adults

149 pg/m3 CAP;2h

Impaired vascular function, increased blood
pressure; no significant change in markers of
inflammation (compared to filtered air)

Ramanathan et
al., 2016

Healthy adults

149 pg/m3 CAP;2h

Decreased antioxidant/anti-inflammatory
capacity when baseline capacity was low

Sivagangabalan et
al., 2011

Healthy adults

150 pg/m3 CAP; 2 h

Increase in indicator of possible arrhythmia; no
significant effect on heart rate

3-61

-------
Kushaet al., 2012

Healthy adults

154 |jg/m3 CAP;2h

No significant effect on indicator of possible
arrhythmia

Gong et al., 2003

Adults with and
without asthma

174 |jg/m3 CAP;2h

Increased heart rate; No significant effect on
indicators of arrhythmia, inflammation,
coagulation; inconsistent effects on blood
pressure

Gong et al., 2004

Older adults
with and
without COPD

200 |jg/m3 CAP; 2 h

Decreased heart rate variability, increase in
markers of inflammation (without COPD only);
inconsistent effect on arrhythmia; no significant
effect on markers of blood coagulation

Liu et al., 2015

Healthy adults

238|jg/m3 CAP; 130 min

Increase in urinary markers of oxidative stress
and vascular dysfunction; no significant effect on
blood markers of oxidative stress, vascular
function, or inflammation

Bellaviaet al.,
2013

Healthy adults

-242 pg/m3 CAP; 130 min

Increased blood pressure

Behbod et al.,
2013

Healthy adults

-250 pg/m3 CAP; 130 min

Increase in markers of inflammation

Tong et al., 2015

Healthy older
adults

253 pg/m3 CAP; 2 h

Impaired vascular function and increased blood
pressure; no significant change in markers of
inflammation or coagulation

Lucking et al.,
2011

Healthy young
men

320 pg/m3 (unfiltered) vs
7.2 pg/m3 (filtered) diesel
exhaust; 1 h

Impaired vascular function and increased
potential for coagulation; no significant effect on
blood pressure, markers of inflammation, or
arterial stiffness

Vieira et al.,
2016a, Vieira et
al., 2016b

Healthy adults;
Heart failure
patients

325 pg/m3 (unfiltered) vs
25 pg/m3 (filtered) diesel
exhaust; 21-min

Increase in marker of potential impairment in
heart function, impaired vascular function (heart
failure patients); no significant effect on blood
pressure, heart rate or heart rate variability,
markers of inflammation, markers of coagulation,
or arterial stiffness

* Study newly assessed in the ISA Supplement

1 The published study reports an average CAP concentration of 41 pg/m3, but communication with the study authors revealed
an error in that reported concentration (Jenkins, 2016).

Most of the controlled human exposure studies in Table 3-4 exposed participants to
average PM2.5 concentrations at or above about 100 |ig/m3, with exposure durations typically up
to about two hours. Statistically significant effects on one or more indicators of cardiovascular
function are often, though not always, reported following 2-hour exposures to average PM2.5
concentrations at and above about 120 |ig/m3, with less consistent evidence for effects following
exposures to concentrations lower than 120 |ig/m3. Impaired vascular function, the effect

3-62

-------
identified in the 2019 ISA as the most consistent across studies (U.S. EPA, 2019, section
6.1.13.2), is shown following 2-hour exposures to PM2.5 concentrations at and above 149 |ig/m3.
Mixed results are reported in the three studies that evaluated longer exposure durations (i.e.,
longer than 2 hours) and lower (i.e., near-ambient) PM2.5 concentrations, with significant effects
for some outcomes reported following 5-hour exposures to 24 |ig/m3 in Hemmingsen et al.
(2015b), but not for other outcomes following 5-hour exposures to 24 |ig/m3 in Hemmingsen et
al. (2015a) and not following 24-hour exposures to 10.5 |ig/m3 in Brauner et al. (2008). Wyatt 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 and found significant effects for some cardiovascular (e.g., systematic inflammation
markers, cardiac repolarization, and decreased pulmonary function) and respiratory effects
following 4-hour exposures to 37.8 |ig/m3. The higher ventilation rate and longer exposure
duration in this study compared to most controlled human exposure studies is roughly equivalent
to a 2-hour exposure of 75-100 |ig/m3 of PM2.5. Therefore, dosimetric consideration may explain
the observed changes in lung function and inflammation in young healthy individuals. While this
study provides evidence of some effects at lower PM2.5 concentrations, overall there is
inconsistent evidence for changes in lung function and inflammation in other controlled human
exposure studies evaluated in the 2019 ISA (U.S. EPA, 2019, sections 5.1.7., 5.1.2.3.3, and
6.1.11.2.1; U.S. EPA, 2022, section 3.3.1).

Taken together, these controlled human exposure studies support biological plausibility
for the serious cardiovascular and respiratory effects that have been linked with ambient PM2.5
exposures and seen in epidemiologic studies (U.S. EPA, 2019, Chapter 6). However, while these
studies are important in establishing biological plausibility, it is unclear how the results alone
and the importance of the effects observed in these studies, particularly in studies conducted at
near-ambient PM2.5 concentrations, should be interpreted with respect to adversity to public
health. For example, impaired vascular function, the effect identified as most consistent across
studies (U.S. EPA, 2019, section 6.1.13.2), can signal an intermediate effect along the potential
biological pathways for cardiovascular effects following short-term exposure to PM2.5 and show
a role for exposure to PM2.5 leading to potential worsening of IHD and heart failure followed
potentially by ED visits, hospital admissions, or mortality (U.S. EPA, 2019, section 6.1 and
Figure 6-1). However, just observing the occurrence of impaired vascular function alone does
not clearly suggest an adverse health outcome. Additionally, associated judgments regarding
adversity or health significance of measurable physiological responses to air pollutants have been
informed by guidance, criteria or interpretative statements developed within the public health
community, including the American Thoracic Society (ATS) and the European Respiratory

3-63

-------
Society (ERS), which cooperatively updated the ATS 2000 statement What Constitutes an
Adverse Health Effect of Air Pollution (ATS, 2000) with new scientific findings, including the
evidence related to air pollution and the cardiovascular system (Thurston et al., 2017).19 With
regard to vascular function, the ATS/ERS statement considers the adversity of both chronic and
acute reductions in endothelial function. While the ATS/ERS statement concluded that chronic
endothelial and vascular dysfunction can be judged to be a biomarker of an adverse health effect
from air pollution, they also conclude that "The health relevance of acute reductions in
endothelial function induced by air pollution is less certain" (Thurston et al., 2017). This is
particularly informative to our consideration of the controlled human exposure studies which are
short-term in nature (i.e., ranging from 2- to 5-hours), including those studies that are conducted
at near-ambient PM2.5 concentrations.

It is also important to consider that controlled human exposure studies include a small
number of individuals compared to epidemiologic studies. Additionally, these studies tend to
include generally healthy adult individuals, who are at a lower risk of experiencing health
effects. These studies, therefore, often do not include including children, or older adults, or
individuals with pre-existing conditions. As such, these studies are somewhat limited in their
ability to inform at what concentrations effects may be elicited in at-risk populations.

Nonetheless, we note the findings in several of these controlled human exposure studies
conducted at near-ambient PM2.5 concentrations and the potential of these studies to provide
some insight into what these controlled human exposure studies may indicate regarding short-
term exposure to peak PM2.5 concentrations and how those relate to ambient PM2.5
concentrations in areas that meet the primary PM2.5 standards. As such, we focus on 2-hour
exposures (the exposure window most often utilized) and consider the degree to which 2-hour
ambient PM2.5 concentrations in locations meeting the current primary standards are likely to
exceed the 2-hour exposure concentrations at which statistically significant effects are reported
in multiple studies for one or more indicators of cardiovascular function. To this end, we refer to
Figure 2-19 (Chapter 2, section 2.3.2.2.3), which presents the frequency distribution of 2-hour
average PM2.5 concentrations from all FEM PM2.5 monitors in the U.S. for 2017-2019. At sites

19 The ATS/ERS described its 2017 statement as one "intended to provide guidance to policymakers, clinicians and
public health professionals, as well as others who interpret the scientific evidence on the health effects of air
pollution for risk management purposes" and further notes that "considerations as to what constitutes an adverse
health effect, in order to provide guidance to researchers and policymakers when new health effects markers or
health outcome associations might be reported in future." The most recent policy statement by the ATS, which
once again broadens its discussion of effects, responses and biomarkers to reflect the expansion of scientific
research in these areas, reiterates that concept, conveying that it does not offer "strict rules or numerical criteria,
but rather proposes considerations to be weighed in setting boundaries between adverse and nonadverse health
effects," providing a general framework for interpreting evidence that proposes a "set of considerations that can
be applied in forming judgments" for this context (Thurston et al., 2017).

3-64

-------
meeting the current primary PM2.5 standards, most 2-hour concentrations are below 10 ng/m3,
and almost never exceed 30 |ag/m3. The extreme upper end of the distribution of 2-hour PM2.5
concentrations is shifted higher during the warmer months (April to September, denoted by red
bars in Figure 2-19), generally corresponding to the period of peak wildfire frequency in the U.S.
At sites meeting the current primary standards, the highest 2-hour concentrations measured
almost never occur outside of the period of peak wildfire frequency (i.e., 99.9th percentile of 2-
hour concentrations is 62 ng/m3 during the warm season). Most of the sites measuring these very
high concentrations are in the northwestern U.S. and California (see Appendix A, Figure A-l),
where wildfires have been relatively common in recent years. When the typical fire season is
excluded from the analysis (blue in Figure 2-19), the extreme upper end of the distribution is
reduced (i.e., 99.9th percentile of 2-hour concentrations is 55 ng/m3).20 Given these results, we
conclude that PM2.5 exposure concentrations evaluated in most of these controlled human
exposure studies are well-above the 2-hour ambient PM2.5 concentrations typically measured in
locations meeting the current primary standards.

Animal Toxicology Studies

The 2019 ISA relies on animal toxicology studies to support the plausibility of a wide
range of PM2.5-related health effects. While animal toxicology studies often examine more
severe health outcomes and longer exposure durations than controlled human exposure studies,
there is uncertainty in extrapolating the effects seen in animals, and the PM2.5 exposures and
doses that cause those effects, to human populations. We consider these uncertainties when
evaluating what the available animal toxicology studies may indicate with regard to the current
primary PM2.5 standards.

Most of the animal toxicology studies assessed in the 2019 ISA have generally examined
short-term exposures to PM2.5 concentrations from 100 to >1,000 |j,g/m3 and long-term exposures
to concentrations from 66 to >400 |j,g/m3 (e.g., see U.S. EPA, 2019, Table 1-2). Two exceptions
are a study reporting impaired lung development following long-term exposures (i.e., 24 hours
per day for several months prenatally and postnatally) to an average PM2.5 concentration of 16.8
Hg/m3 (Mauad et al., 2008) and a study reporting increased carcinogenic potential following
long-term exposures (i.e., 2 months) to an average PM2.5 concentration of 17.7 |j,g/m3 (Cangerana
Pereira et al., 2011). These two studies demonstrate serious effects following long-term
exposures to PM2.5 concentrations similar to the ambient concentrations reported in some PM2.5
epidemiologic studies (U.S. EPA, 2019, Table 1-2), though still above the ambient

20 Similar analyses of 4-hour and 5-hour PM2 5 concentrations are presented in Appendix A, Figure A-2 and Figure
A-3, respectively.

3-65

-------
concentrations likely to occur in areas meeting the current primary standards. However, noting
uncertainty in extrapolating the effects seen in animals, and the PM2.5 exposures and doses that
cause those effects to human populations, animal toxicology studies are of limited utility in
informing decisions on the public health protection provided by the current or alternative
primary PM2.5 standards. As such, the animal toxicological studies are most useful in providing
further evidence to support the biological mechanisms and plausibility of various adverse effects.

3.3.3.2 Ambient PM2.5 Concentrations in Locations of Epidemiologic Studies

As summarized in section 3.1.1 above, epidemiologic studies examining associations
between daily or annual average PM2.5 exposures and mortality or morbidity represent a large
part of the evidence base supporting several of the 2019 ISA's "causal" and "likely to be causal"
determinations. In this section, we consider the ambient PM2.5 concentrations present in areas
where epidemiologic studies have evaluated associations with mortality or morbidity, and what
such concentrations may indicate regarding the primary PM2.5 standards. As noted in section 3.2,
the use of information from epidemiologic studies to inform conclusions on the primary PM2.5
standards is complicated by the fact that such studies evaluate associations between distributions
of ambient PM2.5 and health outcomes, and do not identify the specific exposures that can lead to
the reported effects. Rather, health effects can occur over the entire distribution of ambient PM2.5
concentrations evaluated, and epidemiologic studies conducted to date do not identify a
population-level threshold below which it can be concluded with confidence that PM-associated
health effects do not occur . (U.S. EPA, 2019, section 1.5.3). To address these issues, we
consider the following question:

• To what extent does the evidence from epidemiologic studies that have evaluated
associations with mortality or morbidity provide support for adverse effects
occurring following PM2.5 exposures?

In the absence of discernible thresholds, we consider what information can be provided
from epidemiologic studies. In particular, to address the question above, we consider the study-
reported ambient PM2.5 concentrations reflecting estimated exposure with a focus on the middle
portion of the PM2.5 air quality distribution, which provides the strongest support for reported
health effect associations. The section below discusses the key epidemiologic studies available in
this reconsideration and observations from these studies to inform conclusions on the primary
PM2.5 standards.

3.3.3.2.1 PM2.5 Air Quality Distributions Associated with Mortality or Morbidity in Key
Epidemiologic Studies

In this section, we consider the PM2.5 air quality distributions associated with mortality or
morbidity in key epidemiologic studies. In previous reviews, the decision framework used to

3-66

-------
judge adequacy of the existing PM2.5 standards, and what levels of any potential alternative
standards should be considered, placed significant weight on epidemiologic studies that assessed
associations between PM2.5 exposure and health outcomes that were most strongly supported by
the body of scientific evidence. In doing so, the decision framework recognized that while there
is no specific point in the air quality distribution of any epidemiologic study that represents a
"bright line" at and above which effects have been observed and below which effects have not
been observed, there is significantly greater confidence in the magnitude and significance of
observed associations for the part of the air quality distribution corresponding to where the bulk
of the health events in each study have been observed, generally at or around the mean
concentration. This is the case both for studies of daily PM2.5 exposures and for studies of annual
average PM2.5 exposures.

Studies of daily PM2.5 exposures examine associations between day-to-day variation in
PM2.5 concentrations and health outcomes, often over several years. While there can be
considerable variability in daily exposures over a multi-year study period, most of the estimated
exposures reflect days with ambient PM2.5 concentrations around the middle of the air quality
distributions examined (i.e., "typical" days rather than days with extremely high or extremely
low concentrations). Similarly, for studies of annual PM2.5 exposures, most of the health events
occur at estimated exposures that reflect annual average PM2.5 concentrations around the middle
of the air quality distributions examined. In both cases, epidemiologic studies provide the
strongest support for reported health effect associations for this middle portion of the PM2.5 air
quality distribution, which corresponds to the bulk of the underlying data, rather than the extreme
upper or lower ends of the distribution. Consistent with this, as noted above in section 3.3.1.1,
several epidemiologic studies report that associations persist in analyses that exclude the upper
portions of the distributions of estimated PM2.5 exposures, indicating that "peak" PM2.5
exposures are not disproportionately responsible for reported health effect associations.

An example of the relationship between data density and reported health effect
associations is illustrated in Figure 3-2 below (from Lepeule et al., 2012, Figure 1 in
supplemental material; U.S. EPA, 2019, Figure 6-26). For the years 1974 to 2009, Lepeule et al.
(2012) report a positive and statistically significant association between estimated long-term
PM2.5 exposures and cardiovascular mortality in six U.S. cities. Based on a visual inspection of
the concentration-response function reported in this study (i.e., presented in Figure 3-2), 95%
confidence intervals are narrowest for long-term PM2.5 concentrations near the overall mean
concentration reported in the study (i.e., 15.9 |j,g/m3). Confidence intervals widen at lower and
higher long-term PM2.5 concentrations, particularly at concentrations < -10 |j,g/m3 and > -20
Hg/m3. This widening in the confidence intervals is likely due in part to the comparative lack of

3-67

-------
data at concentrations approaching the lower and upper ends of the air quality distribution (i.e.,
exposure estimates are indicated by hash marks on the horizontal axis).

Cardiovascular mortality

¦*-*
05

Psj

-C

—1

¦IIIMj 1IMII III III I

10 IS 20 2S 30 3S 40
PM2.5(Mg/m3)-moving average
1-3 years

Figure 3-2. Estimated concentration-response function and 95% confidence intervals
between PM2.5 and cardiovascular mortality in the Six Cities Study (1974-2009) (from
Lepeule et al., 2012, supplemental material, figure 1; Figure 6-26 in U.S. EPA, 2019).

Similar to the information presented in Figure 3-2, other studies have also reported that
confidence intervals around concentration-response functions are relatively narrow at PM2.5
concentrations around the overall mean concentrations reported by those studies, likely reflecting
high data density in the middle portions of the distributions (e.g., Crouse et al., 2015; Villeneuve
et al., 2015; Shi et al., 2016 as discussed in U.S. EPA, 2019, section 11.2.4). Thus, consistent
with the approaches in the 2012 and 2020 reviews (78 FR 3161, January 15, 2013; U.S. EPA,
2011, sections 2.1.3 and 2.3.4.1; 85 FR 82716-82717, December 18, 2020; U.S. EPA, 2020,
sections 3.1.2 and 3.2.3), in this reconsideration, we use study-reported means (or medians) of
daily and annual average PM2.5 concentrations over the entire study period as indicators for the
middle portions of the air quality distributions, over which studies generally provide strong
support for reported associations and for which our confidence in the magnitude and significance
of associations observed in the epidemiologic studies is greatest (78 FR 3101, January 15, 2013).
As described further below, when considering the PM2.5 air quality distributions in
epidemiologic studies in this section, we focus on PM2.5 concentrations around these overall
means.

3-68

-------
In addition to the overall study means, we also focus on concentrations somewhat below
the means (e.g., 25th and 10th percentiles), when such information is available from the
epidemiologic studies, which is consistent with approaches used in previous reviews. In so
doing, we continue to note, as in previous reviews, that a relatively small portion of the health
events are observed in the lower part of the air quality distribution and our confidence in the
magnitude and significance of the associations begins to decrease in the lower part of the air
quality distribution. Furthermore, consistent with past reviews, we recognize that there is no
single percentile value within a given air quality distribution that is most appropriate or "correct"
to use to characterize where our confidence in associations becomes appreciably lower.

However, we note that the range from the 25th to 10th percentiles is a reasonable range to
consider as a region where we have appreciable less confidence in the associations observed in
epidemiologic studies compared to the means21 (78 FR 3101, January 15, 2013).

• Does the currently available epidemiologic information call into question the

approaches used in previous reviews for evaluating the relationship between study-
reported mean PM2.5 concentrations and the current or potential alterative primary
PM2.5 standards?

In evaluating the overall study-reported means, and concentrations somewhat below the
means from epidemiologic studies, the focus is on the form, averaging time and level of the
current primary annual PM2.5 standard. Consistent with the approaches used in the 2012 and
2020 reviews (78 FR 3161-3162, January 15, 2013; 85 FR 82716-82717, December 18, 2020),
the annual standard has been utilized as the primary means of providing public health protection
against the bulk of the distribution of short- and long-term PM2.5 exposures. Thus, the evaluation
of the study-reported mean concentrations from key epidemiologic studies lends itself best to
evaluating the adequacy of the annual PM2.5 standard (rather than the 24-hour standard with its
98th percentile form). This is true for the study-reported means from both long-term and short-
term exposure epidemiologic studies, recognizing that the overall mean PM2.5 concentrations
reported in studies of short-term (24-hour) exposures reflect averages across the study population
and over the years of the study. Thus, mean concentrations from short-term exposure studies
reflect long-term averages of 24-hour PM2.5 exposure estimates. In this way, our examination
aims to evaluate the protection provided by the annual PM2.5 standard against the exposures
where our confidence is greatest for associations with mortality and morbidity in key
epidemiologic studies. We note that the protection provided by the annual standard is evaluated

21 As detailed in the 2011 PA, we note the interrelatedness of the distributional statistics and a range of one standard
deviation around the mean which represents approximately 68% of normally distributed data, and in that one
standard deviation below the mean falls between the 25th and 10th percentiles (U.S. EPA, 2011, p. 2-71; U.S.
EPA, 2005, p. 5-22).

3-69

-------
in conjunction with that provided by the 24-hour standard, with its 98th percentile form, which
aims to provide supplemental protection against the short-term exposures to peak PM2.5
concentrations that can occur in areas with strong contributions from local or seasonal sources,
even when overall ambient mean PM2.5 concentrations in an area remain relatively low.

In focusing on the annual standard, and in assessing the range of study-reported exposure
concentrations for which we have the strongest support for adverse health effects, we look to
answer whether the current annual standard provides adequate protection against these exposure
concentrations. This means, as in past reviews, application of a decision framework based on
assessing means reported in key epidemiologic studies must also consider how the study means
were computed and how these values compare to the annual standard metric (including the level,
averaging time and form) and the use of the monitor with the highest PM2.5 design value in an
area for compliance. In the 2012 review, it was recognized that the key epidemiologic studies
computed the study mean using an average across monitor-based PM2.5 concentrations. As such,
the Agency noted that this decision framework applied an approach of using maximum monitor
concentrations to determine compliance with the standard, while selecting the standard level
based on consideration of composite monitor concentrations. Further, the Agency included
analyses (Hassett-Sipple et al., 2010; Frank, 2012) that examined the differences in these two
metrics (i.e., maximum monitor concentrations and composite monitor concentrations) across the
U.S. and in areas included in the key epidemiologic studies and found that the maximum design
value in an area was generally higher than the monitor average across that area, with that amount
varying based on location and concentration. This information was taken into account in the
Administrator's final decision in selecting a level for the primary annual PM2.5 standard the 2012
review and discussed more specifically in her considerations on adequate margin of safety.

The relationship between the mean PM2.5 concentrations and the area design value
continues to be an important consideration in evaluating the adequacy of the current or potential
alternative annual standard levels in this reconsideration. In a given area, the area design value is
based on the monitor in an area with the highest PM2.5 concentrations and is used to determine
compliance with the standard. The highest PM2.5 concentrations spatially distributed in the area
would generally occur at or near the area design value monitor and the distribution of PM2.5
concentrations would generally be lower in other locations and at monitors in that area. As such,
when an area is meeting a specific annual standard level, we would expect the annual average
exposures in that area to be at concentrations lower than that level and the average of the annual
average exposures across that area (i.e., a metric similar to the study-reported mean values) to be
lower than that level.

In this reconsideration, we note that there are a substantial number of different types of
epidemiological studies available since the 2012 review, included in both the 2019 ISA and the

3-70

-------
ISA Supplement, that make understanding the relationship between the mean PM2.5
concentrations and the area design value even more important. While the key epidemiologic
studies in the 2012 review were all monitor-based studies, the newer studies include hybrid
modeling approaches which have emerged in the epidemiologic literature as an alternative to
approaches that only use ground-based monitors to estimate exposure. As assessed in the 2019
ISA and ISA Supplement, a substantial number of epidemiologic studies used hybrid model-
based methods in evaluating associations between PM2.5 exposure and health effects. Hybrid
model-based studies employ various fusion techniques that combine ground-based monitored
data with air quality modeled estimates and/or information from satellites to estimate PM2.5
exposures. While these studies provide a broader estimation of PM2.5 exposures compared to
monitor-based studies (i.e., PM2.5 concentrations are estimated in areas without monitors), the
hybrid modeling approaches result in study-reported means that are more difficult to relate to the
annual standard metric and to the use of maximum monitor design values to assess compliance.
In addition, to further complicate the comparison, when looking across these studies, we find
variations in how exposure is estimated between such studies, and thus, how the study means are
calculated. Two important variations across studies include: (1) variability in spatial scale used
(i.e., averages computed across the national (or large portions of the country) versus a focus on
only CBSAs) and (2) variability in exposure assignment methods (i.e., averaging across all grid
cells, averaging across a scaled-up area like a ZIP code, and population weighting).

Because of these differences and the current state of the science, the application of any
decision framework in considering the study-reported mean PM2.5 concentrations, and whether
the current annual standard provides adequate protection against these reported exposure
concentrations, is more complicated than the approaches used in past reviews. To further
understand how the differences in exposure estimation methods used in the epidemiologic
studies translate into comparisons between the mean PM2.5 concentrations reported in the studies
and the adequacy of the protection provided by the primary annual PM2.5 standard, we seek to
answer the following questions:

• How can the approaches used in epidemiologic studies to estimate exposure affect the
study-reported mean PM2.5 concentrations? How do these approaches and the
resulting means compare to one another? How do these different methods to
estimate exposure and compute mean values affect what these mean values represent
relative to the annual standard?

In answering these questions, we first utilize a simplified example to show differences in
the methods used to estimate exposure can lead to difference in mean concentrations. In Figure
3-3 below, we exhibit the state of Georgia and the CBS A of Atlanta-Sandy Springs-Roswell. In
this figure, the gradient of PM2.5 concentrations are shown for 1 km x 1 km grid cells using one

3-71

-------
of the hybrid approaches described in more detail in Chapter 2, referred to as the DI201922
hybrid approach, from 2014-2016, as well as the monitor locations within the Atlanta-Sandy
Springs-Roswell CBS A and their annual PM2.5 design values for 2016. Using these data, several
metrics were calculated and shown in Table 3-5 below. For all monitors within the CBSA, the
average PM2.5 concentration is 9.3 |ig/m3, while the area design value (based on the highest
monitored PM2.5 concentration in the area) is 10.4 |ig/m3. This comparison helps to illustrate the
fact that composite monitor values tend to be somewhat lower than the highest area monitor
values, consistent with the key points made in the 2012 review. This example also illustrates how
monitors are sited to represent the higher concentrations within the area and that the area's
annual design value, which is used for compliance with the standard, is calculated based on the
highest monitor in the area.

Next, we evaluate the average estimated PM2.5 concentrations from 2014-2016 using the
DI2019 hybrid approach and calculate: (1) the average concentration across the entire state; (2)
the population-weighted average across the entire state; (3) the average concentration across the
CBSA; and (4) the population-weighted average across the CBSA. In doing this, we have
focused on using some of the main approaches used in epidemiologic studies to compute study
means. At the urban level (e.g., Atlanta-Sandy Springs-Roswell CBSA), the average PM2.5
concentration when taking the mean of all grid cells is 9.2 |ig/m3, whereas the population-
weighted mean is 9.6 |ig/m3. Across Georgia, the average PM2.5 concentration using the hybrid
approach and averaged across each grid cell is 8.3 |ig/m3, which is lower than the population-
weighted statewide average of 9.1 |ig/m3. While this is a simple example evaluated in just one
state and one CBSA, it suggests that the lowest mean values tend to result from the approaches
that use concentrations from all or most grid cells (e.g., did not apply population weighting),
both urban and rural, across the study area to compute the mean. Higher mean values are
observed when the approach focuses on the urban areas alone or when the approach incorporates
population weighting. Overall, this example suggests that the means from studies using hybrid
modeling approaches are generally lower than the means from monitor-based approaches, and
means from both approaches are lower than the annual design values for the same area.
Population weighting tends to increase the calculated mean concentration, likely because more
densely populated areas also tend to have higher PM2.5 concentrations. Table 3-5 shows how the
different approaches affect mean concentration estimates for the example discussed above. Note
that while the statewide average using the hybrid approach is quite a bit lower than the mean

22 As discussed above in section 2.3.3.2.4, DI2019 refers to estimated PM2 5 concentrations from a hybrid modeling
approach developed by Di et al. (2019b), which estimates nationwide PM25 concentrations from 2000-2016.

3-72

-------
from either the monitor-based approach or the Atlanta-only hybrid approach, population
weighting the statewide average brings the value closer to the other approaches.

6 6.5 7 7.5 8 8.5 9 9.5 10 10.5 11

Figure 3-3. Estimated PM2.5 concentrations using the DI2019 hybrid approach and

monitoring locations and design values for the state of Georgia and the Atlanta-Sandy
Springs-Roswell, Georgia CBSA. (Note: Additional information on the DI2Q19 hybrid
approach is described in section 2.3.3.1.4 and in Di et al„ 2019a.)

Table 3-5. PM2.5 Concentrations Metrics from Monitor and Modeled Data23

Description of Metric

PM2.5 Concentrations (pg/m3)

Atlanta highest monitor

10.4

Atlanta monitored average

9.3

Atlanta spatial average

9.2

Atlanta population-weighted average

9.6

Georgia spatial average

8,3

Georgia population-weighted average

9.1

To expand upon this example in answering our questions above, we look to the analyses
in Chapter 2 which compared area annual design values, composite monitor PM2.5 concentrations
and mean concentrations from two hybrid approaches. The analyses also included population-

23 "Spatial average" as used in Table 3-5 refers to the average across all grid cells in Atlanta or Georgia using the
DI2019 hybrid modeling approach, while "population-weighted average" uses the DI2019 hybrid modeling
approach and applies population weighting to calculate the mean PM2 5 concentration.

3-73

-------
weighted mean metrics. In the air quality analyses comparing composite monitored PM2.5
concentrations with annual PM2.5 design values in U.S. CBS As, maximum annual PM2.5 design
values were approximately 10% to 20% higher than annual average composite monitor
concentrations (i.e., averaged across multiple monitors in the same CBSA) (section 2.3.3.1,
Figure 2-28 and Table 2-3). The difference between the maximum annual design value and
average concentration in an area can be smaller or larger than this range, depending on factors
such as the number of monitors, monitor siting characteristics, and the distribution of ambient
PM2.5 concentrations.24 Such ratios may also depend on how the average concentrations are
calculated (i.e., averaged across monitors versus across modeled grid cells). Based on these
results, this analysis suggests that there will be a distribution of concentrations and the maximum
annual average monitored concentration in an area (at the design value monitor, used for
compliance with the standard), will generally be 10-20% higher than the average across the other
monitors in the area. Thus, in considering how the annual standard levels would relate to the
study-reported means from monitor-based studies, we can generally conclude that an annual
standard level that is no more than 10-20% higher than monitor-based study-reported mean
PM2.5 concentrations would generally maintain air quality exposures to be below those
associated with the study-reported mean PM2.5 concentrations, exposures for which we have the
strongest support for adverse health effects occurring.

We continue with this assessment to consider information from the epidemiologic studies
that utilized the hybrid modeling approaches. The air quality analyses in Chapter 2 looked at data
from two hybrid modeling approaches. While hybrid modeling approaches are not universal and
the various hybrid approaches all have their different nuances, the analysis in Chapter 2 focused
on the DI2019 and HA2020 approaches, which have been used in several recent epidemiologic
studies. Section 2.3.3.2.4 details a comparison of PM2.5 fields in estimating exposure relative to
design values using these two hybrid modeling surfaces. Annual average PM2.5 concentrations
are estimated per year at a 1 km x 1 km spatial resolution. As exhibited in Figure 2-37, the means
vary when one estimates PM2.5 exposures in urban areas only (CBSAs) versus when the averages
were calculated with all or most grid cells nationwide. This is likely indicative of the fact that
areas included outside of CBS As tend to be more rural and have lower estimated PM2.5

24 Given that higher PM2 5 concentrations have been reported at some near-road monitoring sites, relative to the
surrounding area (section 2.3.2.2.2), recent requirements for PM2 5 monitoring at near-road locations in large
urban areas (section 2.2.3.3) may increase the ratios of maximum annual design values to averaged concentrations
in some areas. In the Georgia example above, a near-road monitor was not included in our analysis. The near-road
monitor was not added until 2015, and data related to DI2019 ended in 2016. For purposes of developing three-
year average concentrations using the most recent data for which we had monitored and modeled data, 2014-2016
data was selected for monitors as well, for which data from 2014-2016 was not available for the near-road
monitor.

3-74

-------
concentrations. This is important to note since the study areas included in the calculation of the
mean, and more specifically whether a study is focused on nationwide, regional, or urban areas,
will affect the calculation of the study mean based on how many rural areas are included with
lower estimated PM2.5 concentrations. While the determination of what spatial scale to use to
estimate PM2.5 concentrations does not inherently affect the quality of the epidemiologic study,
the spatial scale can influence the calculated long-term mean concentration across the study area
and period. As exhibited in Table 2-4, regardless of the hybrid modeling approach assessed, the
annual average PM2.5 concentrations in CBSA-only analyses are 4-8% higher than for
nationwide analyses, likely as a result of higher PM2.5 concentrations in more densely populated
areas, and exclusion of more rural areas. When evaluating comparisons between surfaces that
estimate exposure using population weighting versus surfaces that do not calculate means using
population weighting, surfaces that calculate long-term mean PM2.5 concentrations with
population-weighted averages have higher average annual PM2.5 concentrations, ranging from
8.2-10.2 |ig/m3, compared to annual PM2.5 concentrations that range from 7.0-8.6 |ig/m3 in
analyses that do not apply population weighting. Average maximum annual design values, on the
other hand, exhibit a range from 9.5 to 11.7 |ig/m3. Analyses show that average maximum annual
design values are 40 to 50% higher when compared to annual average PM2.5 concentrations
estimated without population weighting and are 15% to 18% higher when compared to average
annual PM2.5 concentrations with population weighting applied.

The comparisons discussed above for the studies using the hybrid modeling approaches
show a trend generally observed across the various methods employed to calculate the mean.
First, when assessing means from epidemiologic studies that employ hybrid model-based
methods, the lowest means tend to result from the approaches that use estimated PM2.5
concentrations from all or most grid cells across the study area (i.e., both urban and rural) to
compute the mean. When comparing means from these types of studies to the area annual design
values, the annual design values are higher than means by 40-50%. However, when the approach
instead employs methods that population weight the mean, the calculated mean PM2.5
concentrations are higher, regardless of the hybrid method employed, and when compared to the
area annual design values, design values are 15-18% higher than means (similar to the
differences observed for the composite monitor comparison values for the monitor-based
epidemiologic studies). Given these results, it is worth noting that for the studies using the hybrid
modeling approaches, the choice of methodology employed in calculating the study-reported
means (i.e., using population weighting or not), and not a difference in estimates of exposure in
the study itself, can produce substantially different study-reported mean values, with the
approach that doesn't utilize population weighting producing a much lower value.

3-75

-------
Based on these results, and similar to conclusions for the monitor-based studies, we can
generally conclude that the study-reported mean concentrations in the studies that employ hybrid
modeling approaches and population-weight the mean are associated with air quality conditions
that would be achieved by meeting annual standard levels that are 15-18% higher than study-
reported means. Therefore, an annual standard level that is no more than 15-18% higher than the
study-reported means would generally maintain air quality exposures to be below those
associated with the study-reported mean PM2.5 concentrations, exposures for which we have the
strongest support for adverse health effects occurring. For the studies that utilize hybrid
modeling approaches but do not incorporate population weighting in calculating the mean, we
would expect the annual design values associated with these air quality conditions to be much
higher (i.e., 40-50% higher) and this larger difference makes it more difficult to consider how
these studies can be used to determine the adequacy of the protection afforded by the current or
potential alternative annual standards.

While this information can be helpful to inform our understanding of the relationship
between study-reported mean concentrations and the level of the annual standard, we recognize
some limitations to our assessment of this information. First, we note that our comparisons used
only two hybrid modeling approaches. Although the two hybrid modeling surfaces have been
used in a number of recent epidemiologic studies, they represent just two of the many hybrid
modeling approaches that have been used in epidemiologic studies to estimate PM2.5
concentrations. These methods continue to evolve over time, with further development and
improvement to prediction models that estimate PM2.5 concentrations in epidemiologic studies.
In addition to differences in hybrid modeling approaches, we also recognize that epidemiologic
studies use different methods to assign a population-weighted average PM2.5 concentration to
their study population, and our assessment does not evaluate all of the potential methods that
could be used.

Further, we note that some of these epidemiologic studies also provide information on the
broader distributions of exposure estimates and/or health events and the PM2.5 concentrations
corresponding to the lower percentiles of those data (e.g., 25th and/or 10th). Our air quality
analysis focuses on mean PM2.5 concentrations and we did not include a similar comparison for
these lower percentiles, and therefore, any direct comparison of study-reported PM2.5
concentrations corresponding to lower percentiles and annual design values is more uncertain
than such comparisons with the mean. Finally, we also recognize that our air quality analysis
included two hybrid modeling-based approaches that used U.S.-based air quality information for
estimating PM2.5 concentrations. As such, our analyses are most relevant to interpreting the
study-reported mean concentrations from U.S. epidemiologic studies and does not provide
additional information to inform our understanding of how the mean exposures concentrations

3-76

-------
reported in epidemiologic studies in other countries would compare to annual design values
observed in the U.S. In addition, we note that differences in the exposure environments and
population characteristics between the U.S. and other countries can affect the study-reported
mean value and its relationship with the annual standard level. Sources and pollutant mixtures, as
well as PM2.5 concentration gradients, may be different between countries, and the exposure
environments in other countries may differ from those observed in the U.S. Furthermore,
differences in population characteristics and population densities can also make it challenging to
directly compare studies from countries outside of the U.S. to a design value in the U.S.

• What are the epidemiologic studies assessed in the 2019 ISA and ISA Supplement
that have the potential to be most informative in reaching conclusions on the
adequacy of the current primary PM2.5 standards or potential alternative standards?

To evaluate the PM2.5 air quality distributions in key studies in this PA, we first identify
the epidemiologic studies assessed in the 2019 ISA and ISA Supplement that have the potential
to be most informative in reaching conclusions on the primary PM2.5 standards. As with the
experimental studies discussed above, we focus on epidemiologic studies that provide strong
support for "causal" or "likely to be causal" relationships with PM2.5 exposures in the 2019 ISA.
We focus on the health effect associations that are determined in the 2019 ISA and ISA
Supplement to be consistent across studies, coherent with the broader body of evidence (e.g.,
including animal and controlled human exposure studies), and robust to potential confounding by
co-occurring pollutants and other factors. 25

In our assessment of the evidence judged to be most relevant to decisions on the elements
of the primary PM2.5 standards, we place greater weight on U.S. and Canadian studies. This is
because findings of U.S. and Canadian studies are more directly applicable for quantitative
considerations in this reconsideration, as studies conducted in other countries reflect different
populations, exposure characteristics, and air pollution mixtures. Additionally, epidemiologic
studies outside of the U.S. and Canada generally reflect higher PM2.5 concentrations in ambient

25 As described in the Preamble to the ISAs (U.S. EPA, 2015a), "the U.S. EPA emphasizes the importance of
examining the pattern of results across various studies and does not focus solely on statistical significance or the
magnitude of the direction of the association as criteria of study reliability. Statistical significance is influenced
by a variety of factors including, but not limited to, the size of the study, exposure and outcome measurement
error, and statistical model specifications. Statistical significance may be informative; however, it is just one of
the means of evaluating confidence in the observed relationship and assessing the probability of chance as an
explanation. Other indicators of reliability such as the consistency and coherence of a body of studies as well as
other confirming data may be used to justify reliance on the results of a body of epidemiologic studies, even if
results in individual studies lack statistical significance. Traditionally, statistical significance is used to a larger
extent to evaluate the findings of controlled human exposure and animal toxicology studies. Understanding that
statistical inferences may result in both false positives and false negatives, consideration is given to both trends in
data and reproducibility of results. Thus, in drawing judgments regarding causality, the U.S. EPA emphasizes
statistically significant findings from experimental studies, but does not limit its focus or consideration to
statistically significant results in epidemiologic studies."

3-77

-------
air than are currently found in the U.S., and thus, are less relevant to informing questions about
adequacy of the current standards.26 However, we do note that while information from Canadian
studies can be useful in assessing the adequacy of the annual standard, there are still important
differences between the exposure environments in the U.S. and Canada and interpreting the data
(e.g., mean concentrations) from the Canadian studies in the context of a U.S.-based standard
may present challenges in directly and quantitatively informing questions regarding the adequacy
of the current or potential alternative the levels of the annual standard. Lastly, we also emphasize
multicity/multistate studies that examine health effect associations, as such studies are more
encompassing of the diverse atmospheric conditions and population demographics in the U.S.
than studies that focus on a single city or state.

Figure 3-4 to Figure 3-7 below summarize information from U.S. and Canadian studies
that are assessed in the 2019 ISA and ISA Supplement and that meet these criteria. For each
study, Figure 3-4 to Figure 3-7 present the cohort27 and/or geographic area examined, the
approach used to estimate PM2.5 exposures (i.e., monitored or predicted with hybrid modeling
methods28), the study years during which health events occurred, the years of PM2.5 air quality
data used to estimate exposures, and the effect estimate29 with 95% confidence intervals (per 5
Hg/m3 for long-term exposures; 10 |j,g/m3 for short-term exposures). When available, these
figures also include the overall means (or medians if means are not available) of the short- or
long-term PM2.5 exposure estimates reported by the study. Figure 3-4 and Figure 3-5 summarize
information from studies of long-term PM2.5 exposures. Figure 3-6 and Figure 3-7 summarize
information from studies of short-term PM2.5 exposures.

26 This emphasis on studies conducted in the U.S. or Canada is consistent with the approach in the 2012 and 2020
reviews of the PMNAAQS (U.S. EPA, 2011, section 2.1.3; U.S. EPA, 2020, section 3.2.3.2.1) and with
approaches taken in other NAAQS reviews. We recognize the importance of studies in the U.S., Canada, and
other countries in informing an ISA's considerations of the weight of the evidence that informs causality
determinations.

27 The cohorts examined in the studies included in Figure 3-4 to Figure 3-7 include large numbers of individuals in
the general population, and often also include those populations identified as at-risk (i.e., children, older adults,
minority populations, and individuals with pre-existing cardiovascular and respiratory disease).

28 As discussed further below, and in Chapter 2, hybrid methods incorporate data from several sources, often
including satellites and models, in addition to ground-based monitors.

29 The effect estimates presented in the forest plot figures (Figure 3-4 to Figure 3-7) show the associations of long-
or short-term PM2 5 exposures with health endpoints presented either as hazard ratio or odds ratio or relative risk
(for which the bold dotted vertical line is at 1), or as per unit or percent change (for which the bold dotted vertical
line is at 0).

3-78

-------
All-cause mortality (U.S.)

Air Quality Reported PM Mean
Citation Cohort Health Data Data (Range)(ug/m3)

Di etal., 2017b Medicare 2000-2012 2000-2012 11.0 (5th and 95th: 6.21-15.64)

•

Dominici etal., 2019 Medicare 2000-2012 2000-2012 11.0 (NR)

•-

Nurses

Elliott et al., 2020 ,, _ 1988-2008 1988-2007 13.7 (NR)
Health v '

Nurses

Hart etal., 2015 ,, M 2000-2006 1999-2006 12.0 (NR)
Health x '

Lefler et al., 2019 NHIS 1987-2015 1988-2015 10.7 (NR)

Malik etal., 2019 2003-2008 2002-2007 12.0(4.3-20.5)
PREMIER v '

Pope et al., 2015 ACSCPS-II 1982-2004 1999-2004 12.6(1.0-28.0)

Pope etal., 2019 NHIS 1986-2015 1999-2015 10.7(2.5-19.2)

Nurses

Puett etal., 2009 ,, 1992-2002 1988-2002 13.9(5.8-27.6)
Health v '

Puett etal., 2011 Health 1989-2003 1988-2003 17.8 (NR)
Professionals x '

Shi etal., 2016 Medicare 2003-2008 2003-2008 8.12(0.8-20.22)

—

Thurston et al., 2016 NIH-AARP 2000-2009 2000-2008 12.2(2.9-28.0)

r«-

Turner etal., 2016 ACSCPS-II 1982-2004 1999-2004 12.6(1.4-27.9)

Wang etal., 2017 Medicare 2000-2013 2000-2013 NR (Median: 10.7) (6.0-20.6)

—•

Wang etal., 2020 Medicare 2000-2008 2000-2008 10.3 (NR)

Weichenthal etal., ... Iowa: 8.8; North Carolina: 11.1

2Ql4 Ag Health 1993-2009 2001-2006 ^

Wu etal., 2020a Medicare 2000-2016 2000-2016 9.8 (NR)

•-

Eum et al., 2018 Medicare 2000-2012 2000-2012 Overall: 11.7 (NR)

Central region: 9.9 (NR)
Eastern region: 12.3 (NR)
Western region: 11.5 (NR)

•

•
•

•

GOSSetal., 2004 U S-Cystic 1999_2000 2000 13.7 (NR)
Fibrosis v '

Hart et al., 2015 I^T,5 2000-2006 2000-2006 12.7 (NR)
Health

Kiomourtzoglou etal.,.. .. 12.0(Mean Range:9.0-13.0)
2016 y Medicare 2000-2010 2000-2010 y '

. Harvard 1974-2009:15.9; 2000 onwards
Lepeuleetal.,2012 ^ 2001-2009 1979-2009 '

Six-City mean range: <15 <18 (NR)

Lipfertetal., 2006 Veterans 1997-2001 1999-2001 14.3(NR)

•

Zegeretal., 2008 MCAPS 2000-2005 2000-2005 Central^regionrNR (Median:

Eastern region: NR (Median:

14.0) (NR)

Western region: NR (Median:

13.1) (NR)

0.95 1.00 1.05 1.10 1.15 1.20 1.25 1.30 1.35 1.40 1.45 1.50 1.55
Hazard Ratio (95% CI) *

3-79

-------
All-cause mortality (Canada)

Exposure

Proxy Citation Cohort

Modelled Cakmaket al., 2018 CanCHEC

Christidisetal., 2019 mCCHS

Crouse et al., 2012 CanCHEC

Crouse et al., 2015 CanCHEC

Crouse et al., 2020 CanCHEC

Ericksonetal., 2019 CanCHEC

Ericksonetal., 2020 CanCHEC

Pappin et al., 2019 CanCHEC

Pinaultetal., 2016 CCHS
Pinault etal., 2017 CanCHEC

Air Quality Reported PM Mean
Health Data Data (Range)(ug/m3)

1991-2011 1984-2011 6.5 (1.2, 24.1)

2000-2016 1998-2015 5.9 (0.4-17.2)
1991-2001 2001-2006 8.7 (1.9-19.2)

1991-2006 1984-2006 8.9 (0.9-17.6)

7.2 (1-year 1-km mean)

2001-2011 1998-2010 fn *' . '

(0.0-20.0)

7.4 (3-year 1-km mean)
(0.0-20.0)

8.0 (8-year 1-km mean)
(0.3-18.4)

2001-2011 1998-2012 8.4 (NR)

2001-2011 1998-2012 6.7 (NR)

2001-2016 1998-2016 7.5 (Non-Immigrant) (NR)

9.1 (Immigrant >30 yrs) (NR)
9.7 (Immigrant <10 yrs) (NR)

1991-2016 1988-2015 7.9 (Year 1991) (0.4-20.0)

7.2 (Year 1996) (0.4-20.0)
6.7 (Year 2001) (0.4-18.5)

2000-2011 1998-2011 6.3 (1.0-13.0)

1991-2011 1998-2011 7.4 (<0.01-20.0)

Zhang et al., 2021

Ontario Health

Study

Monitor Crouse et al., 2012 CanCHEC
Weichenthal etal.,

2016a

" CanCHEC

2009-2017 2000-2016 7.8 (NR)
1991-2001 1987-2001 11.2 (NR)
1991-2009 1998-2009 9.8 (4.74-13.62)

0.90 0.95 1.00 1.05 1.10 1.15 1.20 1.25 1.30 1.35
Hazard Ratio (95% CI) *

3-80

-------
CVD mortality

Exposure Health Air Quality Reported PM Mean Health
Proxy Country Citation Cohort Data Data (Range)(ug/m3) Outcome

Modelled U.S. 8tal" NIH-AARP J^/9& 1980-2010 NR (Median: 13.3) (2.9-28.0) CV0 mortality
2020 2011 v Age 50-71

—

JerrettetaL, . , IHD mortality
2016 ACS CPS-II 1982-2004 2002-2004 12.0(1.5-26.6) Age 30+

Popeetal., ACSCPS-II 1982-2004 1999-2004 12.6(1.0-28.0) CVD morta ity
2015 A9G30+

IHD mortality
Age 30+

other CVD-

CBVD Age 30+

—

Popeetal., . . CVD mortality
2QUig NHIS 1986-2015 1999-2015 10.7(2.5-19.2) Age 18-84

—

Thurston et . . CVD mortality
al 2016 NIH-AARP 2000-2009 2000-2008 12.2(2.9-28.0) Age 50-71

—

Turner etal., ACSCPS-II 1982-2004 1999-2004 12.6(1.4-27.9) CVD mortality

2016 A9G30+

IHD mortality
Age 30+

Other CVD-
CBVD Age 30+

—

Wang etal., .. , , CVD mortality

* Medicare 2000-2008 2000-2008 10.3 (NR) B
2020 v Age 65-120

Weichenthal , , Iowa: 8.8; North Carolina: 11.1

etal 2014 A9 Health 1993-2009 2001-2006 CVD mortality

Chen etal., , . CVD mortality
Canada EFFECT RCT 1999-2011 2001-2010 10.7 (NR) . _c, '
2016 x ' Age 35+

Chen etal., „ , . CVD mortality
2020 °NPHEC 2001-2016 2000-2016 8.6 (NR) Age 35.35

Crouseetal., . . CVD mortality
2012 CanCHEC 1991-2001 2001-2006 8.7 (1.9-19.2) Agg 25+

Crouseetal., . , CVD mortality
2Q15 CanCHEC 1991-2006 1984-2006 8.9(0.9-17.6) Age 25-90

Crouseetal., 7.4(3-year 1-km mean) CVD mortality
2020 CanCHEC 2001-2011 1998-2010 (Q ^

Ericksonet CanCHEC 2001-2011 1998-2012 8.4 (NR) ^VD 'tY

, _ v ' Age 25-89

al., 2019 y

CVD mortality

mCCHS 2001-2011 1998-2012 6.7 (NR) .

x ' Age 25-89

Pinaultetal., , , CVD mortality
2016 CCHS 2000-2011 1998-2011 6.3(1.0-13.0) Age 25-90

Pinaultetal., , . CVD mortality

CanCHEC 1991-2011 1998-2011 7.4 (<0.01-20.0) .
2017 v ' Age 25-89

—

^ ^ . , CVD mortality
Pinaultetal., CanCHEC 2001-2011 1998-2012 7.4(NR) .

Age 25-90

2018 y

—

. , CVD mortality
mCCHS 2001-2008 1998-2013 6.4 (NR) a

v ' Age 25-90

Villeneuveet CNBSS 1980-2005 1998-2006 9.1(1.3-17.6) ^V°
al, 2015 A9e40"59

IHD mortality
Age 40-59

Zhang etal., Ontario , . CVD mortality
2021 Health Study2009"2017 2000-2016 7.8 (NR) Age3Q+

Hart etal.,

Monitor U.S. 2011 TrIPS 1985-2000 2000 14.1 (NR) CVD mortality

Lepeuleetal., Harvard „ 1974-2009:15.9; 2000 onwards CVD mortality
2012 Six-City 2001-2009 1979-2009 mgan rangg; <15.<18 Age 25_74

Miller, etal. , . CVD mortality
2007 WHI 1994-2002 2000 13.5(3.4-28.3)

Canada CanCHEC 1991-2009 1998-2009 9.8(4.74-13.62) ,"D

etal., 2016a ' ' Age 25-89

0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
Hazard Ratio (95% CI) *

3-81

-------
Respiratory mortality

Exposure Health Air Quality Reported PM Mean
Proxy Country Citation Cohort Data Data (Range)(ug/m3)

Modelled U.S. Pope et a 1., 2015 ACSCPS-II 1982-2004 1999-2004 12.6(1.0-28.0)

Thurston etal., 2016 NIH-AARP 2000-2009 2000-2008 12.2(2.9-28.0)

Turneretal., 2016 ACSCPS-II 1982-2004 1999-2004 12.6(1.4-27.9)

Wang et al., 2020 Medicare 2000-2008 2000-2008 10.3 (NR)

•

Canada Crouseetal., 2015 CanCHEC 1991-2006 1984-2006 8.9(0.9-17.6)

Crouseetal., 2020 CanCHEC 2001-2011 1998-2010 00 1 kmmean^

(0.0-20.0)

—

Pinaultetal., 2016 CCHS 2000-2011 1998-2011 6.3(1.0-13.0)

——

Pinaultetal., 2017 CanCHEC 1991-2011 1998-2011 7.4(<0.01-20.0)

—

Zhang etal., 2021 . 2009-2017 2000-2016 7.8 (NR)
a Health Study v '

Monitor U.S. Hart etal., 2011 TrlPS 1985-2000 2000 14.1 (NR)

Canada ^ei^henthal etal-' CanCHEC 1991-2009 1998-2009 9.8(4.74-13.62)
2016a v '

0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2
Hazard Ratio (95% CI) *

Lung cancer mortality

Exposure Health Air Quality Reported PM Mean
Proxy Country Citation Cohort Data Data (Range)(ug/m3)

.. . Turner etal.,

Modelled U.S. 2015 ACSCPS-II 1982-2004 1999-2004 12.6(1.4-27.9)

Crouse etal.,

Canada 2015 CanCHEC 1991-2006 1984-2006 8.9(0.9-17.6)

Ericksonetal., CanCHEC 2001-2011 1998-2012 8.4(NR)
2019

mCCHS 2001-2011 1998-2012 6.7 (NR)

20^66t a' ' CCHS 2000-2011 1998-2011 6.3(1-0-13.0)

Villeneuveet

al 2015 CNBSS 1980-2005 1998-2006 9.1(1.3-17.6)

Monitor U.S. j^ietaL' Trips 1985-2000 2000 14.1 (NR)

Krewski et al., „ 1979-1983; 1979-1983: 21.2; 1999- 2000:14.0
2009 ACSCPS-" 19822000 1999-2000 (NR)

Laden etal.. Harvard 1979-1987; , . , .
2006 Six-City 1974-1998 19a5.1998 16.4 (Mean Range: 10.2-29.0) (NR)

Lepeuleetal., Harvard „ 1974-2009:15.9; 2000 onwards
2012 Six-City 2001-2009 1979-2009 mean range; <15-<18 (NR)

Thurston et , .
al 2013 ACS CPS-II 1982-2004 2000-2005 14.2 (NR)

Weichenthal

Canada j 2016a CanCHEC 1991-2009 1998-2009 9.8(4.74-13.62)

0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6
Hazard Ratio (95% CI) *

Figure 3-4. Epidemiologic studies examining associations between long-term PM2.5
exposures and mortality.

3-82

-------
Asthma incidence

Exposure

Proxy Health Air Quality Reported PM Mean
(group) Country Citation Cohort Data Data (Range)(ug/m3)

Modelled Canada ^®^aultetal-' QicDSS 1996-2011 2001-2006 9.86(NR)

•

Monitor U.S. McCornell et al, CHS 2003-2005 2003-2004 13.9(6.3-23.7)
2010 '

Nishimuraetal., , „„„„ Mean Range: 8.1-17.0

GALA III1 SAGE II 1986-2003 198o-£i003 , ,
2013 ' (NR)

1.0 1.5 2.0
Odds Ratio (95% CI)

Lung cancer incidence

Exposure
Proxy

Country

Citation

Cohort

Health Air Quality
Data Data

Reported PM Mean
(Range)(ug/m3)

Modelled

Canada

Hystad etal.,2013

NECSS

1994-1997 1975-1994

11.9 (NR) — •

Tomczaketal., 2016

CNBSS

1980-2004 1998-2006

9.1 (1.3-17.6) •

U.S.

Puettetal., 2014

Nurses Health

1994-2010 1988-2007

13.1 (NR) i—•

Monitor

U.S.

Gharibvand et al., 2016

AHSMOG-2

2002-2011 2000-2001

12.9 (NR) : •

0.9 1.0 1.1 1.2 1.3 1.4
Hazard Ratio (95% CI) *¦

Lung development

Exposure Health Air Quality Reported PM Mean

Proxy Country Citation Cohort Data Data (Range)(ug/m3) Health Outcome

Monitor U.S. Breton et al., CHS 1993-2000, 1994-2004 Mean Range: r!?.08?!16'11' (Change)
2011 1996-2004 6.0-28.0 (NR) FVC Age 10-18

Lung Development (Change) -
MMEF Age 10-18

Lung Development (Change)-
FEV1 Age 10-18

Gauderman CHS 1993-2000 1994-2000 Mean Range: W^e, nt (change).
et al., 2004 6.0-28.0 (NR) Age 10-18

Lung Development (Change) -
MMEF Age 10-18

Lung Development (Change)-
FEV1 Age 10-18

-80 -60 -40 -20 0
ml Change in Growth (95% CI)

Lung function

Exposure Health Air Quality Reported PM Mean

Proxy Country Citation Cohort Data Data (Range)(ug/m3) Health Outcome

Monitor U.S. Urmanetal., CHS 2002-2007 2002-2007 6.0-28.0 (NR) Lung Function Decline (%)-
2014 98

Lung Function Decline (%)-
FVC Age 5-7

-1.5 -1.0 -0.5 0.0
Percent Difference (95% CI)

3-83

-------
CVD morbidity

Exposure

Proxy Endpoint

Modelled CHD

Monitor CHD
CVD
Ml

Health Outcome

CHD Incidence

CHD Incidence Ages 45+

CVD (First event) Age
50-79

CVD Incidence

CHF Incidence Ages
35-85

Heart Failure (First HA)
Age 65+

AMI Incidence Ages
35-85

AMI Incidence Age
35-85

Ml Incidence Age 42-67
Ml Incidence Ages 45+
Ml Incidence

Ml (First HA) Age 65+

Stroke Incidence Age
42-67

Stroke Incidence
IS Incidence

HS Incidence

Stroke Incidence Ages
35-85

Stroke (First HA) Age
65+

CHD Incidence Age
50-79

CVD Incidence Age
50-79

Ml Incidence Age 50-79

Stroke Incidence Age
50-79

Citation

Hartetal., 2015b
Loop et al., 2018
Chi etal., 2016a
Hartetal., 2015b
Baietal., 2019
Yazdietal., 2019
Baietal., 2019
Chen etal., 2020
Elliott et al., 2020
Loop etal., 2018
Puett et al., 2011
Yazdi et al., 2019
Elliott etal., 2020
Hart et al., 2015b
Puett etal., 2011
Puett etal., 2011
Shin et al., 2019
Yazdi et al., 2019
Miller, etal. 2007
Miller, etal. 2007
Miller, etal. 2007
Miller, etal. 2007

Cohort

Nurses Health

REGARDS

WHI

Nurses Health

ONPHEC

Medicare

ONPHEC

Nurses Health

REGARDS
Health

Professionals
Medicare

Nurses Health

Nurses Health
Health

Professionals
Health

Professionals
ONPHEC

Medicare

WHI

Health
Data

1988-2006
2003-2012

1993-2005
1988-2006
2001-2015

2000-2012

2001-2015
2001-2016

1988-2008
2003-2012

1989-2003

2000-2012
1988-2008

1988-2006

1989-2003
1989-2003

2001-2015
2000-2012

1994-2002
1994-2002
1994-2002

Reported PM Mean
(Range)(ug/m3)

13.4 (NR)

13.6 (median) (NR)

12.7 (NR)

13.4 (NR)
9.6 (1.1-20.0)
NR (NR)

9.6 (1.1-20.0)
8.6 (NR)

13.7 (NR)

13.6 (median) (NR)

17.8 (NR)

NR (NR)

13.7 (NR)

13.4 (NR)

17.8 (NR)

17.8 (NR)
9.8 (1.0-20.0)
NR (NR)

13.5 (3.4-28.3)
13.5 (3.4-28.3)
13.5 (3.4-28.3)
13.5 (3.4-28.3)

1.0 1.2 1.4 1.6
Relative Risk (95% CI) *

Figure 3-5. Epidemiologic studies examining associations between long-term PM2.5
exposures and morbidity.

3-84

-------
All-cause mortality

Exposure

Proxy Country Citation

Modelled U.S. Dietal., 2017a

Lee etal., 2015b

Cohort

Medicare

State Dept

Health Data

2000-2012

Reported PM Mean
(Range)(ug/m3)

11.6 (5th and 95th: 6.21-
15.64)

2007-2011 11.1 (0.02- 86.2)

Shi et al., 2016
Monitor Canada Burnett etal., 2003
Burnett etal., 2004
Lavigneetal., 2018
Liu etal., 2019
Shin etal., 2021
U.S. Baxter et al., 2017
Dai et al., 2014
Dominic! etal., 2007
Franklin etal., 2007
Franklin etal., 2008
Klemm etal., 2003

Medicare
Statistics Canada

Statistics Canada

Canadian Mortality
Database

MCC

National Vital
Statistics Database

NCHS
NCHS
NMMAPS
NCHS/State Dept
NCHS/State Dept
Harvard Six-City

2003-2008
1986-1996
1981-1999

1998-2011
1986-2011
1984-2012
2001-2005
2000-2006

1999-2000
1997-2002

2000-2005
1979-1988

8.21 (0.8-20.22)

13.3 (NR)

12.8 (NR)

8.8 (<1-98.15)

9.3 (NR)

8.0 (Warm season); 6.0 (Cold
season) (NR)

Cluster Mean Range: 12.2-14.1
(NR)

13.3 (NR)

NR (NR)

15.6 (NR)

14.8 (NR)

14.7 (Median: 9.0) (NR)

Krai I et al., 2013

NCHS

2000-2005 13.6 (NR)

Liu etal., 2019

1987-2006 12.4 (NR)

Zanobetti and Schwartz,

1999-2005 13.2 (NR)

Zanobetti etal., 2014 Medicare

1999-2010 Mean Range: 4.37-17.97 (NR)

0.5 1.0 1.5 2.0
Percent Increase (95% CI)

CVD mortality

Exposure
Proxy

Country

Citation

Cohort

Health Data

Reported PM Mean
(Range)(ug/m3)

Modelled

U.S.

Lee etal., 2015b

State Dept

2007-2011

11.1(0.02-86.2)

Monitor

U.S.

Dai et al., 2014

NCHS

2000-2006

13.3 (NR)

Franklin etal., 2007

NCHS/State Dept

1997-2002

15.6 (NR)

Franklin etal., 2008

NCHS/State Dept

2000-2005

14.8 (NR)

Zanobetti and Schwartz, 2009

NCHS

1999-2005

13.2 (NR)

Canada

Lavigne et al., 2018

Canadian Mortality

1998-2011

8.8 (<1-98.15)

Database

0.0

0.5

1.0 1.5 2.0
Percent Increase (95% CI)

2.5

3.0

3-85

-------
Respiratory mortality

Exposure
Proxy

Country

Citation

Cohort

Health Data

Reported PM Mean
(Range)(ug/m3)

Modelled

U.S.

Lee etal., 2015b

State Dept

2007-2011

11.1 (0.02- 86.2) -

Monitor

U.S.

Dai etal., 2014

NCHS

2000-2006

13.3 (NR)

Franklin etal., 2007

NCHS/State Dept

1997-2002

15.6 (NR)

Franklin etal., 2008

NCHS/State Dept

2000-2005

14.8 (NR)

Zanobetti and Schwartz, 2009

NCHS

1999-2005

13.2 (NR)

Canada

Lavigne etal., 2018

Canadian Mortality
Database

1998-2011

8.8 (<1-98.15) •

Shin et al., 2021

National Vital Statistics
Database

1984-2012

8.0 (Warm season);
6.0 (Cold season) (NR)

-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Percent Increase (95% CI)

Figure 3-6. Epidemiologic studies examining associations between short-term PM2.5
exposures and mortality.30

CVD morbidity (US)

Exposure Proxy

Endpoint

Health Outcome

Citation

Cohort

Reported PM Mean
Health Data (Range)(ug/m3)

Modelled

CVD

CVD (First HA) Age 18+

deSouza et al., 2021

Medicaid adults

2000-2012

11.5 (NR)

•

CVD 30-day Hospital
re-admission

Wyattetal., 2020c

USRDS haemodialysis
patients

2008-2014

9.3 (0.05-155.16)

CVD HA Age 65+

Bravo et al., 2017

Medicare

2002-2006

12.3 (NR)

•

Kloog etal., 2012

Medicare

2000-2006

9.6 (0.01-72.59)

Kloog et al., 2014

Medicare

2000-2006

11.9 (NR)

Monitor

CVD

CVD HA Age 65+

Bell etal., 2008

Medicare

1999-2005

12.9 (NR)

Bell et al., 2014

Medicare

2000-2004

14.0 (Median: 11.7)
(NR)

Bravo etal., 2017

Medicare

2002-2006

12.5 (NR)

Peng etal., 2009

Medicare

2000-2006

NR (Median: 11.8)
(NR)

-1 0 1 2 3 4 5
Percent Increase (95% CI) *

30 As noted above, the overall mean PMSS concentrations reported in studies of short-term (24-hour) exposures
reflect averages across the study population and over the years of the study. Thus, mean concentrations reflect
long-term averages of 24-hour PMis exposure estimates.

3-86

-------
CVD (cause-specific) morbidity (US)

Exposure

Proxy Endpoint Health Outcome

Modelled CHF

Monitor HF

CHF (First HA) Age 18+
CHF ED

CHF 0-7 day Hospital

re-admission

CHF 8-30 day Hospital

re-admission

Citation

deSouza et al., 2021
Kralletal., 2018
Wyattetal., 2020c
Wyatt et al., 2020c

Cohort

Medicaid adults

ED visit databases

USRDS haemodialysis
patients

Health Data

2000-2012

2002-2008
2008-2014
2008-2014

Reported PM Mean
(Range)(ug/m3)

11.5 (NR)

Mean Range: 10.8-15.4
(NR)

9.3 (0.05-155.16)
9.3 (0.05-155.16)

IHD HA Age 65+

Ml Ml (First HA) Age 18+

Stroke is (First HA) Age 18+

Kloog et al., 2014
Kralletal., 2018
deSouza et al., 2021
deSouza et al., 2021

Medicare
ED visit databases
Medicaid adults
Medicaid adults

2000-2006
2002-2008
2000-2012
2000-2012

11.9 (NR)

Mean Range: 10.8-15.4
(NR)

11.5 (NR)
11.5 (NR)

Total Stroke Ages 53-88 Fisher etal., 2019 HPFS
IS Ages 53-88 Fisher etal., 2019 HPFS

1999-2010
1999-2010

12.9 (NR)
12.9 (NR)

HS Ages 53-88

Undetermined stroke
Ages 53-88
Total Stroke Incidence
Ages 50-79

IS Incidence Ages 50-79

HS Incidence Ages
50-79

Total Stroke Incidence
(High vs Low <12 ug/m..
IS Incidence (High vs
Low <12 ug/m3)
HS incidence (High vs
Low <12 ug/m3)

Stroke ED

Heart Failure HA Age
65+

Fisher etal., 2019
Fisher et al., 2019
Sun etal., 2019
Sun etal., 2019
Sun etal., 2019
MCCIure etal., 2017

HPFS

WHI (Post-menopausal
women)

REGARDS

1999-2010
1999-2010
1993-2012
1993-2012
1993-2012
2003-2011

12.9 (NR)

12.4 (Case day) (NR)
12.4 (Case day) (NR)
12.4 (Case day) (NR)
NR (NR)

MCCIure etal., 2017
MCCIure etal., 2017
Kralletal., 2018
Bell et al., 2015
Dominici et al., 2006

REGARDS

ED visit databases

Medicare

2003-2011
2003-2011
2002-2008
1999-2010
1999-2002

NR (NR)

Mean Range: 10.8-15.4
(NR)

12.3 (6.4- 20.2)

13.4 (NR)

Zanobetti et al.,
2009

IHD HA Age 65+

Ml HA Age 65+

Bell etal., 2015

Medicare
Medicare

Dominici et al., 2006 Medicare

Zanobetti et al., .. ..
„„„„ Medicare

2009

2000-2003
1999-2010

1999-2002

2000-2003

15.3 (NR)

12.3 (6.4- 20.2)

13.4 (NR)
15.3 (NR)

0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6
Relative Risk/ Odds Ratio (95% CI) +

CVD (cause-specific) morbidity (Canada)

Exposure Proxy

Endpoint

Health Outcome

Citation

Cohort

Health Data

Reported PM Mean
(Range)(ug/m3)

Monitor

Angina

Angina ED

Szyszkowicz et al., 2009

Hospital Database

1992-2003

8.3 (NR)

Angina/
Ml

Angina/Ml ED

Stieb et al., 2009

Hospital Database

1992-2003

8.2 (6.7-9.8)

Ml ED

Weichenthal et al., 2016b

NACRS

2004-2011

6.9 (NR)

Heart Failure ED

Stieb etal., 2009

Hospital Database

1992-2003

8.2 (6.7-9.8)

0.95 1.00 1.05 1.10 1.15
Relative Risk/ Odds Ratio (95% CI) *

3-87

-------
Respiratory morbidity

Exposure Proxy Country Endpoint Health Outcome Citation

U.S. COPD COPD HA Age 65+ Kloog et al., 2014

Modelled
Monitor

Canada Asthma Asthma ED stieb etal., 2009

Weichenthal et al., 2016c

COPD COPD ED Stieb etal., 2009

Weichenthal et al., 2016c

U.S. Asthma Asthma ED Age 5-18 Alhanti etal., 2016

Asthma ED Age 65+ Alhanti et al., 2016

Asthma HA Age 65+ Bell et al., 2015
Asthma HA Age 1-9:

Central Valley
Asthma HA Age 1-9:
South Coast

Yap et al, 2013
Yap et al, 2013
Asthma ED & HA Malig etal., 2013
Ostro et al., 2016
COPD COPD HA Age 65+ Bell et al., 2015

Dominici et al., 2006
Malig et al., 2013
Ostro etal., 2016

COPD ED & HA

Cohort

Medicare

Hospital Database
NACRS

Hospital Database
Hospital Database
Medicare

Hospital Admissions

Hospital Inpatient and
Outpatient visits
Hospital Inpatient and
Outpatient visits

Medicare
Medicare

Hospital Inpatient and
Outpatient visits
Hospital Inpatient and
Outpatient visits

Health Data

2000-2006

1992-2003
2004-2011

1993-2009
1993-2009

Reported PM Mean
(Range)(ug/m3)

11.9 (NR)
8.2 (6.7-9.8)

7.1 (NR)

8.2 (6.7-9.8)

7.1 (NR)

Mean Range:
11.1-14.1 (NR)
Mean Range:
11.1-14.1 (NR)

1999-2010 12.3 (6.4- 20.2)

2000-2005
2000-2005
2005-2008
2005-2009
1999-2010
1999-2002
2005-2008
2005-2009

Mean Range:
12.8-20.8 (NR)
Mean Range:
14.0-24.6 (NR)
Mean Range:
5.2-19.8 (NR)

16.5 (NR)

12.3 (6.4- 20.2)

13.4 (NR)

Mean Range:
5.2-19.8 (NR)

16.5 (NR)

Relative Risk/ Odds Ratio (95% CI) *

Figure 3-7. Epidemiologic studies examining associations between short-term P1VI2.5
exposures and morbidity.

3-88

-------
• What are the key epidemiologic studies on which the PA should focus for informing
conclusions regarding the adequacy of the current or potential alternative primary
PM2.5 standards? For these key epidemiologic studies, how were the mean PM2.5
concentrations calculated?

Based on the information in Figure 3-4 to Figure 3-7, key epidemiologic studies indicate
generally positive and statistically significant associations between estimated PM2.5 exposures
(short- or long-term) and mortality or morbidity across a range of ambient PM2.5 concentrations.
Drawing from the multicity studies in Figure 3-4 to Figure 3-7, we identify the key
epidemiologic studies most informative to our understanding to evaluate the PM2.5 air quality
distributions in key studies in this reconsideration. Key epidemiologic studies are those that
report overall mean (or median) PM2.5 concentrations and for which the years of PM2.5 air quality
data used to estimate exposures overlap entirely with the years during which health events are
reported. For some studies of long-term PM2.5 exposures, exposure is estimated from air quality
data corresponding to only part of the study period, often including only the later years of the
health data, and are not likely to reflect the full ranges of ambient PM2.5 concentrations that
contributed to reported associations.31 While this approach can be reasonable in the context of an
epidemiologic study that is evaluating health effect associations with long-term PM2.5 exposures,
under the assumption that spatial patterns in PM2.5 concentrations are not appreciably different
during time periods for which air quality information is not available (e.g., Chen et al., 2016),
our interest is in understanding the distribution of ambient PM2.5 concentrations that could have
contributed to reported health outcomes. Therefore, we identify studies as key epidemiologic
studies when the years of air quality data and health data overlap in their entirety.

Additionally, for studies that estimate PM2.5 exposure using hybrid modeling approaches,
we also consider the approach used to estimate PM2.5 concentrations and the approach used to
validate hybrid model predictions when determining those studies that we identify as key
epidemiologic studies. Such studies are identified as those that use hybrid modeling approaches
for which recent methods and models were used (e.g., recent versions and configurations of the
air quality models); studies that are fused with PM2.5 data from national monitoring networks
(i.e., FRM/FEM data); and studies that reported a thorough model performance evaluation for
core years of the study.32 While numerous approaches to estimating PM2.5 concentrations in

31 The following studies do not have an overlap between the years of PM2 5 air quality data and the years during
which health effects are reported: Miller et al., 2007; Hart et al., 2011; Thurston et al., 2013; Weichenthal et al.,
2014; Pope et al., 2015; Villeneuve et al., 2015; Turner et al., 2016; Weichenthal et al., 2016a; Parker et al., 2018;
Pope et al., 2019; and Bevan et al., 2021.

32 The following studies do not meet these criteria: Bravo et al., 2017, Crouse et al., 2015; Puett et al., 2009, Puett et
al., 2011, Hystad et al., 2012; Hystad et al., 2013, Hayes et al., 2020; Elliott et al., 2020; Lefler et al., 2019;;
Pappin et al., 2019; Cakmak et al., 2018; Fisher et al., 2019; Sun et al., 2019; McClure et al., 2017; Loop et al.,
2018 ; and Honda etal., 2017.

3-89

-------
hybrid modeling studies can be reasonable in the context of an epidemiologic study evaluating
health effect associations with PM2.5 exposures (e.g., in studies that use satellite data in fused
surfaces), our interest is in utilizing the most up to date methods based on surfaces fused with
monitored PM2.5 data in order to inform the consideration of the PM NAAQS, as attainment of
the standards is determined based on PM2.5 monitoring data.

While all of the key epidemiologic studies in the 2012 review relied on ground-based
monitoring information to characterize PM2.5 exposure concentrations, as at the time of the 2020
review, a number of the more recent epidemiologic studies in Figure 3-3 to Figure 3-6 utilized
various "hybrid modeling" approaches that include fusion techniques that combine ground-based
monitored data with air quality modeled estimates and/or information from satellites to estimate
PM2.5 exposures. Furthermore, some studies use various mathematical approaches (e.g.,
population weighting, trimmed mean33) to compute the study-reported mean from the estimated
PM2.5 exposure concentrations. The fact that there are more and different techniques utilized to
characterize exposure in the key epidemiologic studies in this reconsideration highlights the
importance of understanding those techniques and how they compare to each other and to
consider how those differences translate into comparisons between the mean PM2.5
concentrations reported in the studies and the level of the primary annual PM2.5 standard.

As noted above, study-reported mean concentrations in Figure 3-3 to Figure 3-6 were
calculated using different methods. This is an important consideration when comparing mean
concentrations across studies, as the methods used to estimate PM2.5 concentrations can vary
from traditional methods using monitoring data from ground-based monitors to those using more
complex hybrid modeling approaches. Studies using hybrid modeling approaches aim to broaden
the spatial coverage of estimated PM2.5 concentrations by bringing in additional information to
provide estimates in areas that do not have ground-based monitors (i.e., areas that are generally
less densely populated and tend to have lower PM2.5 concentrations). As such, the hybrid
modeling approaches tend to broaden the areas captured in the exposure assessment, and in
doing so, the studies that utilize these methods tend to report lower mean PM2.5 concentrations
than monitor-based approaches because they include more suburban and rural areas where
concentrations are lower. Further, other aspects of the method used to calculate mean PM2.5
concentrations can also have an impact on the study-reported mean concentration (i.e.,
population weighting, trim mean).

In those studies that use ground-based monitors alone to estimate long- or short-term
PM2.5 concentrations, approaches include: (1) PM2.5 concentrations from a single monitor within

33 A trimmed mean is a method of averaging that removes a small percentage of the largest and smallest values
before calculating the mean.

3-90

-------
a city/county; (2) average of PM2.5 concentrations across all monitors within a city/county or
other defined study area (e.g., CBSA); or (3) population-weighted averages of exposures. Once
the study location average PM2.5 concentration is calculated, the study-reported long-term
average is derived by averaging daily/annual PM2.5 concentrations across all study locations over
the entire study period. Table 3-6 and Table 3-7 list the key U.S. and Canadian epidemiologic
studies, respectively, that use ground-based monitors to estimate exposure, gives the reported
study mean, and describes the method used to calculate the mean.

3-91

-------
Table 3-6. Key U.S. Epidemiologic Studies: Monitor-Based Exposure

Citation

Health
Endpoint

Geographic Area

Study
Design

Years and Method Used to Calculate Study-reported
Mean PM2.5 Concentrations

Reported Mean
(other percentiles)
|jg/m3

Short-term Exposure Studies

Bell et al.,
2008*

CVD HA
(65+)

202 U.S. Counties
(population >200,000)

Time-series
study

(MEDICARE
enrollees)

Trimmed mean: 1999-2005
Daily PM2 5 concentrations of 202 counties were averaged
to calculate overall mean PM2.5 exposure for the study
location (all and region specific) and study period

12.9

(10th: 9.8, 25th: 11.5)

Bell et al.,
2014

CVD,

asthma,

and

COPD HA
(65+)

4 Counties in MA and CT

Time-series
study

(MEDICARE
enrollees)

2000-2004

Daily PM2.5 concentrations for all four counties (three with

single monitor and one with two monitors that used
population-weighted approach) were used to calculate the
overall mean PM2 5for the study location and period

14.0

Bell et al.,
2015

HF HA
(65+)

213 U.S. Counties

Time-series
study

(MEDICARE
enrollees)

1999-2010

Daily PM2 5 concentrations of 213 counties were averaged
to calculate overall and region-specific mean
PM2.5PM2.5for the study location and period.

12.3

Bravo et al.,
2017

CVD HA
(65+)

418 U.S. Counties
(population >50,000)

Time-series
study

(MEDICARE
enrollees)

2002-2006

Daily PM2 5 concentration of 418 counties were averaged
to calculate overall mean PM2.5for the study location and
period.

12.3

Dai et al., 2014

All-cause,
CVD, and
respiratory
mortality

75 U.S. Cities (available
daily mortality data and
PM2.5 data for at least
400 days 2000-2006)

Time-series
study (NCHS)

2000-2006

Daily PM2.5 concentration of 75 cities were averaged to
calculate overall mean PM2.5for the study location and
period

13.3

Dominici et al.,
2006*

HF and
COPD HA
(65+)

204 Urban U.S. Counties
(population >200,000)

Time-series
study

(MEDICARE
enrollees)

Trimmed mean: 1999-2002
Daily PM2 5 concentrations for 204 US counties were
averaged to calculate overall mean PM2.5 concentration
for the study regions and period.

13.4

3-92

-------
Citation

Health
Endpoint

Geographic Area

Study
Design

Years and Method Used to Calculate Study-reported
Mean PM2.5 Concentrations

Reported Mean
(other percentiles)
|jg/m3

Franklin et al.,
2007*

All-cause,
CVD, and
respiratory
mortality

27 U.S. Communities in
Boston area (with PM2.5
monitoring and daily
mortality data for at least
2 years of 6-year study
period 1997-2000)

Case-
crossover
study (NCHS)

1997-2000

Daily PM2.5 concentrations (from monitors that are highly
correlated in the counties and thus representing general
population exposure) for 27 communities were averaged
to calculate overall mean PM2.5 concentration for the
study location and period.

15.6

(10th: 10.4, 25th: 12.9)

Franklin et al.,
2008*

All-cause,
CVD, and
respiratory
mortality

25 U.S. Communities for
Boston area (with PM2.5
monitoring and daily
mortality data for at least
4 years of 6-year
period 2000-2005)

Case-
crossover
study (NCHS)

2000-2005

Daily PM2.5 concentrations (from monitors that are highly
correlated in the counties and thus representing general
population exposure) for 25 communities were averaged
to calculate overall mean PM2.5 concentration for the
study location and period.

14.8

Klemm and
Mason, 2003 *

All-cause
mortality

Harvard Six-City study
reanalysis

Time-series
study

1979-1988

Daily PM2 5 concentration of six cities were used to
calculate overall mean PM2.5 exposure for the study
location (all and by study center) and period.

Median: 14.7:
(25th: 9.0)

Krall et al.,
2013

All-cause
mortality

72 Urban U.S.
Communities

Time-series
study (NCHS)

2000-2005

Daily PM2.5 concentration (including only the source-
oriented monitors representative of typical population
exposures) of 72 urban communities were used to
calculate overall mean PM2.5 exposure for the study
location and period

13.6

Liu et al., 2019

All-cause

and

cause-

specific

mortality

107 U.S. Cities

Time-series
study (MCC
Collaborative
Research
Network)

1987-2006

Daily PM25 concentration averaged across stations within
each city was used to calculate an average 2-day moving
average PM2.5 concentrations for the city. These data
were then used to calculate overall mean concentration
for the study location and period.

12.4

3-93

-------
Citation

Health
Endpoint

Geographic Area

Study
Design

Years and Method Used to Calculate Study-reported
Mean PM2.5 Concentrations

Reported Mean
(other percentiles)
|jg/m3

Ostro et al.,
2016

Asthma
and

COPD ED

8 Metropolitan
Areas/Counties in CA

Case-

crossover

study

2005-2009

Daily PM2.5 concentrations for eight metropolitan counties
were used to calculate an overall mean PM2.5
concentration for the study location and period.

16.5

Peng et al.,
2009

CVD HA
(65+)

119 U.S. Urban
Counties>150,000
populations

Time-series
study

(MEDICARE
enrollees)

2000-2006

Daily PM2 5 concentrations for 119 counties were used to
calculate an overall median PM2.5 concentration for the
study location and period.

Median: 11.9

Zanobetti et
al., 2009

CVD, HF,
Ml HA
(65+)

26 U.S. Cities

Time-series
study

(MEDICARE
enrollees)

2000-2003

Daily average PM2.5 data for each county was calculated
using an algorithm that accounts for monitor-specific
means and variances. Monitors that were not well
correlated with other monitors were excluded.

15.3

Zanobetti and

Schwartz,

2009*

All-cause,
CVD and
respiratory
mortality

112 U.S. Cities

Time-series
study (NCHS)

1999-2005

Daily PM2.5 concentrations (from monitors that are highly
correlated in the counties and thus representing general

population exposure) for 112 cities were averaged to
calculate overall mean PM2.5 concentration for the study
location and period.

13.2

(10th: 10.3, 25th: 12.5)

Long-term Exposure Studies

Eum et al.,
2018

All-cause
mortality

U.S. Geographic regions:
"East" of the Mississippi
River, "Center" between
the Mississippi River and
the Sierra Nevada
mountain range, and
"West" of the Sierra
Nevada mountain range

Cohort study
(MEDICARE
enrollees)

2000-2012

Annual average PM2.5 concentrations assigned to
individuals living in ZIP codes with centroids within 6 miles
of a valid monitor (monitors with daily measurements for

at least 8 calendar years, with each year having 9+
months, and with 4+ daily measurements) were used to
calculate overall mean PM2.5 concentration for the study
location (all and by study region) and study period.

Overall: 11.65
Central: 9.9
Eastern: 12.3
West: 11.5

3-94

-------
Citation

Health
Endpoint

Geographic Area

Study
Design

Years and Method Used to Calculate Study-reported
Mean PM2.5 Concentrations

Reported Mean
(other percentiles)
|jg/m3

Gharibvand et
al., 2016

Lung

cancer

incidence

U.S. Nationwide

Cohort study
(AHSMOG-2
study)

2000-2001

Monthly PM2.5 concentrations (calculated using at least
75% valid daily data) assigned to study participants based
on residential address were used to calculate overall
mean PM2.5 for the study period.

12.9

Hart et al.,
2015

All-cause
mortality

U.S. Nationwide

Cohort study
(Nurses'
Health study)

2000-2012

Monthly PM2.5 concentrations assigned to study
participants based on the nearest monitor to residence
locations were used to calculate overall mean for the
study period

12.7

Kioumourtzogl
ou et al., 2016

All-cause

mortality

(65+)

207 U.S. cities

Cohort study
(MEDICARE
enrollees)

2000-2010

Annual PM2.5 concentrations for 207 cities were averaged
to calculate overall mean PM2.5 exposure for the study
location (all and region specific) and study period.

12.0

McConnell et
al., 2010

Asthma
Incidence

13CA Communities

Cohort study
(CHS)

2003-2004

Average annual PM2.5 concentrations assigned to study
participants based on their community of residence were
used to calculate overall mean PM2.5 exposure for the
study location and period.

13.9

Zeger et al.,
2008*

All-cause

mortality

65+

668 U.S. Urban Counties

Cohort Study
of

MEDICARE

enrollees

(MCAPS)

2000-2005

Average annual PM2.5 concentrations of ZIP codes
(for ZIP code centroids within 6 miles of a monitor and
with >10 months of data per year) were used to calculate
overall mean PM2.5 exposure for the study location (all
and by region) and the study period.

Central Region
median: 10.7
Eastern Region
median: 14.0
Western region
median: 13.1

* Evaluated in 2012 review

3-95

-------
Table 3-7. Key Canadian Epidemiologic Studies: Monitor-Based Exposure

Citation

Health
Endpoint

Geographic
Area

Study Design

Years and Method Used to Calculate Study-reported Mean
PM2.5 Concentrations

Reported Mean
(other percentiles)
|jg/m3

Short-term Exposure Studies

Burnett and

Goldberg,

2003*

All-cause
mortality

8 Canadian
cities

Time-series
study

1986-1996

Daily PM25 concentrations (day before the death) for 8 Canadian
cities were averaged to get overall mean for the study area and

period

13.3

Burnett et al.,
2004*

All-cause
mortality

12 Canadian
cities

Time-series
study (data from
Statistics
Canada)

1981-1999

PM2 sDaily PM2.5 concentrations for 12 cities (calculated by
averaging all monitors within each city) were used along with
population information to calculate an overall population-weighted
PM2 5 concentration for the study location and period

12.8

Lavigne et al.,
2018

Non-
accidental,
CVD, and
respiratory
mortality

24 Canadian
cities

Case-crossover
study

1998-2011

Daily average PM2.5 concentrations assigned to participants based
on closest monitor(s) to participant's city of residence. Daily PM2.5
concentrations in 24 Canadian cities were used to calculate overall
mean PM2.5 concentration over the study location and period.

8.8

(Median: 7.1)

Liu et al., 2019

All-cause

and
cause-
specific
mortality

25 Canadian
cities

Time-series
Study (MCC
Collaborative
Research
Network)

1986-2011

PM2 5 concentration averaged across stations within each city was

used to calculate an average 2-day moving average PM2.5
concentrations for the city. These data were then used to calculate
overall mean concentration for the study location and period.

9.3

Stieb et al.,
2009

Cardiac

and
respiratory
ED visits

7 Canadian
cities

Time-series
study (Hospital
cases)

1992-2003

Daily PM2 5 concentrations of the cities (calculated by averaging all
monitors within city) were used to calculate the overall mean PM2.5
exposure for the study location (by site) and study period.

8.2

(10th: 6.7, 25th: 6.8)

Szyszkowicz,
2009

Angina ED

7 Canadian
cities

Time-series
study (Hospital
cases)

1992-2003

Daily PM2 5 concentrations of the cities (calculated by averaging all
monitors within city) were used to calculate the overall mean PM2.5
exposure for the study location (all and by cities) and study period.

8.3

(10th: 6.4, 25th: 6.5)

3-96

-------
Citation

Health
Endpoint

Geographic
Area

Study Design

Years and Method Used to Calculate Study-reported Mean
PM2.5 Concentrations

Reported Mean
(other percentiles)
|jg/m3

Weichenthal et
al., 2016b

Ml ED

16 cities in
Ontario

Case-crossover
Design (cases
extracted from
NACRS
database)

2004-2011

Daily PM2 5 concentrations in Ontario were used to calculate the
overall mean PM2.5 exposure for the study location and period

6.9

Weichenthal et
al., 2016c

Asthma

and
COPD ED

15 cities in
Ontario

Case-crossover
design (cases
extracted from
NACRS
database)

2004-2011

Daily PM2 5 concentrations in Ontario were used to calculate the
overall mean PM2.5 exposure for the study location and period.

Asthma: 7.1
COPD: 7.1

Long-term Exposure Studies

Crouse et al.,
2012

All-cause
mortality

11 Canadian
Cities

Cohort study

1987-2001

Annual PM2.5 concentrations from monitors and assigned to
study participants based on the census division of the residence
were used to calculate overall mean PM2.5 for the study population
and duration.

8.7

* Evaluated in 2012 review

3-97

-------
In the studies that use hybrid modeling approaches to estimate long- or short-term PM2.5
concentrations, data can be incorporated from several different sources, including satellites and
air quality models, in addition to ground-based monitors, as described in section 2.3.3 above.
Compared to ground-based monitors alone, hybrid modeling methods have the potential to
improve the characterization of PM2.5 concentrations in areas with relatively sparse monitoring
networks. These approaches also tend to have lower study-reported mean PM2.5 concentrations
since they often include estimates of PM2.5 concentrations in less populated areas compared to
those methods using only ground-based monitored. Studies that use hybrid modeling approaches
can estimate PM2.5 concentrations at different spatial resolutions, including at 1 km x 1 km grid
cells (i.e., Di et al., 2017b and Di et al., 2017a), at 10 km x 10 km grid cells (i.e., Kloog et al.,
2014), or at the census tract level (i.e., Bravo et al., 2017). Estimated PM2.5 concentrations are
then generally averaged up to a larger spatial resolution that corresponds to the spatial resolution
for which health data exists (e.g., ZIP code level). These values are then averaged across all
study locations at the larger spatial resolution (e.g., averaged across all ZIP codes in the study)
over the study period, resulting in the study-reported mean 24-hour average or annual average
PM2.5 concentration. Table 3-8 and Table 3-9 list the key U.S. and Canadian epidemiologic
studies, respectively, that use hybrid modeling approaches to estimate exposure and give the
reported study mean and describes the method used to calculate the mean. Studies included in
these tables are those that report overall mean (or median) PM2.5 concentrations and for which
the years of PM2.5 air quality data used to estimate exposures overlap entirely with the years
during which health events are reported.34 In addition, studies included in Table 3-8 and Table 3-
9 are those for which recent methods and models were used (e.g., recent versions and
configurations of the air quality models); studies that are fused with PM2.5 data from national
monitoring networks (i.e., FRM/FEM data); and studies that reported a thorough model
performance evaluation for core years of the study.

34 In addition to the study-reported mean concentrations, the 10th and 25th percentiles are also included in the tables
when reported by the study.

3-98

-------
Table 3-8. Key U.S. Epidemiologic Studies: Model-Based Exposure

Citation1

Health
Endpoint

Geographic
Area

Study Design

Years, Model Type, and Method Used to Calculate
Study-reported Mean PM2.5 Concentrations

Reported
Mean
(other
percentiles)
Mg/m3

Short-term Exposure Studies

deSouza et al., 2021

First CVD HA

Continental
U.S.

Time-stratified
case-
crossover

design
(Medicaid
Adults)

2000-2012

Ensemble model (integrating machine learning
algorithms)

Daily PM25 estimates of grid cells averaged at ZIP
code were assigned to study participants based on the
ZIP code of residence. Daily PM2.5 concentration from
case days were used to calculate overall case day
mean PM2.5 concentration for the study location and
period.

11.5 (case days
mean)

Di et al., 2017a

All-cause
mortality (65+)

U.S.
Nationwide

Case-
crossover

study
(MEDICARE
enrollees)

2000-2012
Artificial Neural Network (Hybrid method)

Daily PM2 5 concentrations for case and control days

assigned to participants based on ZIP code of
residence were used to calculate overall mean PM2.5
for the study location and period.

11.6
(10th: 4.7, 25th:
6.7)

Kloog et al., 2012

CVD HA (65+)

New
England
Area with 6
U.S. States

Mixed study
design (with
time series and

cohort
components)

2000-2006
Spatiotemporal model

Daily PM2.5 concentration of all grids within the NE
area for acute exposure (0 day lag) were used
to calculate overall mean for short-term PM2.5
exposure, for the study location and period.

9.6
(25th: 6.4)

3-99

-------
Citation1

Health
Endpoint

Geographic
Area

Study Design

Years, Model Type, and Method Used to Calculate
Study-reported Mean PM2.5 Concentrations

Reported
Mean
(other
percentiles)
|jg/m3

Kloog et al., 2014

CVD and
COPD HA
(65+)

7 U.S. Mid-

Atlantic
States and
D.C.

Case-
crossover

design
(MEDICARE
enrollees)

2000-2006
Spatiotemporal model

2-day moving average of PM2.5 concentration of all

grids within the mid-Atlantic states were used
to calculate overall mean (all area and rural/urban
areas) PM2.5 exposure for the study location and
period.

11.9
(25th: 7.9)

Lee et al., 2015

All-cause,
cardiovascular
, respiratory
mortality

3 U.S.
Southeast
States

Case-
crossover
design (Dept.
of Pub Health
data)

2007-2011
Spatiotemporal model PM2.5

Daily PM2 5 concentrations for ZIP codes in the 3
Southeastern states were averaged to calculate the
overall mean PM2.5 concentration (all states and by
state).

11.1

Shi et al., 2016

Total mortality
(65+)

New
England
Area with 6
U.S. States

Open Cohort

study
(MEDICARE
enrollees)

2003-2008
Predicted from 3-stage statistical model

Lag01 PM2.5 concentrations of all grid cells in the
study area were used to calculate overall mean PM2.5
exposure for the study location and period.

8.2
(25th: 4.6)

3-100

-------
Citation1

Health
Endpoint

Geographic
Area

Study Design

Years, Model Type, and Method Used to Calculate
Study-reported Mean PM2.5 Concentrations

Reported
Mean
(other
percentiles)
|jg/m3

2008-2014
Spatiotemporal prediction model

Wyattet al., 2020c

All-cause,
CVD, RD 30-
day hospital
readmissions

530 U.S.
counties

Cohort study

(USRDS
hemodialysis
patients)

Daily PM2.5 concentrations for grid cells were
converted to population-weighted county-level PM2.5

estimates using 2010 census tract population
estimates. Participants were assigned daily PM2.5
based on the county of their last dialysis visit. Daily
estimates at county-level were then used to calculate
overall PM2.5 concentration for the study location and
period.

9.29

Long-term Exposure Studies

Diet al., 2017b

All-cause
mortality (65+)

U.S.
Nationwide

Cohort study
(MEDICARE
enrollees)

2000-2012
Artificial Neural Network (Hybrid method)

Average PM2.5 concentrations over all ZIP codes in
the continental US were used to calculate overall
mean PM2.5 for the study location and period.

11.0
(10th: 7.3, 25th:
9.1)

Dominici et al., 2019

All-cause
mortality (65+)

U.S.
Nationwide

Cohort study
(MEDICARE
enrollees)

2000-2012
Artificial Neural Network (Hybrid method)

Average PM2.5 concentrations for all ZIP codes in the
continental US were used to calculate overall mean
PM2.5 for the study location and period.

11.0

3-101

-------
Citation1

Health
Endpoint

Geographic
Area

Study Design

Years, Model Type, and Method Used to Calculate
Study-reported Mean PM2.5 Concentrations

Reported
Mean
(other
percentiles)
|jg/m3

2000-2012
Spatiotemporal model

Hart et al., 2015

All-cause
mortality

U.S.
Nationwide

Cohort study
(Nurses' Health
study)

Monthly PM2.5 concentrations for the previous 12-
months that were assigned to study participants at
residence locations during the study follow-up period
were used to calculate overall mean for the
participants included in the study.

12.0

Kloog et al., 2012

CVD HA (65+)

New
England
Area with 6
U.S. States

Mixed study
design (with
time series and

cohort
components)

2000-2006
Spatiotemporal model

Daily PM2.5 concentration of all grids within the NE
area for chronic exposure (365 day moving average)
were used to calculate overall mean for long-term
PM25 exposure, for the study location and period.

9.7
(25th: 9.2)

Shi et al., 2016

Total mortality
(65+)

New
England
Area with 6
U.S. States

Open Cohort

study
(MEDICARE
enrollees)

2003-2008
Predicted from 3-stage statistical model

Average annual PM2.5 concentrations of all grid cells in
the study area were used to calculate overall mean
PM25 exposure for the study location and period.

8.1 (25th: 6.2)

Thurston et al., 2016

All-cause,
CVD and
respiratory
mortality

6 U.S.
States and
2 MS As

Cohort study
(NIH_AARP
cohort)

2000-2008
Spatiotemporal model

Average annual PM2.5 concentrations in the prior year
at the census tract of residence over the follow-up
period were used to calculate overall mean PM2.5
exposure for the study participants.

12.2
Mean range:
2.9-28.0

3-102

-------
Citation1

Health
Endpoint

Geographic
Area

Study Design

Years, Model Type, and Method Used to Calculate
Study-reported Mean PM2.5 Concentrations

Reported
Mean
(other
percentiles)
|jg/m3

Wang et al., 2017

Total mortality
(65+)

7 U.S.
Southeast
States

Cohort study
(MEDICARE
enrollees)

2000-2013
Three stage Hybrid model PM2.5

Average annual PM2.5 concentrations of ZIP code
tabulation areas were calculated by averaging annual
mean PM2.5 concentration of all grids in the ZCTA and
then used to calculate overall median PM2.5 exposure
for the study location (overall and by state), and period
(overall and by year).

Median: 10.7
Range: 6.0-20.6
(25th: 9.1)

Wang et al., 2020

Non-
accidental
cause-specific
(respiratory,
CVD, cancer)
mortality

U.S.
Nationwide

Cohort study
(MEDICARE)

2000-2008
Spatiotemporal prediction model

Daily PM2.5 concentrations of grids were matched to
study participants based on the grid point closest to
their residential ZIP code centroid. The estimates
were used to calculate overall annual mean PM2.5
exposure for the study period.

10.3

Wu et al., 2020

All-cause
mortality

U.S.
Nationwide

Cohort study
(MEDICARE)

2000-2016

Ensemble model (integrating machine learning
algorithms)

Daily PM2.5 concentration at grid cells whose centroids
were inside the ZIP code boundary were averaged for
each year and assigned to participants based on the
ZIP code of residence. These data were used to
calculate overall mean PM2.5 concentration for the
study period.PM2.5PM2.5

9.8
(<12 ug/m3:
8.4)

1 None of the studies presented in this table were evaluated in the 2012 review.

3-103

-------
Table 3-9. Key Canadian Epidemiologic Studies: Model-Based Exposure

Citation1

Health
Endpoint

Geographic
Area

Study
Design

Years, Model Type, and Method Used to Calculate Study-reported
Mean PM2.5 Concentrations

Reported Mean
(other percentiles)
|jg/m3

Long-term Exposure Studies

Bai et al.,
2019

CHF and

AMI
incidence

Ontario

Cohort study
(ONPHEC)

1998-2012

Fused surface (AOD, GEOS-Chem & geographically weighted regression)

Annual estimates of PM2.5 concentrations assigned to participants based on
postal code of residence used to calculate 3-year moving average PM2.5
concentration for each year of follow-up in the study. The 3-year moving
averages for study participants at the baseline residence location was used
to calculate overall mean PM2.5 concentration at the beginning of the follow-

up period in 2001.

9.6
(25th: 7.9)

Chen et al.,
2020

CVD
mortality
and AMI
incidence

Ontario

Cohort study
(ONPHEC)

2000-2016

Fused surface (AOD, GEOS-Chem & geographically weighted regression)

Annual estimates of PM2.5 concentrations were assigned to participants
based on postal code of residence. Annual PM2.5 concentrations across all
postal code areas in the Ontario region were then used to calculate overall
mean PM2.5 concentration for the study location and period.

8.61

Christidis et
al., 2019

Non-
accidental
mortality

Canada
Nationwide

Cohort study
(mCHHS)

1998-2015

Fused surface (AOD, GEOS-Chem & geographically weighted regression)

Annual PM2.5 estimates assigned to study participants based on the postal
code for residence were used to calculate 3-year moving average based on
the location and year of follow-up. The average PM2.5 concentrations were
then used to calculate overall mean PM2.5 concentration for the study

period.

5.9

(Median: 5.5; 25th:
4.3)

3-104

-------
Citation1

Health
Endpoint

Geographic
Area

Study
Design

Years, Model Type, and Method Used to Calculate Study-reported
Mean PM2.5 Concentrations

Reported Mean
(other percentiles)
|jg/m3

Crouse et al.,
2020

Non-
accidental,

CVD,
respiratory
mortality
and lung
cancer

Canada
Nationwide

Cohort study
(CanCHEC)

1998-2010

Fused surface (AOD, GEOS-Chem & geographically weighted regression)

Annual PM2.5 estimates assigned to study participants based on the postal
code for residence were used to calculate moving average at various
temporal and spatial scales based on the location and year of follow-up.
The moving average PM2.5 concentrations at the baseline residence location
were then used to calculate overall mean PM2.5 concentration at the
beginning of the follow-up period in 2001 at various temporal and spatial

scales.

1-year in 1 km:
Mean: 7.2,

3-year in 1 km:
Mean: 7.4,

8-year in 1 km:
Mean: 8.0

Erickson et
al., 2020

Non-
accidental,
CVD, and
respiratory
mortality
and lung
cancer
mortality

Canada
Nationwide

Cohort study
(CanCHEC)

1998-2016

Fused surface (AOD, GEOS-Chem & geographically weighted regression)

Annual PM2.5 estimates assigned to study participants based on the postal
code for residence were used to calculate 3-year moving average based on
the location and year of follow-up. The average annual PM2.5 concentrations
of the study participants were then used to calculate overall mean PM2.5
concentration for the study period by immigrant status and duration in

Canada.

Non-immigrant: 7.5
Immigrant: 9.3
Pre-1971: 9.1
1971-1980: 9.3
1981-1990: 9.5
1991-2001:9.7

Erickson et
al., 2019

All-cause,
CVD,,
and lung
cancer
mortality

Canada
Nationwide

CanCHEC
(Primary
data), CCHS
(Ancillary
data),
mCCHS
(Additional
data)

1998-2012

Fused surface (AOD, GEOS-Chem & geographically weighted regression)

Annual PM2.5 estimates assigned to study participants based on the postal
code for residence were used to calculate 3-year moving average based on
the location and year of follow-up. The average annual PM2.5 concentrations
of the study participants were then used to calculate overall mean PM2.5
concentration for the study period.

CanCHEC: 8.4,
CCHS: 6.7

3-105

-------
Citation1

Health
Endpoint

Geographic
Area

Study
Design

Years, Model Type, and Method Used to Calculate Study-reported
Mean PM2.5 Concentrations

Reported Mean
(other percentiles)
|jg/m3

Pinault et al.,
2016

All-cause,
CVD and
lung
cancer
mortality

Canada
Nationwide

Cohort
Study
(mCCHS)

1998-2012

Fused surface (AOD, GEOS-Chem & geographically weighted regression)

Annual PM2.5 estimates assigned to study participants based on the postal
code for residence were used to calculate 3-year moving average based on
the location and year of follow-up. The average annual PM2.5 concentrations
of the study participants were then used to calculate overall mean PM2.5
concentration for the study period.

6.3

Pinault et al.
2017

All-cause,

CVD,
Respirator
y, and
lung
cancer
mortality

Canada
Nationwide

Cohort study
(CanCHEC)

1998-2012

Fused surface (AOD, GEOS-Chem & geographically weighted regression)

Annual PM2.5 estimates assigned to study participants based on the postal
code for residence were used to calculate 3-year moving average based on
the location and year of follow-up. The average PM2.5 concentrations of the
study participants were then used to calculate overall mean PM2.5
concentration for the study period.

7.4

Pinault et al.,
2018

CVD
mortality

Canada
Nationwide

Cohort study
(CanCHEC,
mCCHS)

1998-2012

Fused surface (AOD, GEOS-Chem & geographically weighted regression)

Annual PM2.5 estimates assigned to study participants based on the postal
code for residence were used to calculate 3-year moving average based on
the location and year of follow-up. The average PM2.5 concentrations of the
study participants were then used to calculate overall mean PM2.5
concentration for the study period.

CanCHEC: 7.4
mCHHS: 6.4

Shin et al.,
2019

AF and
Stroke
Incidence
(1st HA)

Ontario

Cohort study
(ONPHEC)

1998-2012

Fused surface (AOD, GEOS-Chem & geographically weighted regression)

Annual PM2.5 estimates assigned to study participants based on the postal
code for residence were used to calculate 5-year moving average based on
the location and year of follow-up. The average PM2.5 concentrations were
then used to calculate overall mean PM2.5 concentration for the study

period.

9.8
(25th: 8)

3-106

-------
Citation1

Health
Endpoint

Geographic
Area

Study
Design

Years, Model Type, and Method Used to Calculate Study-reported
Mean PM2.5 Concentrations

Reported Mean
(other percentiles)
|jg/m3

Zhang et al.,
2021

Non-
accidental,
CVD, and
respiratory
mortality

Ontario

Cohort study
(Ontario
Health
Study)

Modeled from AOD satellite retrievals
2000-2016

Annual PM2.5 estimates assigned to study participants based on the postal

code for residence was used to calculate 3-year and 5-year moving
averages based on the location and year of follow-up. The 5-year average
PM2.5 concentrations were then used to calculate overall mean PM2.5
concentration for the baseline year.

Baseline: 7.8
(Median: 8.0; 25th:
6.7)

1 None of the studies presented in this table were evaluated in the 2012 review.

3-107

-------
To further expand our evaluation of study-reported mean PM2.5 concentrations, we
specifically consider the following questions:

• What are the overall mean PM2.5 concentrations reported by key epidemiologic
studies? For studies with available information on the broader distributions of
exposure estimates and/or health events, what are the PM2.5 concentrations
corresponding to the lower percentiles of those data (e.g., 25th and/or 10th)?

Figure 3-8 and Figure 3-9 highlight the overall mean (or median) PM2.5 concentrations
reported in key U.S. and Canadian studies, respectively, that use ground-based monitors alone to
estimate long- or short-term PM2.5 exposures. For the small subset of studies with available
information on the broader distributions of underlying data, Figure 3-8 and Figure 3-9 also
identify the study-period mean PM2.5 concentrations corresponding to the 25th and 10th
percentiles of health events35 (see Appendix B, Section B.2 for more information).

Figure 3-10 and Figure 3-11 present overall means of predicted PM2.5 concentrations for
key U.S. and Canadian model-based epidemiologic studies, respectively, and the concentrations
corresponding to the 25th and 10th percentiles of estimated exposures or health events36 when
available (see Appendix B, section B.3 for additional information).

35 That is, 25% of the total health events occurred in study locations with mean PM2 5 concentrations (i.e., averaged
over the study period) below the 25th percentiles identified in Figure 3-8 and Figure 3-9 and 10% of the total
health events occurred in study locations with mean PM2 5 concentrations below the 10th percentiles identified.

36 For most studies in Figure 3-10 and Figure 3-11, 25th percentiles of exposure estimates are presented. The
exception is Di et al. (2017b), for which Figure 3-10 presents the short-term PM2 5 exposure estimates
corresponding to the 25th and 10th percentiles of deaths in the study population (i.e., 25% and 10% of deaths
occurred at concentrations below these concentrations). In addition, the authors of Di et al. (2017b) provided
population-weighted exposure values (Chan, 2019). The 10th and 25th percentiles of these population-weighted
exposure estimates are 7.9 and 9.5 ng/m3, respectively.

3-108

-------
ST Exposure
and Mortality

ST Exposure
and Morbidity

LT Exposure
and Mortality

LT Exposure
and Morbidity

Franklin 2007 (US: 27 Cities)

Franklin 2008 (US: 25 Cities)

Klemm 2003 (US: Harvard 6 City)

Krall 2013 (US: 72 Cities)

Dai 2014 (US: 75 Cities)

Zanobetti and Schwartz 2009 (US: 112 Cities)

Liu 2019 (US: 107 cities; ST Exposure)*

Ostro 2016 (US: 8 California Counties)

Zanobetti 2009 (US: 26 cities)

Bell 2014 (US: 4 Counties in MA & CT)

Dominici 2006 (US: 204 Urban Counties)

Bell 2008 (US: 202 Counties)

Bravo 2017 (US: 418 Counties)

Bell 2015 (US: 70 Urban Counties)

Peng 2009 (US: 119 Urban Counties)

Zeger 2008 (US: 421 Eastern Region Counties)

Zeger 2008 (US: 62 Western Region Counties)

Hart 2015 (US: Nationwide)

Eum 2018 (US: Eastern Geographic Region; LT Exposure)*
Kioumourtzoglou 2016 (US: 207 Cities)

Eum 2018 (US: 3 Geographic Regions overall; LT Exposure)*
Eum 2018 (US: Western Geographic Region; LT Exposure)*
Zeger 2008 (US: 185 Central Region Counties)

Eum 2018 (US: Central Geographic Region; LT Exposure)*
McConnell 2010 (US: 13 California Communities)
Gharibvand 2016 (US: Nationwide)

Study Type

| ST Exposure and Mortality
| ST Exposure a ad Morbidity
| LT Exposure and Mortality
| LT Exposure and Morbidity

Summary Statistics
O 10th percentile
£ 25th percentile
¦ Mean

9.5 10.0 10.5 11.0

11.5 12.0 12.5 13.0 13.5 14.0 14.5 15.0
Overall PM2.5 Concentration for the Study Period (ng/m3)

15.5 16.0 16.5

Figure 3-8. Monitor-based PM2.5 concentrations in key U.S. epidemiologic studies. (Asterisks denote studies included in the ISA
Supplement).

3-109

-------
ST Exposure and
Mortality

Burnett 2003 (Canada: 8 Cities)

Burnett 2004 (Canada: 12 Cities)

¦ NS

Liu 2019 (Canada: 25 cities; ST Exposure)*

Lavigne 2018 (Canada: 24 Cities; ST Exposure)*

ST Exposure and
Morbidity

Szyszkowicz 2009 (Canada: 6 Cities)

C* ¦

Stieb 2009 (Canada: 6 Cities)

0# ¦

Weichenthal 2016c (Canada: 15 Ontario Cities)

Weichenthal 2016b (Canada: 16 Ontario Cities)

¦ NS

LT Exposure and
Mortality

Crouse 2012 (Canada: 11 Cities)

6.5 7.0 7.5 8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0 13.5

Overall PM2.5 Concentration for the Study Period (tig/m3)

Study Type

| ST Exposure and Mortality

ST Exposure and Morbidity

| LT Exposure and Mortality

Summary Statistics
O 10th percentile
£ J5th percentile
¦ Mean

Figure 3-9. Monitor-based PM2.5 concentrations in key Canadian epidemiologic studies. (Asterisks denote studies included in the
ISA Supplement).

3-110

-------
ST Exposure & Mortality

Di 2017a (US: Nationwide; ST Exposure)

O • ¦

Lee 2015b (US: 3 SE States; ST Exposure)

Shi 2016 (US: 6 NE States; ST Exposure)

• ¦

ST exposure & Morbidity

Kloog 2014 (US: 7 Mid-Atlantic States and D.C.; ST Exposure)

• ¦

deSouza 2021 (US: Nationwide; ST Exposure)*

Kloog 2012 (US: 6 NE States; ST Exposure)

• ¦

Wyatt 2020c (US: 530 US counties; ST Exposure)*

LT exposure & Mortality

Thurston 2016 (US: 6 States and 2 MSAs; LT Exposure)

Hart 2015 (US: Nationwide; LT Exposure)

Di 2017b (US: Nationwide; LT Exposure)

O • ¦

Dominici 2019 (US: Nationwide; LT Exposure)*

Wang 2017 (US: 7 SE States; LT Exposure)

• ¦

Wang 2020 (US: Nationwide; LT Exposure)*

Wu 2020a (US: Nationwide; LT Exposure)*

Shi 2016 (US: 6 NE States; LT Exposure)

• ¦

LT Exposure & Morbidity

Kloog 2012 (US: 6 NE States; LT Exposure)

• ¦

5 6 7 8 9 10 11 12
Overall PM2.5 Concentration for the Study Period (pg/m3)

Study Type

¦ ST Exposure & Mortality

¦ ST exposure & Morbidity

¦ LTexposure & Mortality
I LT Exposure & Morbidity

Summary Statistics

O 10th percentile
• 25th percentile

¦ Mean

Figure 3-10. Hybrid model-predicted PM2.5 concentrations in key U.S. epidemiologic studies. (Asterisks denote studies included
in the ISA Supplement).

3-111

-------
LT exposure & Mortality

LT Exposure a Morbidity

Erickson 2020 (Canada: Nationwide; LT Exposure; Immigrants 10 years)*

Erickson 2020 (Canada: Nationwide; LT Exposure; immigrants: 11-20 years)*

Erickson 2020 (Canada: Nationwide; LT Exposure; immigrants: 21-30 yea rs)*

Erickson 2020 (Canada: Nationwide; LTExposure;lmmigrant >30 years)*

Chen 2020 (Canada: Ontario; LT Exposure)*

Erickson 2019 (Canada: Nationwide; LT exposure; CanCHEC)*

Crouse 2020 (Canada: Nationwide; LT Exposure; 8-year)*

Zhang 2021 (Canada: Ontario; LT Exposure)*

Erickson 2020 (Canada: Nationwide; LT Exposure; Non-immigrant)*

Crouse 2020 (Canada: Nationwide; LT Exposure; 3-year)*

Pinauit2017 (Canada: Nationwide; LT exposure)*

Pinauit 2018 (Canada: Nationwide; LT Exposure; CanCHEC)*

Crouse 2020 (Canada: Nationwide; LT Exposure; 1-year)*

Erickson 2019 (Canada: Nationwide; LTexposure; mCCHS}*

Pinauit 2018 (Canada: Nationwide; LT Exposure; mCHHS)*

Pinauit 2016 (Canada: Nationwide; LT Exposure)

Christidis 2019 (Canada: Nationwide; LT Exposure)*

Shin 2019 (Canada: Ontario; LT Exposure)*

Bai 2019 (Canada: Ontario; LT Exposure)*

Chen 2020 (Canada: Ontario; LT Exposure)*

Study Type

¦ LT exposure & Mortality

¦ LT Exposure & Morbidity

Summary Statistics

• 25th percentile

¦ Mean

5 6 7 8 9 10

Overall PM2.5 Concentration for the Study Period ftig/m3)

Figure 3-11. Hybrid model-predicted PM2.5 concentrations in key Canadian epidemiologic studies. (Asterisks denote studies
included in the ISA Supplement).

3-112

-------
In further examining these data, we also ask:

• For the key epidemiologic studies using hybrid modeling approaches, what are the
study-reported means for the general categories of methods of calculating the study
mean and how do the study-reported means vary and compare to each other?

Figure 3-12 and Figure 3-13 present the same key model-based epidemiologic studies
from the figures above but focus on the U.S. studies and group them based on their approach to
estimating PM2.5 concentrations, including the study-reported mean. For Figure 3-12, the studies
are grouped by the geographical spatial scale at which the modeling was conducted (i.e.,
nationwide, regional, rural). Figure 3-13 presents the same key U.S. model-based epidemiologic
studies, but subset by the method used to average grid cells in study-reported long-term mean
PM2.5 concentrations. For the key U.S. model-based epidemiologic studies, the various methods
include the average of all grid cells; grid cells averaged up to ZIP code, postal code or census
tract; or population-weighted grid cell averaged up to ZIP code or census tract. Lastly, Figure 3-
14 subsets the key U.S. epidemiologic studies that used hybrid exposure models by both spatial
scale and the method used to average grid cells in study-reported long-term mean PM2.5
concentrations. Grouping the key epidemiologic studies in such ways allows for visual
comparisons of the study-reported mean PM2.5 concentrations across the different spatial scales
and methods of averaging the grid cells.

3-113

-------
Nationwide US Hart 2015 (US: Nationwide; LT Exposure)

Di 2017a (US: Nationwide; ST Exposure)
deSouza 2021 (US: Nationwide; ST Exposure)*

Di 2017b (US: Nationwide; LT Exposure)

Dominici 2019 (US: Nationwide; LT Exposure)*

Wang 2020 (US: Nationwide; LT Exposure)*

Wu 2020a (US: Nationwide; LT Exposure)*

Regional US Thurston 2016 (US: 6 States and 2 MSAs; LT Exposure)

Kloog 2014 (US: 7 Mid-Atlantic States and D.C.; ST Exposure)

Lee 2015b (US: 3 SE States; ST Exposure)

Wang 2017 (US: 7 SE States; LT Exposure)

Kloog 2012 (US: 6 NE States; LT Exposure)

Kloog 2012 (US: 6 NE States; ST Exposure)

Wyatt 2020c (US: 530 US counties; ST Exposure)*

Shi 2016 (US: 6 NE States; ST Exposure)

Shi 2016 (US: 6 NE States; LT Exposure)

6 7 8 9 10 11

Overall PM2.5 Concentration forthe Study Period (pg/m3)

Study Type

¦ ST Exposure & Mortality

¦ ST exposure a Morbidity

¦ LT exposure & Mortality

¦ LT Exposure & Morbidity

Summary Statistics

O 10th percentile
• 25th percentile

¦ Mean

Figure 3-12. Hybrid model-predicted PM2.5 concentrations in key U.S. epidemiologic studies, subset by spatial scale. (Asterisks
denote studies included in the ISA Supplement).

3-114

-------
Avg of all grid cells

Grid cell averaged up to zip, postal code, US
or census tract

Population weighted grid cell averaged
up to census tract or zip code

Kloog 2014 (US: 7 Mid-Atlantic States and D.C.; ST
Exposure)

Kloog 2012 (US: 6 NE States; LT Exposure)

Kloog 2012 (US: 6IME States; ST Exposure)
Shi 2016 (US: 6 NE States; ST Exposure)

Shi 2016 (US: 6 NE States; LT Exposure)

Thurston 2016 (US: 6 States and 2 MSAs; LT
Exposure)

Hart 2015 (US: Nationwide; LT Exposure)

Di 2017a (US: Nationwide; ST Exposure)
deSouza 2021 (US: Nationwide; ST Exposure)*
Lee 2015b (US: 3 SE States; ST Exposure)

Di 2017b (US; Nationwide; LT Exposure)

Dominici 2019 (US: Nationwide; LT Exposure)*
Wang 2017 (US: 7 SE States; LT Exposure)

Wang 2020 (US: Nationwide; LT Exposure)*
Wu 2020a (US: Nationwide; LT Exposure)*

Wyatt 2020c (US: 530 US counties; ST Exposure)*

6 7 8 9 10 11 12

Overall PM2.5 Concentration for the Study Period (ng/m3)

Study Type

¦ ST Exposure & Mortality

¦ ST exposu re 8 Morbidity

¦ LT exposure & Mortality

¦ LT Exposure & Morbidity

Summary Statistics
O 10th percentile
• 25th percentile

¦ Mean

Figure 3-13. Hybrid model-predicted PM2.5 concentrations in key U.S. epidemiologic studies, subset by method used to average
grid cells in study-reported long-term mean PM2.5 concentrations. (Asterisks denote studies included in the ISA Supplement).

3-115

-------
Nationwide Grid cell averaged up to zip, postal code, US
or census tract

Regional Avg of all grid cells

Grid cell averaged up to zip, postal code, us
or census tract

Population weighted grid cell averaged
up to census tract or zip code

Hart 2015 (US: Nationwide; LT Exposure)
Di 2017a (US: Nationwide; ST Exposure)
deSouza 2021 (US: Nationwide; ST Exposure)*
Oi 2017b (US: Nationwide; LT Exposure)
Dominici 2019 (US: Nationwide; LT Exposure)*
Wang 2020 (US: Nationwide; LT Exposure)*

Wu 2020a (US: Nationwide; LT Exposure)*

Kloog 2014 (US: 7 Mid-Atlantic States and
D.C.; ST Exposure)

Kloog 2012 (US: 6 NE States; LT Exposure)
Kloog 2012 (US: 6 NE States; ST Exposure)
Shi 2016 (US: 6 NE States; ST Exposure)

Shi 2016 (US: 6 NE States; LT Exposure)

Thurston 2016 (US: 6 States and 2 MSAs; LT
Exposure)

Lee 2015b (US: 3 SE States; ST Exposure)

Wang 2017 (US: 7 SE States; LT Exposure)

Wyatt 2020c (US: 530 US counties; ST
Exposure)*

11 12

Overall PM2.5 Concentration for the Study Period (ng/m3)

Study Type

¦ ST Exposure & Mortal ity

¦ ST exposure & Morbidity

¦ LT exposure & Mortality

¦ LT Exposure & Morbidity

Summary Statistics
O 10th percentile
• 25th percentile

¦ Mean

Figure 3-14. Hybrid model-predicted PM2.5 concentrations in key U.S. epidemiologic studies, subset by spatial scale and

method used to average grid cells in study-reported long-term mean PM2.5 concentrations. (Asterisks denote studies included
in the ISA Supplement).

3-116

-------
• What is the range of mean PM2.5 concentrations reported by key epidemiologic
studies that use monitor-based and hybrid modeling-based approaches and how do
these values compare to each other? What is the range of concentrations
corresponding to the lower percentiles of the data in those studies with the available
information?

Based on the information above with regard to the key U.S. and Canadian epidemiologic

studies, we summarize some of our observations:

• For key U.S. epidemiologic studies that use monitors to estimate PM2.5 exposures (Figure 3-
8), overall mean PM2.5 concentrations range between 9.9 |j,g/m3 37 to 16.5 |j,g/m3.

• For key U.S. epidemiologic studies that use hybrid model-predicted exposure (Figure 3-10),
mean PM2.5 concentrations range from just above 8.0 |j,g/m3 to just above 12.0 |j,g/m3.

o The majority of these studies estimate mean PM2.5 exposure by averaging up from the
grid cell spatial resolution used in the modeling approach to the spatial resolution of
health study data (e.g., ZIP code or census tract). This incorporates an aspect of
population weighting in the calculation of the mean. Based on our air quality analyses,
we would expect these epidemiologic studies to report means similar to those from
monitor-based studies.

averaging from the grid cell spatial resolution across the entire study area, whether that
be the nation or a region of the country. These studies do not weight the estimated
exposure concentrations based on population density or location of health events.
Because of this, these reported mean concentrations are most different (and much
lower) than those means reported in monitor-based studies. Due to the methodology
employed in calculating the study-reported means and not necessarily a difference in
estimates of exposure, we would expect these epidemiologic studies to report some of
the lowest mean values.

- For these studies, the reported mean PM2.5 concentrations range from 8.1 [j,g/m3 to
11.9 ^ig/m3.

• Of the key epidemiologic studies evaluated in the 2019 ISA and ISA Supplement, a subset of
studies report PM2.5 concentrations corresponding to the 25th and 10th percentiles of health
data or exposure estimates to provide insight into the concentrations that comprise the lower
quartiles of the air quality distributions.

In key U.S. epidemiologic studies that use monitors to estimate PM2.5 exposures,
25th percentiles of health events correspond to mean PM2.5 concentrations (i.e.,

37 This is generally consistent with, but slightly below, the lowest study-reported mean PM2 5 concentration from
monitor-based studies available in the 2020 PA, which was 10.7 |ig/m3 (U.S. EPA, 2020, Figure 3-7).

3-117

-------
averaged over the study period for each study city) at or above 11.5 |ag/m3 and
10th percentiles of health events correspond to mean PM2.5 concentrations at or
above 9.8 |j,g/m3 (i.e., 25% and 10% of health events, respectively, occur in study
locations with mean PM2.5 concentrations below these values).

Of the key U.S. epidemiologic studies that use hybrid modeling approaches to
estimate long-term PM2.5 exposures, the ambient PM2.5 concentrations
corresponding to 25th percentiles of estimated exposures are 6.2 and 9.1 |j,g/m3.
In key U.S. epidemiologic studies that use hybrid modeling approaches to
estimate short-term PM2.5 exposures, the ambient concentrations corresponding to
25th percentiles of estimated exposures, or health events, are generally at or above
6.4 |j,g/m3. In the one study with lower concentrations, the ambient PM2.5
concentration corresponding to the 25th percentile of estimated exposures is 4.7
Hg/m3.38 In the one study with information available on the 10th percentile of
health events, the ambient PM2.5 concentration corresponding to that 10th
percentile is 4.7 |j,g/m3.

• Generally, the study-reported mean concentrations in Canadian studies are lower than those
reported in the U.S. studies for both monitor-based and hybrid model methods.

- For the majority of key Canadian epidemiologic studies that use monitor-based
exposure (Figure 3-9), mean PM2.5 concentrations generally ranged from 7.0
l_ig/m3 to 9.0 |j,g/m3. For these studies, 25th percentiles of health events correspond
to mean PM2.5 concentrations at or above 6.5 |j,g/m3 and 10th percentiles of health
events correspond to mean PM2.5 concentrations at or above 6.4 |j,g/m3.

- For the key Canadian epidemiologic studies that use hybrid model-predicted
exposure (Figure 3-11), the mean PM2.5 concentrations are generally lower than in
U.S. model-based studies (Figure 3-10), ranging from approximately 6.0 |j,g/m3 to
just below 10.0 |j,g/m3.

The majority of the key Canadian epidemiologic studies that used hybrid
modeling were completed at the nationwide scale, while four studies were
completed at the regional geographic spatial scale. In addition, all the key
Canadian epidemiologic studies apply aspects of population weighting, where all
grid cells within a postal code are averaged, individuals are assigned exposure at
the postal code resolution, and study mean PM2.5 concentrations are based on the
average of individual exposures.

The majority of studies estimating exposure nationwide range between just below
6.0 |j,g/m3 to 8.0 |j,g/m3. One study (Erickson et al. (2020)) presents an analysis
related immigrant status and length of residence in Canada versus non-immigrant
populations, which accounts for the four highest mean PM2.5 concentrations in
Figure 3-11, ranging between 9.0 |j,g/m3 and 10.0 |j,g/m3.

The four studies that estimate exposure at the regional scale report mean PM2.5
concentrations that range from 7.8 |j,g/m3 to 9.8 |j,g/m3.

38 As noted above, in this study (Shi et al., 2016), the authors report that most deaths occurred at or above the 75th
percentile of annual exposure estimates (i.e., 10 |-ig/m3). The short-term exposure estimates accounting for most
deaths are not presented in the published study.

3-118

-------
In two Canadian studies with information available on the 25th percentile of health
events, the ambient PM2.5 concentration corresponding to that 25th percentile is
approximately 8.0 |j,g/m3 in two studies, and 4.3 |j,g/m3 in a third study.

For the studies that utilize hybrid modeling approaches but do not incorporate population
weighting in calculating the mean, we would expect the associated annual design values to be
much higher (i.e., 40-50% higher) than the study-reported means (i.e., 8.1 [j,g/m3 to 11.9 (J,g/m3).
This larger difference between design values and study-reported mean concentrations makes it
more difficult to consider how these studies can be used to determine the adequacy of the
protection afforded by the current or potential alternative annual standards. In addition to the key
epidemiologic studies, the 2019 ISA and ISA Supplement also include a subset of studies that
assess the relationship between PM2.5 exposure and health effects that have emerged and so we
ask:

• To what extent has information emerged to further inform our understanding of
PM2.5 in ambient air and associations with health effects? Are there studies that
explore alternative methods for assessing the relationship between PM2.5 exposure
and health effects or studies that observe changes in health effects with changes in
PM2.5 concentrations in ambient air over time?

In addition to the expanded body of evidence from the key epidemiologic studies
discussed above, there are also a subset of studies that have emerged that further inform our
understanding of the relationship between PM2.5 exposure and health effects (U.S. EPA, 2019,
U.S. EPA, 2021a U.S. EPA, 2022).

The first type are studies that examine health effect associations in analyses with the
highest exposures excluded, restricting analyses to daily exposures less than the 24-hour primary
PM2.5 standard and annual exposures less than the annual PM2.5 standard. The restricted analyses
can be informative in assessing the nature of the association between long-term exposures (e.g.,
annual average concentrations < 12.0 |ig/m3) or short-term exposures (e.g., daily concentrations
<35 |ig/m3) when looking only at exposures to lower concentrations, including whether the
association persists in such restricted analyses compared to the same analyses for all exposures,
as well as whether the association is stronger, in terms of magnitude and precision, than when
completing the same analysis for all exposures. While these studies are useful in supporting the
confidence and strength of associations at lower concentrations, these studies also have inherent
uncertainties and limitations, including uncertainty in how studies exclude concentrations (e.g.,
are they excluded at the modeled grid cell level, the ZIP code level) and in how concentrations in
studies that restrict air quality data relate to design values for the annual and 24-hour standards.
Further, these studies often do not report descriptive statistics (e.g., mean PM2.5 concentrations,
or concentrations at other percentiles) that allow for additional consideration of this information.
As such, while these studies can provide additional supporting evidence for associations at lower

3-119

-------
concentrations, there are also limitations in how to interpret these studies when evaluating the
adequacy of the current or potential alternative standards. These studies, as assessed in the 2019
ISA and ISA Supplement, are summarized in Table 3-10 below.

3-120

-------
Table 3-10. Epidemiologic studies examining the health impacts associated with ambient PM2.5 concentrations when studies
are conducted with restricted air quality exposures.

Citation

Study Area

(health
endpoint)

Years of
PM2.5 Air
Quality
(monitored)

AQ in restricted
analysis
(Mg/m3)

Study-
reported
Mean in
restricted
analysis
(ug/m3)

Study-
reported
Mean in

main
analysis
(ug/m3)

Effect Estimate in
restricted analysis
(95% CI)

Effect Estimate in main
analysis (95% CI)

U.S.-based Studies and Long-term Exposure (per 5 ng/m3)

Di et al.,
2017b

Nationwide
(All-cause
mortality 65+)

2000-2012

< 12.0

9.6

11.0

1.07 (1.06-1.07)

1.04(1.04-1.04)

Dominici et al.,
2019

Nationwide
(All-cause
mortality)

2000-2012

< 12

9.6

11.0

1.06 (1.06-1.07)

1.03(1.03-1.04)

Shi et al., 2016

6 NE States

2003-2008

< 10.0

8.1

1.04 (1.00, 1.09)

1.04(1.01, 1.06)

Yazdi et al.,
2019

7 SE States

(CVD
morbidity)

2000-2012

< 12

Stroke: 1.29 (1.27-
1.31)

Ml: 1.18(1.16-1.20)
HF: 1.44 (1.43-1.46)

Stroke: 1.16(1.16-1.17)
Ml: 1.14(1.13-1.15)
HF: 1.29 (1.29-1.30)

Canadian Studies and Long-term Exposure (per 5 ng/m3)

Zhang et al.,
2021

Ontario (Non-
accidental and
CVD mortality)

2000-2016

< 10.0 and
<8.8

7.8

Non-accidental
mortality: < 10.0:1.22
(1.10-1.36); and <8.8:
1.04(0.91-1.17)
CVD mortality: < 10.0:
1.38 (1.10-1.73); and <
8.8: 1.05(0.80-1.38)

Non-accidental mortality: 1.20
(1.09-1.32)

CVD mortality: 1.49 (1.22-1.83)

U.S. Studies and Short-term Exposure (per 10 Mg/m3)

deSouza et al.,
2021

Nationwide
(First CVD HA)

2000-2012

<25

11.5

1.3% (0.9-1.6%)

0.9% (0.6-1.1 %)

3-121

-------
Di et al.,
2017a

Nationwide
(All-cause
mortality 65+)

2000-2012

<25.0

11.6

1.61 (1.48-1.74)

1.18(1.09-1.28)

Lee et al.,
2015 1

3 SE States

(Non-
accidental)

2007-2011

In ZIP codes
where annual
average <12.0
and only on
days < 35.0

11.1

Non-accidental: 2.08%
(1.99-2.17)%

Non-accidental 1.56% (1.19-
1.94%)

Lee et al.,
20152

3 SE States

(Non-
accidental)

2007-2011

In ZIP codes
where annual
average < 12.0

11.1

Non-accidental: 2.06%
(1.97-2.15%)

Non-accidental 1.56% (1.19-
1.94%)

Shi et al., 2016

6 NE States

2003-2008

<30.0

8.2

2.14% (1.34-2.95%)

2.14% (1.38, 2.89%)

Wei et al.,
2019

Nationwide
(CVD HA)

2000-2012

<25
(WHO air quality
guideline value
for daily PM2.5)

Relative increase in
risk for HA with 1
|jg/m3 increase in
Iag0-1 PM25:
Ml: 0.16 (0.09, 0.24)
CHF: 0.16(0.11,0.22)

Relative increase in risk for HA
with 1 pg/m3 increase in Iag0-1
PM25:

Ml: 0.11 (0.07, 0.16)
CHF: 0.14(0.10, 0.17)

1 First, restricted ZIP code areas to where the annual average of predicted PM2.5 is < 12 |jg/m3 to assess the acute effect of PM2.5 on mortality only areas with annual average
concentrations < 12 |jg/m3.

2 In terms of daily standard, conducted analysis on the days < 35 |jg/m3 and only in ZIP codes with annual average concentrations < 12 |jg/m3.

3-122

-------
There are a number of U.S. and Canadian studies that examine health effect associations
in analyses with the highest exposures excluded. These restricted analyses provide support for
positive and statistically significant effect estimates at lower mean PM2.5 concentrations than
their main effect analysis means as shown in Table 3-10 and in many cases, exhibit greater effect
estimates in magnitude than their corresponding main analyses. With regard to these studies, we
particularly note the following:

• In the four U.S. studies that estimate effects associated with long-term exposure to PM2.5, the
effect estimates are greater in the restricted analyses than in the main analyses.

o Di et al. (2017a) and Dominici et al. (2019) report positive and statistically significant
associations in analyses restricted to concentrations less than 12.0 |j,g/m3 for all-cause
mortality Di et al. (2017b) and stroke, MI, and HF Dominici et al. (2019), and effect
estimates are greater in the restricted analyses than effect estimates reported in main
analyses. In addition, both studies report mean PM2.5 concentrations of 9.6 |j,g/m3

o Shi et al. (2016) and Yazdi et al. (2019) report positive and statistically significant
associations in analyses restricted to concentrations less than 10.0 |j,g/m3 and 12.0
Hg/m3, respectively. Shi et al. (2016) does not report overall mean PM2.5
concentrations in restricted analyses, though such means are presumably somewhat
below the main analysis reported mean of 8.1 |j,g/m3. Yazdi et al. (2019) does not
report the overall mean PM2.5 concentration in either the restricted analysis or main
analysis, but the effect estimates for stroke, MI, and HF are all higher in the restricted
analyses compared to main analyses.

• While none of the U.S. studies of short-term exposure present mean PM2.5 concentrations for
the restricted analyses, these studies generally have mean 24-hour average PM2.5
concentrations in the main analyses below 12.0 |ag/m\ and report increases in the effect
estimates in the restricted analyses compared to the main analyses.

o With the exception of Wei et al. (2019), short-term exposure studies report mean 24-
hour average PM2.5 concentration in main analyses all below 12.0 ng/m3, and ranging
from 8.2 |j,g/m3 Shi et al. (2016) to 11.6 (Di et al. (2017a).39

o These studies, except for Shi et al. (2016), report increases in effect estimates in

restricted analyses compared to main analyses. Shi et al. (2016) reports the same effect
estimates for both the restricted and main analyses.

• In the one Canadian study of long-term PM2.5 exposure, Zhang et al. (2021) conducted
analyses where annual PM2.5 concentrations were restricted to concentrations below 10.0
Hg/m3 and 8.8 |ag/m\ which presumably have lower mean concentrations than the mean of
7.8 |j,g/m3 reported in the main analyses, though restricted analysis mean PM2.5
concentrations are not reported.

o Effect estimates for non-accidental mortality are greater in analyses restricted to PM2.5
concentrations less than 10.0 |ag/m\ but less in analyses restricted to < 8.8 |j,g/m3.

39 93.6% of all case and control days in this study had PM2 5 concentrations below 25

3-123

-------
Effect estimates for CVD mortality are lower in restricted analyses than the main
analysis.

Overall, these studies provide additional information on the nature of the association
between long- or short-term exposures when analyses are restricted to lower PM2.5
concentrations. Further, these studies indicate that effect estimates are generally greater in
magnitude in the restricted analyses for long- and short-term PM2.5 exposure compared to the
main analyses.

The second type of studies that have recently emerged and can further inform our
understanding of the relationship between PM2.5 exposure and health effects are those that
employ alternative methods for confounder control. Alternative methods for confounder control
seek to mimic randomized experiments through the use of study design and statistical methods to
more extensively account for confounders and are more robust to model misspecification. The
studies that employ alternative methods for confounder control assessed in the 2019 ISA and ISA
Supplement are summarized in Table 3-11 below.

3-124

-------
Table 3-11. Summary of information from studies that use alternative methods for confounder control.

Study
Reference

Statistical
Method1

Study Area

AQ Years

Health
Endpoint
(populatio
n)

Study-reported
Mean (ng/m3)

Results

Awad et
al., 2019

IPW

U.S.
Nationwide

2000-
2012

LT
mortality
(65+)

Mean change in
exposure the year
before move and
the second year
after move:
Whites: -0.73
Blacks: -0.90

Per a 10 pg/m3 increase in annual PM2 5 concentrations:
White individuals: HR= 1.21 (95% CI: 1.20,1.22)

Black individuals: HR = 1.12 (95% CI: 1.08,1.15)
All-cause mortality: HR = 1.12 (95% CI: 1.08,1.15)

Awad et
al., 2019
(restricted
analysis)

IPW

U.S.
Nationwide

2000-
2012

LT
mortality
(65+)

Restricted <12.0:
NR

Per a 10 pg/m3 increase in annual PM2.5 concentrations:
White individuals: HR = 1.25 (95% CI: 1.24,1.27)

Black individuals: HR = 1.08 (95% CI: 1.01,1.14)

Higbee et
al., 2020

IPW

U.S.
Nationwide

1986-
2015

LT
mortality
(18+)

10.7

For a 10 pg/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)

Qiu et al.,
2020

IPW

New England

2000-
2012

ST CVD
HA (65+)

AMI: 10.3
CHF: 10.08
IS: 10.1

Percent increase HA rate for a 10 pg/m3 increase in

PM2.5 concentrations

AMI: 4.31% (95% CI: 2.21, 6.42)

CHF: 3.95% (95% CI: 2.37,5.53)

IS: 2.56% (95% CI: 0.44, 4.69)

Schwartz
et al.,
2018a

3 approaches:
Instrumental
approach
Marginal structural
models
Time-series
analysis

135 U.S.
Cities

1999-
2010

ST
mortality
(18+)

12.8

Percent change in daily mortality per 10 pg/m3 increase
in PM2.5 concentrations

Instrumental approach: 1.54% (95% CI: 1.12,1.97)
Marginal structural models: 0.75% (95% CI: 0.35,1.15)
Time-series: 0.60%: (95% CI: 0.34, 0.85%)

Schwartz
et al.,
2018a

3 approaches:
Instrumental
approach

135 U.S.
Cities

1999-
2010

ST
mortality
(18+)

Restricted < 25.0:
NR

Percent change in daily mortality per 10 pg/m3 increase
in PM2.5 concentrations

Instrumental approach: 1.70% (95% CI: 1.11, 2.29)

3-125

-------
(restricted
analysis)

Marginal structural
models
Time-series
analysis

Marginal structural models: 0.83% (95% CI: 0.39,1.27)
Time-series: 0.62%: (95% CI: 0.32, 0.93)

Schwartz
et al.,
2018b

GPS IPW

Northeastern

and Mid-
Atlantic States

2000-
2012

Life
expectancy

Estimated mean age at death for an annual average
exposure of 12 pg/m3 was 0.89 years (95% CI:
0.88,0.91) than estimated for a counterfactual PM2.5
exposure of 7.5 |jg/m3

Schwartz
et al.,
2021

DID

U.S.
Nationwide

2000-
2016

probability
of dying
(65+)

10.3

Probability of dying in each year increased by 3.85x104
(95% CI 1.95x104, 5.76x104) for each 1 pg/m3 increase
in annual PM2.5 concentrations

Schwartz
et al.,

2021

(restricted
analysis)

DID

U.S.
Nationwide

2000-
2016

probability
of dying
(65+)

Probability of dying in each year increased by 4.26x104
(95% CI 1.43x104, 7.09x104) for each 1 pg/m3 increase
in annual PM2.5 concentrations

Wu et al.,
2019

RC-GPS and 3
GPS approaches:
Subclassification
GPS
IPTW GPS
GPS matching

New England

2000-
2012
(modeled)

LT
mortality
(65+)

Exposure levels of low (< 8.0 pg/m3) versus moderate
PM2.5 concentrations (8.0-10.0 Dg/m3) to low exposure
Subclassification: 1.025 (95% CI: 1.006,1.045)

IPTW GPS: 1.022 (95% CI: 1.007, 1.038)

Matching GPS: 1.028 (1.012,1.045)

Comparison of exposure levels of < 8.0 pg/m3 vs. >
10.0 pg/m3

Subclassification: 1.035 (95% CI: 0.999,1.072)

IPTW GPS: 1.030 (95% CI: 1.005, 1.056)

Matching GPS: 1.035 (95% CI: 1.015,1.055)

Wu et al.,
2020

Three GPS
approaches:
GPS matching
GPS weighting
GPS adjustment

U.S.
Nationwide

2000-
2016
(modeled)

LT
mortality
(65+)

9.8

Reported hazard ratios for a decrease in mortality risk
per 10 pg/m3 decrease in annual PM2.5
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 = 1.072 (95% CI: 1.061,1.082)

Wu et al.,
2020
(restricted
analysis)

Three GPS
approaches:
GPS matching
GPS weighting

U.S.
Nationwide

2000-
2016
(modeled)

LT
mortality
(65+)

Restricted <12.0:
8.4

Reported hazard ratios for a decrease in mortality risk
per 10 pg/m3 decrease in annual PM2.5
GPS matching: HR = 1.261 (95% CI: 1.233,1.289)
GPS weighting: HR = 1.268 (95% CI: 1.237,1.300)

3-126

-------

GPS adjustment

GPS adjustment: HR = 1.231 (95% CI: 1.180,1.284)

Yazdi et
al., 2021

Doubly Robust
Additive Model
(DRAM)

U.S.
Nationwide

2000-
2016
(modeled)

Cardiovasc

ular
hospitalizat

ion
outcomes
(65+)

10.21

% increase in the risk with 1 pg/m3 increase in PM2.5:
Ml: 0.002; Stroke: 0.009; Al: 0.006

Yitshak-
Sade et
al., 2019

DID

Northeastern

and mid-
Atlantic States
(14 U.S.
States)

2000-
2013

LT
mortality
(65+)

Range: 6.5-14.5

4.04% (95% CI: 3.49,4.59) increase in mortality rates for
an IQA (3 pg/m3) increase in annual PM2.5
concentrations

Zigler et
al., 2018

Propensity Score
within Bayesian

Hierarchical
Regression model

Eastern U.S.

2000-
2012
(baseline
year:
2000-
2005, and
follow-up
year
2010-
2012)

LT
mortality

and
hospitalizat

ion for
Respiratory
and CVD
outcomes
(65+)

2002-2004:
Attainment areas:
11.59 ug/m3
Non-attainment
areas: 14.48 ug/m3

2010-2012:
Attainment areas:
9.39 ug/m3
Non-attainment
areas: 11.13 ug/m3

All-cause mortality: Overall average effects across non-
attainment areas, the non-attainment designations
reduced the mortality rate by 1.251 deaths per 1000
beneficiaries, with 95% posterior interval (-2.631,

0.108).

Hospitalizations: Overall average effects of the
designations on hospitalization outcomes ranged from
an average reduction of 0.651 for CV stroke to an
increase of 0.440 for respiratory tract infections per
1000 person-years, although 95% posterior intervals for
all hospitalization outcomes included 0.

Average associative effects: a reduction in the all-cause
mortality rate of 3.16 (95% CI: -5.19, -1.21) deaths per
1000 beneficiaries. A reduction in the hospitalization
rate ranged from 2.37 for heart failure to 2.61 for
respiratory tract infections per 1000 person-years. This
provided evidence that the nonattainment designations
causally reduced rates of health outcomes among areas
where the designations causally reduced ambient
PM2.5.

3-127

-------
1 GPS: generalized propensity score; IPW: inverse probability weighting; DID: Difference-in-difference; HR: hazard ratio; IRR: incidence rate ratio; IPTW: inverse probability
treatment weighting; IV: instrument variable; OLS: Ordinary Least Squares; RC: regression calibration

3-128

-------
The 2019 ISA and ISA Supplement assess epidemiologic studies that implemented
alternative methods for confounder control. As presented in Table 3-11 above, these studies
employ a variety of statistical methods, such as GPS, IPW, and DID. We particularly note the
following:

• These studies reported consistent results among large study populations across the U.S. The
results from studies that use alternative methods for confounder control further inform the
relationship between long- and short-term PM2.5 exposure and total mortality.

• Studies that employ alternative methods for confounder control to assess the association
between long-term exposure to PM2.5 and mortality provide additional support for the
associations reported in the broader body of cohort studies that examined long-term PM2.5
exposure and mortality.

- For example, Wu et al., 2020 used three different alternative methods for

confounder control, in addition to two more traditional statistical method methods
(Cox proportional hazards modeling and Poisson time-series regression model),
finding consistent positive and statistically significant results between the five
statistical methods and with HRs per a 10 |j,g/m3 increase in PM2.5 ranging from
1.062 (95% CI: 1.055,1.069) using the poisson statistical method to 1.076 (95%
CI: 1.065, 1.088) with the GPS matching statistical method.

Lastly, there is also a smaller subset of epidemiologic studies that assess whether long-
term reductions in ambient PM2.5 concentrations result in corresponding reductions in health
outcomes. These include studies that evaluate the potential for improvements in public health,
including reductions in mortality rates, increases in life expectancy, and reductions in respiratory
disease as ambient PM2.5 concentrations have declined over time. Some of these studies are
accountability analyses, which 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 . Given the nature of these
studies, the majority tend to focus on time periods in the past during which ambient PM2.5
concentrations were substantially higher than those measured more recently (e.g., see Chapter 2,
Figure 2-16). These studies, as assessed in the 2019 ISA and ISA Supplement, are summarized in
Table 3-12 below.

3-129

-------
Table 3-12. Epidemiologic studies examining the health impacts of long-term reductions in
ambient PM2.5 concentrations.

Study
Reference

Study Area

Years of
PM2.5 Air
Quality
(monitored)

Starting Mean

PM2.5
Concentration
(ug/m3)

Ending Mean

PM2.5
Concentration
(ug/m3)

Study Results

Pope et al.,
20091

211 U.S.
counties

1979-1983

Statistically significant
association between

compared to
1999-2000

20.6

14.1

declining ambient PM2.5
and increasing life
expectancy

Correia et al.,
20131

545 U.S.
counties

2000

Statistically significant
association between

compared to
2007

13.2

11.6

declining ambient PM2.5
and increasing life
expectancy

8.0

Reductions in PM2.5

Bennett et al.,
2019 1

U.S.
Nationwide
and 1339
U.S. counties

1999-2015

13.6 (Pop-
weighted

(Population-
weighted
mean; Mean

since 1999 have
increased life
expectancy in men and

mean)

range in
counties: 2.8-
13.2)

women in all but 14
counties where PM2.5
increased slightly

Berhane et al.,
20162

4,602 children

1992-2000;

Statistically significant
decrease in bronchitic

in 8 California
communities

1995-2003;
2002-2011

20.5

14.4

symptoms in 10-year old
children with and without
asthma

Gauderman et
al., 20152

2,120 children
in 5 California
communities

1994-1997;
1997-2000;
2007-2010

21.3-31.5

11.9-17.8

Statistically significant
improvements in 4-year
growth of lung function

Corrigan et al.,

619 U.S.
counties (486
attainment

2000-2004:
All: 12.0
Attainment:

2005-2010:
All: 10.8
Attainment:
10.2
Nonattainment:
13.2

Fewer CV deaths per
year for each 1 pg/m3
decrease in PM2.5.

20183

counties and

2000-2010

11.1

133

Nonattainment:

nonattainment

15.3

counties)

Henneman et al.,

Multiple U.S.
states

Reduced exposure to
total PM2.5 and coal

2019b3

2005-2012

2005:10.0

2012: 7.2

emissions led to

reduced rates in CVD-
related HA.

Sanders et al.,
2020b3

600-700 U.S.
counties

2000-2013

Before 2006:

Non-
attainment:
15.3 and

After 2006:

Non-
attainment:
12.0
Attainment: 9.3

By 2005 PM2.5
designation status
(attainment or non-
attainment), PM2.5 levels

Attainment:
11.0

and corresponding
mortality rates

3-130

-------
Fan and Wang,
20203

Eastern US

1999-2013

Fewer CVD deaths per
year for each 1 pg/m3
reduction in annual
PM2.5 concentrations

Peterson et al.,
20203

2132 counties

1990-2010

Fewer CVD deaths for
each 1 pg/m3 reduction
in annual PM2.5
concentrations

Zigler et al., 2018

Eastern US

Baseline
year: 2000-
2005 and
Follow-up
year: 2010-
2012

2002-2004:
Attainment
areas: 11.59
ug/m3

Non-attainment
areas: 14.48
ug/m3

2010-2012:
Attainment
areas: 9.39
ug/m3

Non-attainment
areas: 11.13
ug/m3

Substantial reductions in
all-cause mortality; and
rate of hospitalizations
for respiratory and CVD
outcomes were
observed in the subset
of areas designated as
nonattainment and
provided evidence for
effectiveness of targeted
local control measures
beyond the regional
strategies.

Wyatt et al.,
2020b4

2132 counties
in the U.S.
(population
>20,000)

1990-2010

The annual change in
cardiovascular mortality
rate ranged from 6.5-7.6
fewer deaths/year (per
100,000 person-years)
per 1 pg/m3 decrease in
PM2.5 over time.

1 Life expectancy s
evaluating change

tudies;2 Studies of respiratory disease in children;3 Accountability studies; and 4Study
in PM25 concentration over time and associated change in cardiovascular mortality rate

The studies assessed in the 2019 ISA and ISA Supplement that evaluate health impacts of
long-term reductions in ambient PM2.5 concentrations provide support for the conclusion that
public health benefits are associated with decreases in ambient PM2.5 concentrations. In
particular, we note the following key observations from these studies:

• Of the accountability studies evaluated in the 2019 ISA and ISA Supplement, which account
for changes in PM2.5 concentrations due to policy or the implementation of an intervention to
assess whether there was evidence of changes in associations with mortality or
cardiovascular effects due to changes in annual PM2.5 concentrations, Corrigan et al. (2018),
Henneman et al. (2019b) and Sanders et al. (2020b) present analyses with starting
concentrations below 12.0 |ig/m3.

- Henneman et al. (2019b) explored the changes in modeled PM2.5 concentrations
following the retirement of coal fired power plants in the U.S., and found that
reductions from mean annual PM2.5 concentrations of 10.0 |ig/m3 in 2005 to mean
annual PM2.5 concentrations of 7.2 |ig/m3 in 2012 from coal-fueled power plants
resulted in corresponding reductions in the number of cardiovascular-related

3-131

-------
hospital admissions, including for all cardiovascular disease, acute myocardial
infarction, stroke, heart failure, and ischemic heart disease in those aged 65 and
older.

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 implementation based on mortality data from the
National Center for Health Statistics. They reported 1.10 (95% CI: 0.37, 1.82)
fewer cardiovascular deaths per year per 100,000 people for each 1 [j,g/m3
reduction in annual PM2.5 concentrations. When comparing whether counties met
the annual PM2.5 standard, there were 1.96 (95% CI: 0.77, 3.15) fewer
cardiovascular deaths for each 1 [j,g/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. (2020b) examined whether policy actions (i.e., the first annual
PM2.5 NAAQS implementation rule in 2005 for the 1997 annual PM2.5 standard
with a 3-year annual average of 15 (J,g/m3) reduced PM2.5 concentrations and
mortality rates in Medicare beneficiaries between 2000-2013. They found
evidence of changes in associations with mortality (a decreased mortality rate of ~
0.5 per 1,000 in attainment and non-attainment areas) due to changes in annual
PM2.5 concentrations in both attainment and non-attainment areas, which had
starting concentrations below 12.0 |ig/m3 following implementation of the annual
PM2.5 NAAQS in 2005. In addition, following implementation of the annual
PM2.5 NAAQS, annual PM2.5 concentrations decreased by 1.59 [j,g/m3 (95% CI:
1.39, 1.80) which corresponded to a reduction in mortality rates among
individuals 65 years and older (0.93% [95% CI: 0.10%, 1.77%]) in non-
attainment counties relative to attainment counties.

• Bennett et al. (2019) reports increases in life expectancy in all but 14 counties (1325 of 1339
counties) that have exhibited reductions in PM2.5 concentrations from 1999 to 2015.

• While Fan and Wang (2020), Peterson et al. (2020), and Wyatt et al. (2020a) do not report
starting and ending concentrations, these studies lend support to the conclusions that
reductions in PM2.5 concentrations lead to public health improvements, including reductions
in cardiovascular mortality.

The information in Table 3-10, Table 3-11, and Table 3-12 provide additional support to
inform the relationship between long- and short-term PM2.5 exposure and total mortality.
Analyses that are restricted only to concentrations at or below the levels of the current primary
PM2.5 standards find positive and significant associations with exposure to PM2.5 and health
outcomes. These restricted analyses often report greater effect estimates compared to effect
estimates in the main analysis that uses the full distribution of PM2.5 concentrations. Studies that
use alternative methods for confounder control to assess the relationship between PM2.5 and
health outcomes provide additional support for the associations reported in other epidemiologic
studies. Finally, new studies assessed in the ISA Supplement evaluate the relationship between

3-132

-------
declines in ambient PM2.5 concentrations over time and the potential for improvements in public
health, and support the conclusion in the 2020 PA; improvements in air quality are associated
with improvements in public health. Some of these new studies have lower starting
concentrations than similar studies included in the 2019 ISA.

3.3.4 Uncertainties in the Health Effects Evidence

• To what extent have important uncertainties identified in prior reviews been

reduced and/or have additional uncertainties emerged?

We have not identified any new uncertainties in the evidence since the 2020 review.
However, we continue to recognize uncertainties that persist from the previous reviews. This
array of important areas of uncertainty related to the current health effects evidence, including
that assessed in the 2019 ISA and the ISA Supplement, is summarized below.

Although the epidemiologic studies clearly demonstrate associations between long- and
short-term PM2.5 exposures and health outcomes, as in previous reviews, we continue to
recognize several uncertainties and limitations in the health effects evidence remain.
Epidemiologic studies evaluating short-term PM2.5 exposure and health effects have reported
heterogeneity in associations between cities and geographic regions within the U.S.

Heterogeneity in the associations observed across epidemiologic studies may be due in part to
exposure error related to measurement-related issues, the use of central fixed-site monitors to
represent population exposure to PM2.5, and our limited understanding of factors that could be
due to a number of factors including exposure error related to measurement-related issues,
variability in PM2.5 composition regionally, and factors that result in differential exposures (e.g.,
topography, the built environment, housing characteristics, personal activity patterns).
Heterogeneity is expected when the methods or the underlying distribution of covariates vary
across studies (U.S. EPA, 2019, p. 6-221). Studies assessed in the 2019 ISA and ISA Supplement
have advanced the state of exposure science by presenting innovative methodologies to estimate
PM exposure, detailing new and existing measurement and modeling methods, and further
informing our understanding of the influence of exposure measurement error due to exposure
estimation methods on the associations between PM2.5 and health effects reported in
epidemiologic studies (U.S. EPA, 2019, section 1.2.2; U.S. EPA, 2022). Data from PM2.5
monitors continue to be commonly used in health studies as a surrogate for PM2.5 exposure, and
often provide a reasonable representation of exposures throughout a study area (U.S. EPA, 2019,
section 3.4.2.2; U.S. EPA, 2022, section 3.2.2.2.2). However, an increasing number of studies
employ hybrid modeling methods to estimate PM2.5 exposure using data from several sources,
often including satellites and models, in addition to ground-based monitors. These hybrid models
typically have good cross-validation, especially for PM2.5, and have the potential to reduce

3-133

-------
exposure measurement error and uncertainty in the health effect estimates from epidemiologic
models of long-term exposure (U.S. EPA, 2019, section 3.5; U.S. EPA, 2022, section 2.3.3).

While studies using hybrid modeling methods have demonstrated reduced exposure
measurement error and uncertainty in the health effect estimates, these studies use a variety of
approaches to estimate PM2.5 concentrations and to assign exposure to assess the association
between health outcomes and PM2.5 exposure. This variability in methodology has inherent
limitations and uncertainties, as described in more detail in section 2.3.3.1.5, and the
performance of the modeling approaches depends on the availability of monitoring data which
varies by location. Factors likely contributing to poorer model performance often coincide with
relatively low ambient PM2.5 concentrations, in areas where predicted exposures are at a greater
distance to monitors, and under conditions where the reliability and availability of key datasets
(e.g., air quality modeling) are limited. Thus, uncertainty in hybrid model predictions becomes
an increasingly important consideration as lower predicted concentrations are considered.

Regardless of whether a study uses monitoring data or a hybrid modeling approach when
estimating PM2.5 exposures, one key limitation that persists is associated with the interpretation
of the study-reported mean PM2.5 concentrations and how they compare to design values, the
metric that describe the air quality status of a given area relative to the NAAQS.40 As discussed
above, the overall mean PM2.5 concentrations reported by key epidemiologic studies reflect
averaging of short- or long-term PM2.5 exposure estimates across location (i.e., across multiple
monitors or across modeled grid cells) and over time (i.e., over several years). For monitor-based
studies, the comparison is somewhat more straightforward than for studies that use hybrid
modeling methods, as the monitors used to estimate exposure in the epidemiologic studies are
generally the same monitors that are used to calculate design values for a given area. It is
expected that areas meeting a PM2.5 standard with a particular level would be expected to have
average PM2.5 concentrations (i.e., averaged across space and over time in the area) somewhat
below that standard level. Analyses of recent air quality in U.S. CBS As indicate that maximum
annual PM2.5 design values for a given three-year period are often 10% to 20% higher than
average monitored concentrations (i.e., averaged across multiple monitors in the same CBSA
(U.S. EPA, 2020, Appendix B, section B.7). The difference between the maximum annual design
value and average concentration in an area can be smaller or larger than this range, likely
depending on factors such as the number of monitors, monitor siting characteristics, and the
distribution of ambient PM2.5 concentrations. For studies that use hybrid modeling methods to
estimate PM2.5 concentrations, the comparison between study-reported mean PM2.5

40 For the annual PM2 5 standard, design values are calculated as the annual arithmetic mean PM2 5 concentration,
averaged over 3 years. For the 24-hour standard, design values are calculated as the 98th percentile of the annual
distribution of 24-hour PM2 5 concentrations, averaged over three years (Appendix N of 40 CFR Part 50).

3-134

-------
concentrations and design values is more complicated given the variability in the modeling
methods, temporal scales (i.e., daily versus annual), and spatial scales (i.e., nationwide versus
urban) across studies. A recent comparison between two hybrid modeling surfaces explored the
impact of these factors on the resulting mean PM2.5 concentrations and provided additional
information about the relationship between mean concentrations from studies using hybrid
modeling methods and design values (see section 2.3.3.1.4). However, the results of those
analyses only reflect two surfaces and two types of approaches, so uncertainty remains in
understanding the relationship between estimated modeled PM2.5 concentrations and design
values more broadly across hybrid modeling studies. Moreover, this analysis was completed
using two hybrid modeling methods that estimate PM2.5 concentrations in the U.S., thus an
additional uncertainty includes understanding the relationship between modeled PM2.5
concentrations and design values reported in Canada.

In addition, where PM2.5 and other pollutants (e.g., ozone, nitrogen dioxide, and carbon
monoxide) are correlated, it can be difficult to distinguish whether attenuation of effects in some
studies results from copollutant confounding or collinearity with other pollutants in the ambient
mixture (U.S. EPA, 2019, section 1.5.1; U.S. EPA, 2022, section 2.2.1). Studies evaluated in the
2019 ISA and ISA Supplement further examined the potential confounding effects of both
gaseous and particulate copollutants on the relationship between long- and short-term PM2.5
exposure and health effects. 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 change in copollutant models. As noted in
the Appendix (Table A-l) to the 2019 PM ISA (U.S. EPA, 2019), 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. However, the studies continue to
provide evidence indicating that associations with PM2.5 are relatively unchanged in copollutants
models (U.S. EPA, 2019, section 1.5.1; U.S. EPA, 2022, section 2.2.1). Another area of
uncertainty is associated with other potential confounders, beyond copollutants. Some studies
have expanded the examination of potential confounders to not only include copollutants, but
also systematic evaluations of the potential impact of inadequate control from long-term
temporal trends and weather (U.S. EPA, 2019, section 11.1.5.1). Analyses examining these
covariates further confirm that the relationship between PM2.5 exposure and mortality is unlikely
to be biased by these factors. Other studies have explored the use of alternative methods for
confounder control to more extensively account for confounders and are more robust to model
misspecification that can further inform the causality determination for long-term and short-term
PM2.5 and mortality and cardiovascular effects (U.S. EPA, 2019, section 11.2.2.4; U.S. EPA,

3-135

-------
2022, sections 3.1.1.3, 3.1.2.3, 3.2.1.2, and 3.2.2.3). These studies indicate that bias from
unmeasured confounders can occur in either direction, although controlling for these
confounders did not result in the elimination of the association, but instead provided additional
support for associations between long-term PM2.5 exposure and mortality when accounting for
additional confounders (U.S. EPA, 2022, section 3.2.2.2.6).

Another important limitation associated with the evidence is that, while epidemiologic
studies indicate associations between PM2.5 and health effects, they do not identify particular
PM2.5 exposures that cause effects. Rather, health effects can occur over the entire distribution of
ambient PM2.5 concentrations evaluated, and epidemiologic studies conducted to date do not
identify a population-level threshold below which it can be concluded with confidence that
PM2.5-related effects do not occur.

Overall, evidence assessed in the 2019 ISA and ISA Supplement continues to indicate a
linear, no-threshold concentration-response relationship for PM2.5 concentrations > 8 [j,g/m3.
However, uncertainties remain about the shape of the C-R curve at PM2.5 concentrations < 8
[j,g/m3, with some recent studies providing evidence for either a sublinear, linear, or supralinear
relationship at these lower concentrations (U.S. EPA, 2019, section 11.2.4; U.S. EPA, 2022,
section 2.2.3.2).

3.4 RISK INFORMATION

To inform conclusions regarding the primary PM2.5 standards that are "requisite" to
protect public health (i.e., neither more nor less stringent than necessary; section 1.2), it is
important to consider the health risks that would be allowed under those standards. For the
current standards, this means evaluating PM2.5-related health risks in locations with three-year
annual PM2.5 design values of 12.0 |j,g/m3 and/or three-year 24-hour design values of 35 |j,g/m3
(i.e., neither above nor below the levels of the current standards). Therefore, in addition to our
evaluation of PM2.5 concentrations in locations of key epidemiologic studies (which are based on
existing air quality; section 3.3.3.2), we assess PIVh.s-attributable risk associated with either:

• PM2.5 air quality that has been adjusted to simulate "just meeting" the current standards (i.e.,
design values equal to 12.0 |j,g/m3 and/or 35 |ag/m3) or lower alternative annual and/or 24-
hour standards.

• The change in risk associated with moving from PM2.5 air quality "just meeting" the current
standards to "just meeting" alternative annual and/or 24-hour standards.

These risk estimates, when considered alongside analyses of the evidence discussed in
section 3.3.3, are meant to inform conclusions on the primary standards that would be requisite
to protect the public health against long- and short-term PM2.5 exposures. Our consideration of
estimated risks focuses on addressing the following policy-relevant questions:

3-136

-------
• What are the estimated PM2.5-associated health risks for air quality just meeting the
current primary PM2.5 standards?

• To what extent are risks estimated to decline when air quality is adjusted to just
meet potential alternative standards with lower levels?

• What are the uncertainties and limitations in these risk estimates?

The sections below summarize our approach to estimating risks (section 3.4.1) and the
results of the risk assessment (section 3.4.1.8). Additional detail on the risk assessment is
provided in Appendix C.

3.4.1 Risk Assessment Overview

Risk assessments combine data from multiple sources and involve various assumptions
and uncertainties. Below we summarize key aspects of the risk modeling approach. Input data for
these analyses includes concentration-response functions from epidemiologic studies (section
3.4.1.1) for each health outcome (section 3.4.1.2) and ambient annual or 24-hour PM2.5
concentrations (sections 3.4.1.3 and 3.4.1.4) for the study areas (section 3.4.1.5) utilized in the
risk assessment. Quantitative and qualitative methods used to characterize variability and
uncertainty in the risk estimates are discussed in section 3.4.1.7.

Information on other data inputs, such as baseline health incidence rate and population
demographic information, can be found in the Estimating PM2.5 and Ozone-Attributable Health
Benefits Technical Support Document (TSD) (U.S. EPA, 2021b; associated with the 2021
Revised Cross-State Air Pollution Rule Update (86 FR 23054, April 30, 2021). Additional detail
on the risk assessment approach is provided in Appendix C (section C. 1).

3.4.1.1 Concentration-Response Functions

Concentration-response functions used in this risk assessment are from large, multicity
U.S. epidemiologic studies that evaluate the relationship between PM2.5 exposures and mortality.
Specific epidemiologic studies and concentration-response functions used here to estimate risk
were identified using criteria that take into account factors such as study design, geographic
coverage, demographic populations, and health endpoints. Information about the studies used in
this risk assessment is summarized in Table 3-13 and additional detail regarding the selection of
epidemiologic studies and specification of concentration-response functions can be found in
Appendix C (section C. 1.1) and the EstimatingPM2.5 and Ozone-Attributable Health Benefits
TSD (U.S. EPA, 2021b).

3.4.1.2 Health Outcomes

Consistent with the overall approach for this reconsideration, this risk assessment has a
targeted scope that focuses on all-cause or nonaccidental mortality associated with long-term and

3-137

-------
short-term PM2.5 exposures (Table 3-13 and Appendix C, section C. 1.1).41 Evidence for these
outcomes supports the determination of a "causal relationship" in the 2019 ISA (U.S. EPA,
2019).42

Table 3-13. Epidemiologic studies used to estimate PM2.5-associated risk.

Epidemiology Study

Study Population3

Age Range
(years)

Mortality Categories Covered

Long-term mortality studies

Di et al., 2017b

Medicare

65+

All-cause

Turner et al., 2016

ACS

+
O
CO

All-cause

Short-term mortality

Baxter et al., 2017

77 cities

All ages

Non-accidental

Ito et al., 2013

NPACT

All ages

All cause

Zanobetti et al., 2014

121 communities

65+

All cause

aACS (American Cancer Survey), NPACT (National Particle Components Toxicity). See Appendix C Table C-1 for
additional study details.

3.4.1.3 Air Quality Scenarios

We first estimate health risks associated with air quality adjusted to simulate "just
meeting" the current primary PM2.5 standards (i.e., the annual standard with its level of 12.0
|ig/m3 and the 24-hour standard with its level of 35 |ig/m3). We then use air quality modeling to
simulate air quality just meeting an alternative standard with a level of 10.0 |ig/m3 (annual) and
30 |ig/m3 (24-hour). In addition to the model-based approach, for the subset of 30 areas
controlled by the annual standard we also employ linear interpolation and extrapolation to
simulate just meeting alternative annual standards with levels of 11.0 (interpolated between 12.0
and 10.0 |j,g/m3), 9.0 |ag/m3, and 8.0 |j,g/m3 (both extrapolated from 12.0 and 10.0 |ag/m3) 43
Figure 3-15 provides an example of the interpolation and extrapolation calculations performed
for a single grid cell. In this example grid cell, modeled annual PM2.5 concentrations are 11.23

41 Epidemiologic studies tend to attribute risk to either long- or short-term PM2 5 exposures, but rarely to both,
leading to uncertainties in the relationship between health effects from long- and short-term exposures. When
biologically plausible pathways leading to health effects are similar, estimates of impacts from long-term
exposures may include impacts due to short-term exposures and vice-versa. However, if pathways diverge,
impacts due to long- and short-term exposures may be the sum, or even greater than the sum, of the two exposure
durations.

42 While the 2019 ISA also found that evidence supports the determination of a "causal relationship" between long-
and short-term exposures and cardiovascular effects, cardiovascular mortality was not included as a health
outcome as it will be captured in the estimates of all-cause mortality.

43 Modeled air quality surfaces are simulated to just meet standards at the design value monitors and not necessarily
in all grid cells. As the extrapolated alternative annual standard decreases, the proportion of grid cells at or above
the modeled standard increases. Appendix Figure C-31 provides the full distribution of grid cell concentrations at
each modeled and extrapolated standard.

3-138

-------
when the corresponding design value monitor just meets the current annual standard and 9.87
when the corresponding design value monitor just meets the alternative annual standard of 10.0
Hg/m3. The interpolated and extrapolated values for the example grid cells are provided in green
and blue text, respectively.44

11.23

attainment of attainment of attainment of attainment of attainment of

current alternate alternative alternate alternate

standard (12) standard (11) standard (10) standard (9) standard (8)

Figure 3-15. Illustration of approach to adjusting air quality to simulate just meeting
annual standards with levels of 11.0, 9.0, and 8.0 jig/m3.

There is greater uncertainty regarding whether a revised 24-hour standard (i.e., with a
lower level) is needed to further limit "peak" PM2.5 concentration exposure45 and whether a
lower 24-hour standard level would most effectively reduce PM2.5-associated health risks
associated with "typical" daily exposures. However, we do estimate health risks associated with
air quality adjusted to meet a revised 24-hour standard with a level of 30 |ig/m3, in conjunction
with estimating the health risks associated with meeting a revised annual standard with a level of
10 |ig/m3 4647

3.4.1.4 Model-Based Approaches to Adjusting Air Quality

Air quality modeling was used to develop 12 km gridded PM2.5 concentration fields for
the risk assessment in the 2020 PM PA, and the same air quality simulations used in that

44 Modeling to "just meet" annual standards involves adjusting the design value monitor to the standard, and not
necessarily all grid cells modeled. Therefore, it is possible to have estimated PM2.5 concentrations above the
annual standard modeled in individual grid cells.

45 As noted in section 3.3.2.1, while controlled human exposure studies provided consistent evidence for
cardiovascular effects following PM2.5 exposures for less than 24 hours (i.e., < 30 minutes to 5 hours), exposure
concentrations in the studies were well-above the ambient concentrations typically measured in locations meeting
the existing standards.

46 The simulated air quality surface, which just meets both an alternative annual standard of 10.0 |ig/m3 and
alternative 24-hour standard of 30 |ig/ml was subset into areas that are controlled by either the alternative annual
standard of 10.0 ng/m3 or 24-hour standard of 30 ng/m3 to assess risk associated with just meeting each
alternative standard.

47 We also estimate population risks for recent (i.e., unadjusted) ambient PM2.5 concentrations (Appendix C).

3-139

-------
assessment are used here (U.S. EPA, 2020). Additional information on the generation of air
quality inputs to the risk assessment can be found in Appendix C. 1.4. Briefly, a PM2.5
concentration field for 2015 was developed using a Bayesian statistical model (Downscaler) that
calibrates chemical transport model (CTM) predictions of PM2.5 to surface measurements
(section 2.3.3). As air quality surfaces are adjusted to "just meet" current and alternative
standards before use in the risk assessment, the base year used for air quality has negligible
influence on results and 2015 is an appropriate base year for this analysis. As wildfire influences
on PM2.5 design values are often excluded when judging NAAQS attainment, a limited number
of days clearly associated with summertime wildfires were removed from ambient monitoring
data.

The 2015 PM2.5 concentration field was adjusted using response factors developed from
CTM modeling with emission changes relative to 2015. The modeling approach applies realistic
spatial response patterns from CTM modeling to a concentration field, similar to those used in a
number of recent epidemiologic studies, to characterize PM2.5 concentration fields at 12 km
resolution for study areas. The adjusted concentration fields correspond to:

(1) Just meeting the existing annual and 24-hour standards of 12.0 |ig/m3 and 35 |ig/m3, and

(2) Just meeting potential alternative annual and 24-hour standards of 10.0 |ig/m3 and 30 |ig/m3.

The adjustments to simulate just meeting the current standards and alternative standards
are approximations of these air quality scenarios. In reality, changes in PM2.5 in an area will
depend on what emissions changes occur and the concentration gradients of PM2.5 will vary
across an area accordingly. As the type of emission changes can be difficult to predict, for this
risk assessment, two different adjustment approaches were applied to provide two outcomes that
could represent a potential range of PM2.5 concentration changes in the study areas. The two
adjustment approaches used to guide the generation of these modeled surfaces were selected to
span a wide range of possible PM2.5 spatial response patterns:

• Reductions in primary PM2.5 (Pri-PM)\ This modeling approach simulates air quality
scenarios of interest by preferentially adjusting direct/primary PM emissions. As such, the
changes in PM2.5 tend to be more localized near the direct emissions sources of PM.48

• Reductions in secondary PM2.5 (Sec-PM)\ This modeling approach simulates air quality
scenarios of interest by preferentially adjusting SO2 and NOx precursor emissions to simulate

48 In locations for which air quality scenarios cannot be simulated by adjusting modeled directly emitted PM alone,
modeled SO2 and NOx precursor emissions are additionally adjusted to simulate changes in secondarily formed
PM2 5 (Appendix C, section C.1.4).

3-140

-------
changes in secondary PM2.5. In this case, the reductions in PM2.5 tend to be more evenly
spread across a study area.49

The two distinctly different PM2.5 modeling approaches (Pri-PM and Sec-PM) reflect
antithetical PM2.5 spatial response patterns that could cause PM2.5 concentrations to just meet
NAAQS, and real-world spatial response patterns would likely fall somewhere between the two
modeling approaches; we view them as bounding scenarios. As such, air quality surfaces
generated using these two approaches are not additive. Rather, they should be viewed as two
different broad strategies for adjusting ambient PM2.5 concentrations.

3.4.1.5 Study Area Selection

The following factors were considered most important when selecting U.S. study areas
for inclusion in the risk assessment:

• Available Ambient Monitors: There is a complex relationship between monitor locations and
hybrid modeling approaches, with hybrid modeling methods performing better in areas with
more monitors. We have greater confidence in estimating and simulating air quality
concentrations over urban areas with relatively dense ambient monitoring networks, as the
modeled air quality surfaces can be directly compared with monitored concentrations
(additional detail available in Appendix C, section C.1.4). In addition, comparing results
from the two adjustment approaches using very different spatial response patterns (Pri-PM
and Sec-PM) can inform as to the magnitude of variability with regards to monitor
prevalence and location.

• Geographical Diversity. Risk assessments including areas that represent a variety of regions
across the U.S. and a substantial portion of the U.S. population can be more representative.

• Ambient PM2.5 Air Quality Concentrations: Based on 2014-2016 design values, only 16
CBSAs50, also called urban study areas here, exceeded either or both the current annual and
24-hour PM2.5 NAAQS. To include a larger portion of the U.S. in this risk assessment, we
also identified CBSAs with ambient PM2.5 concentrations below, but near, the current annual
and/or 24-hour PM2.5 NAAQS. Inclusion of such areas in the risk assessment necessitates an
upward adjustment to PM2.5 air quality concentrations in order to simulate just meeting the
current standards. Given uncertainty in how such increases could potentially occur, we select
areas requiring a relatively modest upward adjustment (i.e., no more than 2.0 |j,g/m3 for the
annual standard and 5 |j,g/m3 for the 24-hour standard, based on the 2014-2016 design value
period). Areas that appeared to be strongly influenced by exceptional events were also
excluded (section C.1.4). Using these criteria, 47 urban study areas were identified, which
include nearly 60 million people aged 30-99, or approximately 30% of the U.S population in
this age range (Figure 3-16 and Appendix C, section C.1.3). Of the 47 study areas, there were

49 In locations for which air quality scenarios cannot be simulated by adjusting modeled precursor emissions alone, a
proportional adjustment of air quality is subsequently applied. This behavior occurs in areas where emission
changes in addition to NOx and S02 would be needed to adjust design values to just meet the standard.

(Appendix C, Figure C-19).

511 CBSAs (core-based statistical areas) can include one or more counties. Each CBS A selected included at least one
monitor with valid design values and several CBSAs had more than 10 monitors. See Table C-3 in Appendix C.

3-141

-------
30 study areas where just meeting the current standards is controlled by the annual
standard,51 11 study areas where just meeting the current standards is controlled by the daily
standard,52 and 6 study areas where the controlling standard differed depending on the air
quality adjustment approach (Figure 3-16).53

u PS

Jm 1 / l I 1 \ k K—v__j i 1 • I J // I

\y-TVvJ rn ri rh n /7y^\

I * jP

^ 3 * **

Number of Urban Study
Areas (CBSAs)

Controlling
Standard

Population (>30
years old)

Annual (Blue)

-50M

Daily (Green)

~4M

Mixed (Grey)

~5M

Total: 47

-60M

Figure 3-16. Map of 47 urban study areas included in risk modeling.

51 For these areas, the annual standard is the "controlling standard" because when air quality is adjusted to simulate
just meeting the current or potential alternative annual standards, that air quality also would meet the 24-hour
standard being evaluated.

52 For these areas, the 24-hour standard is the controlling standard because when air quality is adjusted to simulate
just meeting the current or potential alternative 24-hour standards, that air quality also would meet the annual
standard being evaluated. Some areas classified as being controlled by the 24-hour standard also violate the
annual standard.

m In these 6 areas, the controlling standard depended on the air quality adjustment method used and/or the standard
scenarios evaluated.

3-142

-------
3,4.1.6 At-Risk Analysis

To inform conclusions regarding the primary PM2.5 standards that are "requisite" to
protect public health (i.e., neither more nor less stringent than necessary; section 1.2) and provide
an adequate margin of safety, it is important to consider the health risks of specific populations
identified as at increased risk (at-risk) that would be allowed under current and alternative
standards, recognizing associated uncertainties (section 3.4.1.8). We use the term "at-risk" to
denote populations and lifestages for which the 2019 ISA and ISA Supplement found scientific
evidence indicating disproportionately increased risk. Our consideration of estimated risks
among potentially at-risk populations focuses on addressing the following policy-relevant
questions:

• How does PM2.5 exposure and risk compare between demographic groups when air
quality just meets the current and potential alternative primary PM2.5 annual
standards?

• To what extent are impacts estimated to change within each demographic group
when air quality is adjusted to just meet potential alternative annual standards with
lower levels?

Assessing PM2.5-attributable risk stratified by the value of another covariate (e.g., race or
ethnicity) can provide insight into population-specific risk. As described in section 3.3.2, the
2019 ISA and ISA Supplement cite extensive evidence indicating that "both the general
population as well as specific populations and lifestages are at-risk for PM2.5-related health
effects" (U.S. EPA, 2019, p. 12-1; U.S. EPA, 2022). Factors that may contribute to increased risk
of PM2.5-related health effects include life-stage (children and older adults), pre-existing diseases
(cardiovascular disease and respiratory disease), race/ethnicity, and socioeconomic status. In
considering the strength of the available scientific evidence and recognizing that this risk
assessment is focused on the health endpoint of mortality, we assess long-term PM2.5-attributable
exposure and mortality risk, stratified by racial/ethnic demographics. Specifically, we evaluate
exposure and risk, stratified by race-specific concentration-response functions when available, of
White, Black, Asian, Native American, Non-Hispanic, and Hispanic individuals.

Concentration-response functions used in this at-risk analysis are from large, multicity
U.S. epidemiologic studies that evaluate the relationship between PM2.5 exposures and mortality.
Eight epidemiologic long-term exposure studies of PM2.5 exposure and all-cause, nonaccidental,
or total mortality in minority populations were identified in the 2019 ISA and ISA Supplement
(U.S. EPA, 2019; U.S. EPA, 2022). Associations from those eight studies relating long-term
PM2.5 exposure and mortality outcomes in minority populations are available in Figure 3-17.

Specific epidemiologic studies and concentration-response functions used here to
estimate risk were identified using criteria that take into account factors such as study design,

3-143

-------
geographic coverage, demographic populations, and health endpoints. Of the studies available
from the 2019 ISA, Di et al., 2017b was identified as best characterizing potentially at-risk
minority populations across the U.S.54 Additional information on input parameters used in the at-
risk analysis can be found in Appendix C, section C.3.

At-risk estimates presented in section 3.4.2.4, when considered alongside estimates of
risk across all populations in the 47 study areas (sections 3.4.2.1, 3.4.2.2, and 3.4.2.3) are meant
to inform conclusions on the primary annual PM2.5 standards that would be requisite to protect
the public health of minority populations potentially at increased risk of long-term PIVh.s-related
mortality effects.

Demographic Population

Citation

Cohort

Location

White

Awad et al., 2019

Medicare

National US

•

Di et al., 2017

Medicare

National US

•

Lipfert et al., 2020

Veterans

31 VA clinics in 27 states

Parker etal., 2018

NHIS

National US

Son et al., 2020

North Carolina

Wang etal., 2017

Medicare

7 southeastern states

•

Wang etal., 2020

Medicare

National US |

•

White (75tn percentile cities) Kioumourtzoglou et al., 2016

Medicare

National US (207 cities)

Black

Awad etal., 2019

Medicare

National US

—•-

Di etal., 2017

Medicare

National US

Lipfert et al., 2020

Veterans

31 VA clinics in 27 states

Parker etal., 2018

NHIS

National US

Son et al., 2020

North Carolina

Wang eta!., 2017

Medicare

7 southeastern states

•

Wang etal., 2020

Medicare

National US J

•—

Black (75th percentile cities)

Kioumourtzoglou et al., 2016

Medicare

National US (207 citiesn

Asian

Di etal.. 2017

Medicare

National US

Son et al., 2020

North Carolina

Wang etal., 2020

Medicare

National US J

—•

Asian (75th percentile cities)

Kioumourtzoglou et al., 2016

Medicare

National US (207 cities) j

Hispanic

Di et al., 2017

Medicare

National US

•-

Parker etal., 2018

NHIS

National US

Son etal., 2020

North Carolina

Wang etal., 2020

Medicare

National US [

j0.85 0.90 0.95 1.00 1.05 1.10 1.15
Hazard/Risk/Odds Ratio (95% CI)

Figure 3-17. Available epidemiologic associations between long-term PM2.5 exposure and
mortality outcomes in demographic populations.55

3.4.1.7 Characterization of Variability and Uncertainty in the Risk Assessment

Both quantitative and qualitative methods have been used to characterize variability and
uncertainty in the risk estimates (Appendix C, section C.3), including:

54 Additional details on concentration-response function identification can be found in Appendix C, section C.3.2. Di
et al., 2017b was identified as best characterizing potentially at-risk minority populations across the U.S. using
study and risk estimate criteria described in the Estimating PM2.5 and Ozone-Attributable Health Benefits TSD
(U.S. EPA, 2022). Additional information on all available at-risk epidemiologic studies is available in Appendix
C, section C.3.2.

55 All studies estimated median or average long-term PM2 5 exposures between 10-12 iig/1113. other than Lipfert and
Wyzga (2020), which reported an approximate average exposure concentration of 14 |ig/nr\ Kioumourtzoglou et
al. (2016) reported associations in cities ranking at or about the 75th percentile proportionally with regards to
demographic population only. VA, Veterans Affairs: NI-IIS, National Health Insurance Service.

3-144

-------
• 95th percentile confidence intervals: We use an iterative Monte Carlo simulation that samples
from the standard error associated with each epidemiologic concentration-response function.
We present the resulting 2.5th and 97.5th percentile values from this distribution as a 95th
percentile confidence interval around the risk estimate. Monte Carlo methods are a well-
established means of characterizing random sampling error associated with concentration-
response functions.

• Health endpoint sensitivity analyses: We include multiple concentration-response functions
reflecting epidemiology studies differing in various ways, such as the population (e.g.,
geographic locations and demographics), exposure estimation methods (e.g., monitor-based
or hybrid techniques), and potential confounders included in the epidemiologic model (e.g.,
ozone).56

• Air quality adjustment sensitivity analyses: We simulate just meeting the current and
alternative standards using two approaches, which represent potential bounding scenarios of
PM2.5 concentration changes across the study areas. The Pri-PM adjustment method
preferentially adjusts direct (i.e., primary, directly-emitted) PM2.5 emissions, whereas the
Sec-PM method preferentially adjusts SO2 and NOx precursor emissions to simulate changes
in secondarily formed PM2.5.

• Qualitative uncertainty assessment: We perform additional qualitative evaluations of the
potential for key sources of uncertainty to impact the magnitude and direction of risk
estimates (Appendix C, section C.3.2).

3.4.1.8 Characterization of Variability and Uncertainty in the At-Risk Analysis

While considering exposure and health risks of individual at-risk racial and ethnic
populations can be policy-relevant, these estimates will be more uncertain than similar estimates
from the overall risk assessment (sections 3.4.2.1 and 3.4.2.2). This is due to additional sources
of uncertainty specific to the at-risk analysis, such as using concentration-response functions
derived from smaller epidemiologic sample sizes, being combined with the sources of
uncertainty that apply to the overall risk assessment. For example, the 2019 ISA and ISA
Supplement find evidence of PM2.5-related health risk disparities specifically demonstrating that
Black populations were at higher risk for PM2.5-related health outcomes, such as mortality.
Therefore, we have higher confidence estimating race-stratified PM2.5 health effects for Black
and White populations than for other races/ethnicities and place those two populations together
at the top of figures and tables. Stratified mortality risk information for other racial and ethnic
populations (i.e., Asians, American Indians, and Hispanics) are provided as there is some
evidence of disparities in PM2.5 exposure and health risk among different racial and ethnic
groups, but the main populations of comparison in the at-risk analysis are Blacks and Whites.

56 Additional information on long-term epidemiologic study identification can be found in the Estimating PM2.5 and
Ozone-Attributable Health Benefits TSD (U.S. EPA, 2021b). Specifically, additional information on the identified
long-term epidemiologic studies can be found in the Study Information Table (U.S. EPA, 2021b).

3-145

-------
The augmentation of existing uncertainness is exemplified by the exposure estimates in
the White populations in the simulated air quality scenarios. White populations make up a
greater proportion of rural areas (-60% vs -80%, USD A, 2018), and rural areas tend to have
lower ambient PM2.5 concentrations. Therefore, as these scenarios are restricted to the 47 urban
study areas, we expect that the average exposure estimated in this assessment is an over-estimate
of the overall national average exposure in the White population.

For characterizing risk in at-risk populations, we used air quality fields from the Pri-PM
adjustment case alone, because the Pri-PM air quality adjustments are largely associated with
emission reductions within the study areas, due to the local nature of air quality impacts from
primary PM sources57. In contrast, Sec-PM air quality adjustments may be strongly associated
with sources located outside of the study areas. Since the at-risk analyses are performed for
population groups within the 47 areas alone, the Pri-PM adjustment case (in which air quality
adjustments are primarily associated with emission sources within the 47 areas) is most
appropriate for this at-risk analysis. However, limiting the analysis to a single simulation
decreases the potential representativeness of simulated PM2.5 concentrations changes across the
study area.

3.4.2 Results of the Risk Assessment

This section presents estimates of PM2.5-associated mortality risks for populations in the
identified urban study areas (additional results available in Appendix C, section C.2).58 Results
are shown as point estimates with 95th percentile confidence intervals for air quality adjusted to
simulate just meeting the current, and potential alternative, standards. We provide tables that
include the total mortality risk associated with air quality just meeting the current or potential
alternative standards, the change in mortality risk (also called delta risk) when moving from air
quality just meeting the current standard to just meeting potential alternative standards, and the
percent risk reduction when moving from air quality just meeting the current standard to just

57 The Pri-PM and Sec-PM adjustment approaches are described in section 3.4.1.4.

58 As the purpose of the risk assessment is to estimate risk allowed when "just meeting" the current or alternative
standards, air quality in the study areas is adjusted up or down to be as high as would be permitted under each
standard. This objective should not be conflated with a risk assessment of recent conditions, which would not
adjust any areas to "just meet" the NAAQS, or an analysis downwardly adjusting only areas exceeding the
NAAQS. While the latter are worthwhile research questions, assumptions that PM concentrations would not
increase would make results difficult to interpret.

3-146

-------
meeting potential alternative standards.59 We also quantify the percent of baseline incidence,
which estimates the percent of total incidence that is associated with ambient PM2.5 exposure
(e.g., percent of mortality attributable to PM2.5 exposure out of all deaths in the specified
population).60 In addition to tables, we provide figures to illustrate how risks are distributed
across annual average ambient PM25 concentrations. Figures present results for all-cause
mortality associated with long-term PM2.5 exposures, based on a key epidemiologic study by
(Turner et al., 2016). Additional results are presented in Appendix C (section C.2).61

The sections below present risk estimates for the full set of 47 urban study areas (section

3.4.2.1), the subset of 30 areas for which the annual PM2.5 standard is controlling (section

3.4.2.2), and the subset of 11 areas for which the 24-hour PM2.5 standard is controlling (section

3.4.2.3). Risk estimates from populations potentially at increased risk of PM-related effects are
available in section 3.4.2.4. Uncertainties in the risk assessment are summarized in section
3.4.2.5.

3.4,2.1 Summary of Risk Estimates for the Full Set of 47 Urban Study Areas

Risk estimates for the 47 urban study areas are presented in Table 3-14 and Table 3-15.
Table 3-14 presents all-cause and non-accidental mortality risk estimates attributable to PM2.5
when just meeting the current primary PM2.5 standards and just meeting either an alternative
modeled annual standard of 10.0 |j,g/m3 or an alternative modeled 24-hour standard of 30 |j,g/m3.
Table 3-14 also provides the percent of total all-cause mortality attributable to PM2.5 in 2015
estimated by each epidemiologic concentration-response function.

Table 3-15 presents the reduction in estimated risk when moving from air quality
scenarios just meeting the current standard to air quality just meeting alternative standards. Areas
are again subset into those just meeting either an alternative annual standard of 10.0 |j,g/m3 or an
alternative 24-hour standard of 30 |ag/m\ based on which standard is controlling in that study
area. Smaller reductions estimated for the alternative 24-hour standard reflect the reduced
number of study areas controlled by the 24-hour standard and the lesser population in those
areas.

59 Total risk refers to risk associated with the full increment of exposure associated with each air quality scenario.
Both delta risk and percent risk reduction reflect the change in risk in going from the current standard to a
specific alternative standard, with delta risk referring to the change in incidence (i.e., premature PM2 5-attributable
mortality) and percent risk reduction referring to the percent change when comparing risk under the current
standard to risk under simulation of an alternative standard. Percent risk reduction is calculated by dividing the
delta risk by the total risk.

60 In other words, the percent of the health effect attributable to PM2 5 exposure. For example, risk results estimate
that 6-8% of all-cause mortality in 2015 was associated with PM2 5 exposure (Table 3-14).

61 As the concentration-response functions used in the risk assessment (section 3.4.1.1) are derived from national-
level analyses, results are correspondingly presented as national estimates.

3-147

-------
Key observations for the full set of 47 study areas from Table 3-14 and Table 3-15, which

include approximately 30% of the U.S. population aged 30-99, are as follows:

• Substantially larger risk reductions are associated with lowering the annual standard than
with lowering the 24-hour standard (Table 3-15). Impacts are estimated to decrease by 13-
17% when air quality is adjusted to just meet an alternative annual standard with a level of
10.0 |j,g/m3 or by 1-2% when adjusted to just meet an alternative 24-hour standard with a
level of 30 |j,g/m3. This corresponds to up to 7,440 (5,040-9,830) fewer deaths per year
attributable to long-term PM2.5 exposures.62

• Up to 45,100 deaths in 2015 are attributable to long-term PM2.5 exposures associated with air
quality just meeting the current annual and 24-hour PM2.5 standards, with a 95th percentile
confidence interval of 30,800-59,000. This constitutes up to 8% of total baseline mortality in
adults aged 30-99 (Table 3-14).

• Short-term PM2.5 exposures are estimated to be associated with up to 3,870 (2,570-5,160)
deaths annually. This accounts for between 0.2-0.7% of mortality in adults ages 30-99 in
2015. Smaller reductions in the absolute number and percent risk of mortality when "just
meeting" only the 24-hour standard in all 47 study areas suggests the annual standard may be
protective of short-term exposure risks. However, the relatively small number of western
study areas included in the risk assessment is a relevant limitation to this conclusion.

62 In most study areas, the risk reductions presented for an annual standard with a level of 10.0 |ig/m3 reflect the
difference between air quality with a maximum three-year annual PM2 5 design value of 12.0 |ig/m3 and air
quality with a maximum three-year annual PM2 5 design value of 10.0 |ig/m3. Similarly, in most study areas, the
risk reduction presented for a 24-hour standard with a level of 30 |ig/m3 reflects the difference between air quality
with a maximum three-year 24-hour PM2 5 design value of 35 |ig/m3 and air quality with a maximum three-year
24-hour PM2 5 design value of 30 |ig/m3. However, in a small number of study areas, the "starting concentration"
for the annual standard are below 12.0 |ig/m3 (four study areas: Riverside-San Bernardino-Ontario, CA; Stockton-
Lodi, CA; Bakersfield, CA; and Hanford-Corcoran, CA) or the starting concentration for the 24-hour standard are
below 35 |ig/m3 (two study areas Pittsburgh, PA and South Bend-Mishawaka, IN-MI:). This is because, in these
areas, the controlling standard for air quality adjusted to just meet the current standards is different from the
controlling standard for air quality adjusted to simulate just meeting the alternative standards evaluated.

3-148

-------
Table 3-14. Estimates of PM2.5-associated mortality for air quality adjusted to just meet the
current or alternative standards (47 urban study areas).

Exposure

Study &
Ages

Simulation Method

Total Mortality
Under the Current
Standard (12/35-0)

% of Baseline
Mortality
Attributable to the

Total Mortality
Under an
Alternative Annual

Total Mortality

Under an
Alternative 24-Hr

Current Standard

Standard (10-0)

Standard (30-0)

Pri PM

40,600

7.4

35,400

40,100

(39,600 to 41,700)

(34,400 to 36,300)

(39,100 to 41,200)

Long-Term

(65-99)

SecPM

41,200
(40,200 to 42,300)

7.5

34,800
(33,900 to 35,700)

40,600
(39,500 to 41,600)

Pri PM

44,400

6.1

38,600

43,900

Turner

(30,300 to 58,200)

(26,300 to 50,700)

(30,000 to 57,500)

(30-99)

SecPM

45,100
(30,800 to 59,000)

6.2

38,000
(25,900 to 49,900)

44,400
(30,300 to 58,200)

Pri PM

2,490

0.4

2,160

2,460

Baxter

(982 to 3,990)

(850 to 3,460)

(970 to 3,950)

(0-99)

SecPM

2,530
(997 to 4,050)

0.4

2,120
(837 to 3,400)

2,490
(982 to 3,990)

Pri PM

1,180

0.2

1,020

1,160

Short-Term

Ito

(-15.8 to 2,370)

(-13.7 to 2,050)

(-15.6 to 2,340)

(0-99)

SecPM

1,200
(-16.0 to 2,400)

0.2

1,000
(-13.5 to 2,020)

1,180
(-15.8 to 2,370)

Pri PM

3,810

0.7

3,300

3,760

Zanobetti

(2,530 to 5,080)

(2,190 to 4,400)

(2,500 to 5,020)

(65-99)

SecPM

3,870
(2,570 to 5,160)

0.7

3,250
(2,160 to 4,330)

3,810
(2,530 to 5,070)

3-149

-------
Table 3-15. Estimated reduction in PM2.5-associated mortality for alternative annual and
24-hour standards (47 urban study areas).

Exposure

Study &
Ages

Simulation
Method

Risk Change When
Moving from the Current
to an Alternative Annual
Standard of 10

Risk Change When
Moving from the Current
to an Alternative 24-Hr
Standard of 30

% Risk Reduction When
Moving fromthe Current
to an Alternative Annual
Standard of 10

% Risk Reduction When
M oving from the Cu rrent
to an Alternative 24-Hr
Standard of 30

Long-Term

Di
(65-99)

PriPM

5,630
(5,400 to 5,780)

501
(488 to 514)

13.9

1.2

Sec PM

6,820
(6,640 to 7,000)

675
(657 to 692)

16.6

1.6

Turner
(30-99)

PriPM

6,120
(4,140 to 8,090)

555
(375 to 734)

13.8

1.2

Sec PM

7,440
(5,040 to 9,830)

714
(483 to 943)

16.5

1.6

Short-Term

Baxter
(0-99)

PriPM

335
(132 to 537)

30.2
(11.9 to 48.4)

13.4

1.2

Sec PM

408
(160 to 654)

38.7
(15.2 to 62.1)

16.1

1.5

Ito
(0-99)

PriPM

158
(-2.12 to 317)

14.4
(-0.194 to 29.0)

13.4

1.2

Sec PM

192
(-2.58 to 386)

18.4
(-0.246 to 36.9)

16.1

1.5

Zanobetf
(65-99)

PriPM

513
(341 to 684)

45.5
(30.2 to 60.7)

13.5

1.2

Sec PM

622
(413 to 830)

61.5
(40.8 to 82.0)

16.1

1.6

3.4.2,2 Summary of Risk Estimates for the 30 Areas Controlled by the Annual
Standard

This section presents the results for the range of alternative annual standard levels for the
30 urban study areas for which the annual standard is controlling under all air quality scenarios
evaluated.63'64 Table 3-16 presents total all-cause and non-accidental mortality risk estimates
attributable to PM2.5 when just meeting the current standard of 12.0 |ig/m3 and just meeting
potential alternative annual standards with levels of 11.0, 10.0, 9.0, and 8.0 |ig/m3. It also
provides the percent of baseline risk attributable to PM2.5 when just meeting the current annual
standard. Table 3-17 presents the reduction in estimated mortality incidence and percent of risk
reduction when moving from air quality scenarios just meeting the current annual standard to air
quality just meeting the various alternative annual standards.

After presenting mortality impact results from the various epidemiologic studies in Table
3-16 and Table 3-17, we focus on a single epidemiologic concentration-response function from

63 These 30 areas controlled by the annual standard under all scenarios evaluated include a population of
approximately 48 million adults aged 30-99, which corresponds to about 75% of the population included in the
full set of 47 areas or approximately 25% of the total U.S. population.

64 Alternative annual air quality surfaces in addition to the modeled surface just meeting 10.0 |ig/m3 were developed
using interpolation and extrapolation of modeled PM2.5 concentrations (section 3.4.1.4 and Appendix C section
C.1.4).

3-150

-------
Turner et al. (2016) to provide additional insight into the distribution of health impacts across
long-term ambient PM2.5 concentrations.65 Figure 3-18 presents distributions of total risk
attributable to annual PM2.5 concentration bins of 1 |ig/m3 when just meeting the current and
alternative annual standards.66 Figure 3-19 presents distributions as a heat map, again binned in 1
|ig/m3increments, associated with moving from just meeting the current standard to just meeting
each alternative annual standard.67

Drawing from the information in Table 3-16, Table 3-17, Figure 3-18, and Figure 3-19
for the subset of 30 study areas (approximately 25% of the U.S. population) in which the annual
standard is controlling, we note the following key observations:

• There is a potential for significant public health impacts in locations just meeting the current
primary PM2.5 standards. The majority of PIVh.s-associated deaths fall well-within the range
of long-term average concentrations over which key epidemiologic studies provide strong
support for reported positive and statistically significant PM2.5 health effect associations.

• Compared to the current annual standards, air quality adjusted to meet alternative annual
standards with lower levels is associated with reductions in estimated all-cause mortality
impacts (i.e., 7-9% reduction for an alternative annual level of 11.0 |ig/m3, 15-19% reduction
for a level of 10.0 |ig/m3, 22-28% reduction for a level of 9.0 |ig/m3, and 30-37%) reduction
for a level of 8.0 |ig/m3) (Table 3-17 and Figure 3-18).

• The magnitude of estimated risk reduction increases as alternative annual standards with
lower levels are simulated, and these estimated risk reductions are associated with lower
ambient PM2.5 concentrations. Specifically, for air quality adjusted to simulate just meeting
an alternative annual standard, the majority of risk reduction occurs in grid cells with
ambient PM2.5 concentrations between the alternative standard and 2 |ig/m3 lower (e.g., for
air quality adjusted to simulate just meeting an annual standard with a level of 8.0 |ig/m3, the

65 The Estimating PM2.5 and Ozone- Attributable Health Benefits TSD details the approach and criteria used to
identify studies and concentration-response functions from the 2019 ISA used in this risk assessment (U.S. EPA,
2021b). Briefly, two studies were again identified as best characterizing mortality risk across the U.S., Di et al.,
2017b and Turner et al., 2016. While both studies used sophisticated techniques to relate PM2 5 exposure and all-
cause mortality across large portions of the U.S population, Di et al., 2017b evaluated Medicare beneficiaries
aged 65+, whereas Turner et al., 2016 included adults ages 30+ from the ACS cohort. The concentration-response
function identified in the Estimating PM2.5 and Ozone- Attributable Health Benefits TSD (U.S. EPA, 2021b) from
Turner et al., 2016 was selected for use in this risk assessment due to the broader age range, although it should be
noted that the concentration-response function from Di et al., 2017b typically generates mortality risk estimates
within approximately 5% of the Turner et al., 2016 concentration-response function.

66 Bins correspond to the lower whole number and include up to, but not including the next whole number. For
example, the bin for 8 |ig/ml includes all risk occurring at PM2 5 concentrations from 8.00 |ig/m3 to 8.99 |ig/m3.
Previously this data was presented as a line graph, which can be found in Appendix C, Figure C-30.

67 As noted above, Figure 3-18 and Figure 3-19 present estimates of all-cause mortality associated with long-term
PM2 5 exposures, based on the study by Turner et al., 2016.

3-151

-------
majority of risk reduction occurs in grid cells with ambient PM2.5 concentrations between 6
and 8 |ig/m3) (Figure 3-18 and Figure 3-19).68

• For air quality just meeting the current annual standard, long-term PM2.5 exposures are
estimated to be associated with as many as 39,000 (26,000-51,000) total deaths from long-
term exposure annually, accounting for approximately 6-8% of baseline mortality.

Table 3-16. Estimates of PM2.5-associated mortality for the current and potential
alternative annual standards in the 30 study areas where the annual standard is
controlling.

Exposure

Study &
Ages

Simulation
Method

Total Risk Under

the Current
Standard (12/35-0)

%of
Baseline
Risk
Attributable
to the
Current
Standard

Total Risk Under

an Alternative
Annual Standard
(11-0)

Total Risk Under

an Alternative
Annual Standard
(10-0)

Total Risk Under

an Alternative
Annual Standard
(9-0)

Total Risk Under

an Alternative
Annual Standard
(8-0)

Long-Term

Di
(65-99)

PriPM

34,900
(34,000 to 35,800)

7.6

32,400
(31,600 to 33,300)

29,900
(29,200 to 30,700)

27,400
(26,700 to 28,100)

24,900
(24,200 to 25,500)

SecPM

35,600
(34,700 to 36,500)

7.7

32,500
(31,700 to 33,300)

29,400
(28,600 to 30,100)

26,300
(25,600 to 26,900)

23,100
(22,500 to 23,700)

Turner
(30-99)

PriPM

38,200
(26,100 to 50,100)

6.3

35,500
(24,200 to 46,500)

32,700
(22,300 to 42,900)

29,900
(20,400 to 39,300)

27,200
(18,500 to 35,700)

SecPM

38,900
(26,600 to 51,000)

6.4

35,500
(24,200 to 46,600)

32,100
(21,900 to 42,100)

28,700
(19,500 to 37,600)

25,200
(17,100 to 33,100)

Short-Term

Baxter
(0-99)

PriPM

2,150
(846 to 3,440)

0.4

1,990
(784 to 3,190)

1,830
(721 to 2,930)

1,670
(658 to 2,680)

1,510
(595 to 2,420)

SecPM

2,190
(862 to 3,510)

0.4

1,990
(785 to 3,190)

1,790
(707 to 2,880)

1,600
(630 to 2,560)

1,400
(552 to 2,250)

Ito
(0-99)

PriPM

1,010
(-13.6 to 2,040)

0.2

939

(-12.6 to 1,880)

864
(-11.6 to 1,730)

789

(-10.6 to 1,580)

713

(-9.57 to 1,430)

SecPM

1,030
(-13.9 to 2,070)

0.2

940

(-12.6 to 1,890)

847
(-11.4 to 1,700)

754

(-10.1 to 1,510)

661

(-8.87 to 1,330)

Zanobetti
(65-99)

PriPM

3,280
(2,180 to 4,370)

0.7

3,040
(2,020 to 4,050)

2,790
(1,860 to 3,730)

2,550
(1,700 to 3,400)

2,310
(1,540 to 3,080)

SecPM

3,340
(2,220 to 4,450)

0.7

3,040
(2,020 to 4,050)

2,740
(1,820 to 3,650)

2,440
(1,620 to 3,260)

2,140
(1,420 to 2,860)

68 Compared to adjusting primary PM2.5 emissions, adjustment of PM precursor emissions resulted in substantially
larger estimated risk reductions at 7 |ig/ml

3-152

-------
Table 3-17. Estimated delta and percent reduction in PM2.5-associated mortality for the
current and potential alternative annual standards in the 30 study areas where the
annual standard is controlling.

Exposure

Study &
Ages

Simulation
Method

Risk Change
When Moving
from the Current
to an Alternative
Annual Standard
of 11

Risk Change
When Moving
from the Current
to an Alternative
Annual Standard
of 10

Risk Change
When Moving
from the Current
to an Alternative
Annual Standard
of 9

Risk Change
When Moving
from the Current
to an Alternative
Annual Standard
of 8

% Risk Reduction

When Moving
from the Current
to an Alternative
Annual Standard
of 11

% Risk Reduction

When Moving
from the Current
to an Alternative
Annual Standard
of 10

% Risk Reduction

When Moving
from the Current
to an Alternative
Annual Standard
of 9

% Risk Reduction

When Moving
from the Current
to an Alternative
Annual Standard
of 8

Long-Term

Di
(65-99)

Pri PM

2,680
(2,610 to 2,750)

5,350
(5,210 to 5,490)

8,000
(7,790 to 8,210)

10,600
(10,400 to 10,900)

7.7

15.3

22.9

30.5

Sec PM

3,320
(3,230 to 3,400)

6,610
(6,440 to 6,780)

9,880
(9,620 to 10,100)

13,100
(12,800 to 13,500)

9.3

18.6

27.8

36.9

Turner
(30-99)

Pri PM

2,920
(1,970 to 3,860)

5,830
(3,940 to 7,700)

8,720
(5,900 to 11,500)

11,600
(7,860 to 15,300)

7.6

15.2

22.8

30.3

Sec PM

3,610
(2,440 to 4,770)

7,200
(4,870 to 9,510)

10,800
(7,290 to 14,200)

14,300
(9,710 to 18,900)

9.3

18.5

27.7

36.8

Short-Term

Baxter
(0-99)

Pri PM

160
(62.8 to 256)

319
(126 to 512)

478
(188 to 767)

638
(251 to 1,020)

7.4

14.9

22.3

29.7

Sec PM

197
(77.6 to 316)

394
(155 to 632)

592
(233 to 948)

789
(310 to 1,260)

9.0

18.0

27.0

36.0

Ito
(0-99)

Pri PM

75.2
(-1.01 to 151)

150
(-2.02 to 302)

226
(-3.03 to 453)

301
(-4.03 to 604)

7.4

14.8

22.3

29.7

Sec PM

93.1
(-1.25 to 187)

186
(-2.49 to 374)

279
(-3.74 to 561)

372
(-4.99 to 748)

9.0

18.0

27.0

36.0

Zanobetti
(65-99)

Pri PM

244
(162 to 325)

487
(324 to 650)

731
(486 to 975)

974
(647 to 1,300)

7.4

14.9

22.3

29.7

Sec PM

301
(200 to 402)

603
(400 to 804)

904
(600 to 1,210)

1,200
(800 to 1,610)

9.0

18.0

27.0

36.0

3-153

-------
Annual PM Concentration (1 (jg/m3 bins) / Simulation Method

12 13

Just meeting
12 |jg/m3

15K
10K
5K

¦ —

Interpolated
to 11 pg/m3

15K
10K
5K

Just meeting
10 ijg/m3

15K

10K
5K

Extrapolated
to 9 pg/m3

15K
10K
5K

Extrapolated
to 8 pg/m3

15K
10K
5K

Pri PM
Sec PM

Pri PM
3ec PM

-c 0
CL V
¦ CD

Pri PM
3ec PM

lAJd 09S
lAld Md

lAld Md

3ec PM

Pri PM
3ec PM

Pri PM

Sec PM

Pri PM

Sec PM

Pri PM
Sec PM
Pri PM
Sec PM

Figure 3-18. Distribution of total risk estimates (P!Vl2.5-attibutable mortality) for the

current and alternative annual standards for the subset of 30 urban study areas where
the annual standard is controlling (blue and green bars represent the Pri-PMi.s and
Sec-PM2.5 estimates, respectively).69

69 Risk is estimated in this figure using Turner et al., 2016. Risk estimates are rounded toward zero into whole PM2.5
concentration values (e.g., risk estimate at 10 |ig/nr includes risk occurring at 10.0-10.9 iig/m3). For each
standard, a small amount of risk is estimated at concentrations higher than the level of the annual standard (e.g.,
some risk is estimated at an average concentration of 13 |ig/m3 when air quality is adjusted to just meet the
current standard). This can result because risk estimates are for a single year (i.e., 2015) within the 3-year design
value period (i.e., 2014 to 2016). While the three-year average design value is 12.0 |ig/m\ a single year can have
grid cells with annual average concentrations above or below 12.0 |ig/m\

3-154

-------
Annual PM Concentration of Lower Standard (1 |jg/m3 bins)

Annual
Standard
Change

Simulation
Method

Sum

12-11

(interpolated)
Mg/m3

Pri PM
Sec PM

0
0

17
26

110
122

381
628

1,534
1,836

763
858

62
89

2,920
3,611

12-10 |jg/m3

Pri PM
Sec PM

1
1

18
23

12
23

81
89

116
287

569
1,632

3,205
3,377

1,720
1,681

103
87

5,826
7,201

12-9

Pri PM

106

596

4,467

3,252

185

8,718

(extrapolated)
Mg/m3

Sec PM

529

2,754

4,953

2,334

10,768

12-8

Pri PM

161

368

5,324

5,408

238

11,595

(extrapolated)
Mg/m3

Sec PM

129

1,116

3,527

6,390

3,101

14,314

Figure 3-19. Distribution of the difference in risk estimates between the current annual
standard (level of 12.0 jig/m3) and alternative annual standards with levels of 11.0,10.0,
9.0, and 8.0 jig/m3 for the subset of 30 urban study areas where the annual standard is
controlling.70

3.4.2.3 Summary of Risk Estimates for the 11 Areas Controlled by the 24-Hour
Standard

Table 3-18 presents annual risk information for the subset of 11 urban study areas in
which the 24-hour standard controls the simulated attainment of all modeled standard levels.71
For air quality just meeting the current 24-hour standard, PM2.5 exposures are estimated to be
associated with as many as 2,570 (1,750-3,370) deaths annually, accounting for up to 7% of the
baseline mortality in those 11 areas. Compared to the current standard, air quality just meeting an
alternative 24-hr standard with a level of 30 |ig/m3 is associated with reductions in estimated risk
of 9-13%.

711 Risks are presented as integers rounded to three significant digits and aggregated into 1 ng/m3 bins. Bins begin at
the whole number value indicated and include values up to, but not including the next whole number (e.g., risk
occurring at PM concentrations of 6.00 to 6.99 are shown in the bin at 6). Risk is estimated in this figure using
Turner et al., 2016.

71 These 11 areas controlled by the 24-hour standard under all scenarios evaluated include a population of

approximately 10 million adults aged 30-99, or about 17% of the population included in the full set of 47 areas.

3-155

-------
Table 3-18. Estimates of PM2.5-associated mortality for the current 24-hour standard, and
an alternative, in the 11 study areas where the 24-hour standard is controlling.

Exposure

Study &
Ages

Simulation
Method

Total Risk
Under the
Current
Standard (12/35-
0)

%of
Baseline

Total Risk Under

an Alternative
Annual Standard
(30-0)

Risk Change When
Moving from the

Current to an
Alternative 24-Hr
Standard of 30

% Risk Reduction
When Moving from
the Current to an
Alternative 24-Hr
Standard of 30

Di
(65-99)

Pri PM

2,320
(2,260 to 2,380)

6.7

2,040
(1,990 to 2,090)

304
(296 to 312)

13.1

SecPM

2,300
(2,250 to 2,360)

6.7

2,100
(2,050 to 2,150)

218
(212 to 224)

9.4

Long-Term

Turner
(30-99)

Pri PM

2,570
(1,750 to 3,370)

5.6

2,250
(1,530 to 2,960)

334
(226 to 442)

13.0

SecPM

2,550
(1,740 to 3,340)

5.6

2,320
(1,580 to 3,050)

241
(163 to 318)

9.4

Short-Term

Baxter
(0-99)

Pri PM

142
(56.1 to 228)

0.3

124
(49.0 to 199)

18.1
(7.11 to 29.0)

12.7

SecPM

141
(55.6 to 226)

0.3

128
(50.5 Id 206)

13.0
(5.12 to 20.9)

9.2

Ito
(0-99)

Pri PM

68.6
(-0.920 to 138)

0.1

59.9
(-0.803 to 120)

8.70
(-0.117 to 17.5)

12.7

SecPM

68.0
(-0.912 to 137)

0.1

61.8
(-0.828 to 124)

6.25
(-0.0838 to 12.6)

9.2

Zanobetti
(65-99)

Pri PM

217
(145 Id 290)

0.6

190
(126 to 253)

27.7
(18.4 to 36.9)

12.7

SecPM

216
(143 Id 287)

0.6

196
(130 to 261)

19.8
(13.1 to 26.4)

9.2

3.4.2.4 Summary of Risk Estimates for At-Risk Populations

Current scientific evidence indicates that some populations, such as different racial/ethnic
groups, face higher health burdens from PM2.5, including for higher levels of exposure and for
increased risk of adverse health responses to a given level of exposure (section 3.3.2). Therefore,
we evaluate how changes to, or retention of, the current standard could impact different
racial/ethnic populations evaluated in this at-risk analysis. Given that this risk and exposure
assessment focuses on mortality endpoints, a quantitative assessment is supported by evidence in
the 2019 ISA and ISA Supplement for racial and ethnic differences in PM2.5 exposures and in
PM2.5-related health risk supports a quantitative assessment (U.S. EPA, 2019, section 12.5.4;
U.S. EPA, 2022, section 3.3.3.2).72 Evidence strongly supports that populations such as Blacks

72 For characterizing risk in at-risk populations, we used air quality fields from the Pri-PM adjustment case alone. In
the Pri-PM case, the air quality adjustments for a given area are largely associated with emission reductions
within that area due to the local nature of air quality impacts from primary PM sources. For the Sec-PM case, the
air quality adjustments may be strongly associated with sources located outside of the area. Since the at-risk
analyses are performed for population groups within the 47 areas alone, the Pri-PM adjustment case (in which air
quality adjustments are primarily associated with emission sources within the 47 areas) is most appropriate for the
at-risk analysis.

3-156

-------
and Hispanics, have higher PM2.5 exposures than White and non-Hispanic populations,
respectively, thus contributing to increased risk of PM-related effects. Additionally, Di et al.,
2017b provides race- and ethnicity-stratified concentration-response functions for ages 65 and
over. Therefore, we quantitatively assess risk for certain racial and ethnic populations of older
adults in the full set of 47 areas and the subset of 30 areas controlled by the annual PM2.5
standard under all Pri-PM air quality simulations evaluated.73 As the 2019 ISA and ISA
Supplement find evidence that Black populations are at higher risk for PM2.5-related health
outcomes, we have higher confidence estimating race-stratified PM2.5 health effects for Black
and White populations than for other races/ethnicities. To visualize the increased confidence in
health estimates of those two populations, we place them at the top of figures and tables in the at-
risk analysis. Additional information on this at-risk analysis is available throughout Appendix C,
section C.2.

For this analysis, we first compare the estimated changes in air quality occurring within
each demographic population when just meeting current and alternative annual PM2.5 standards
(Figure 3-20, left side).74 Across all simulated air quality scenarios in the full set of 47 and subset
of 30 study areas, Blacks experience the highest average PM2.5 concentrations of the
demographic groups analyzed. This increase was typically around 2-5% and was highest in
modeling scenarios just meeting the current suite of standards. Native American populations
typically experienced the lowest average PM2.5 concentrations, especially in the full set of 47
study areas. White, Hispanic, and Asian populations were exposed to fairly similar average PM2.5
concentrations, although White populations tended to be at the higher end of that range in the
subset of 30 areas and the lower end of that range in the full set of 47 areas. Additionally, there is
comparatively less disproportionate exposure between demographic populations as the
alternative annual standard decreases.

While exposure is an important aspect to evaluate when considering potentially
disproportionate impacts, risk estimates provide additional information. Notably, risk estimates
also generate information regarding:

• The number of people affected by the air pollution reduction. In this instance, the population
is further divided by demographic group.

73 Each individual is categorized by both race and ethnicity in this analysis. In other words, the sum of White, Black,
Asian, and Native American individuals equals the total population, as well as the sum of Hispanic and non-
Hispanic individuals. Though Di et al., 2017b did not provide a non-Hispanic concentration-response
relationship, results for non-Hispanics appears similar to Whites when the overall concentration-response
relationship was applied to non-Hispanics (Appendix C, Figures C-33 and C-34).

74 Changes in air quality are estimated using the same approach used in the general risk assessment (sections 3.4.2.1,
3.4.2.2, and 3.4.2.3), summarized in section 3.4.1.4 and detailed in Appendix C.

3-157

-------
• The relationship between exposure and health impact baseline incidence rates, or more
specifically, the percentage change in the risk of an adverse health effect due to a one-unit
change in ambient air pollution. These concentration-response functions are generally
derived from epidemiologic studies.

• The average number of people who die in a given population over a given period of time.
This is commonly referred to as the baseline mortality incidence rate.

For this quantitative analysis of demographic populations potentially at increased risk of PM2.5
exposure, we utilize race/ethnicity-specific, or race/ethnicity-stratified, concentration-response
functions and baseline incidence rates, to more accurately estimate risk within each demographic
group.75 Population-normalized mortality risk occurring within each demographic population is
available on the right side of Figure 3-20. Across all scenarios and demographic groups
evaluated, Black populations are associated with the largest PM2.5-attributable mortality risk rate
per 100,000 people. An example of the 95th percentile confidence interval is available in
Appendix Figure C-32.

75 Information on how the race-stratified concentration-response functions and baseline incidence rates impact the
results can be found in Appendix C, section C.4. Briefly, race-stratified concentration-response functions
increased risk estimated in minority populations, with the greatest magnitude increase occurring in Black
populations, and decreased risk estimated in White populations. Race-stratified baseline incidence rates decreased
risk estimated in all demographic populations analyzed, with the greatest magnitude decreases occurring in White
and Black populations.

3-158

-------
Study Modeling
Areas Scenario

47 areas Just meeting
12/35 pg/m3

•White
Black •
• Hispanic
Asian

Native American

•White

Black •

• Hispanic

Asian

Native American

Just meeting
10/30 pg/m3

•White

• Black
• Hispanic
Asian
Native American

•White

• Black

• Hispanic
Asian
Native American

30 areas Just meeting
12/35 pg/m3

•White
Black •
Hispanic*

Asian

Native American

•White

Black •

• Hispanic

Asian

Native American

Interpolated
to 11 pg/m3

• White
• Black
• Hispanic
Asian

Native American

•White

• Black

• Hispanic

Asian

« Native American

Just meeting
10/30 |jg/m3

•White

• Black
• Hispanic
Asian

Native American

•White

• Black

• Hispanic
Asian
Native American

Extrapolated
to 9 |jg/m3

•White

• Black
• Hispanic
Asian

Native American

•White

• Black

• Hispanic
Asian
Native American

Extrapolated
to 8 |jg/m3

•White

• Black
• Hispanic
Asian

Native American

• White

• Black

• Hispanic
Asian
Native American

7.5 8^0 8.5 a0 9.5 10.0 10.5 11.0 11.5
Average PM Concentration (|jg/m3)

200 300 400 500 600 700 800
Average Mortality Risk Rate (per 100k)

Figure 3-20. Average PM2.5 exposure concentration and PIVh.s-attributable risk estimates
by demographic population when just meeting current or alternative P1VI2.5 standards.

We next estimate demographic-specific average exposure and risk changes when
modeled air quality shifts from just meeting the current annual standard to just meeting potential
alternative annual standard scenarios (Figure 3-21). Simulated PM2.5 concentration reductions
are shown on the left side of the figure and reductions in population-normalized mortality risk
are shown on the right side. As the alternative annual PM standard decreases in the subset of 30
areas controlled by the annual standard, the average reduction in PM2.5 concentration and
mortality risk rates increase across all demographic populations assessed.

3-159

-------
Study Modeling
Areas Scenario

47 areas 12/35-10/30
pg/m3

30 areas 12/35-11

(interpolated)
|jg/m3

•White
• Black
• Hispanic

Asian
Native American

•White

• Black

• Hispanic
Asian

Native American

•White

• Black

• Hispanic
Asian

Native American

•White

• Black
• Hispanic
Asian

• Native American

12/35-10/30
|jg/m3

•White

• Black

• Hispanic
Asian

Native American

•White

• Black

• Hispanic
Asian

Native American

12/35-9

(extrapolated)

Mg/m3

•White

• Black

• Hispanic
Asian

i Native American

•White

• Black

• Hispanic
Asian

• Native American

12/35-8

(extrapolated)

|jg/m3

•White
Black •
Hispanic •
Asian
Native American

•White

Black*

• Hispanic

Asian
Native American

10 15 2.0 2.5 3^0 3^5
Average PM Concentration Reduction (Mg/m3)

50 100 150 200 250
Average Mortality Risk Rate Reduction (per 100k)

Figure 3-21. Average change in PM2.5 exposure concentration and PM2.5-attributable
mortality risk estimates by demographic population when moving from the current to
alternative PM2.5 standards.

We also directly compare the reductions in average national PM2.5 concentrations and
risk rates within each demographic population. Table 3-19 and Table 3-20 provide the percent of
national average PIVh.s-attributable exposures and risk reductions, when shifting from the current
annual PM2.5 standard (12.0 ug/nr) to potential alternative annual PM2.5 standards (11.0 |ig/m\
10.0 |ig/m3, 9.0 pg/m\ and 8.0 jag/m3). The percent PM2.5 and risk reductions are greater in the
Black population than in the White population for each alternative standard evaluated for both
the full set of study areas and the subset controlled by the annual standard. Additionally, the
difference in percent risk reduction increases more in Blacks than in Whites as the potential
alternative annual standard decreases. In other words, Blacks will experience proportionally
greater benefit from successively lower annual standards, although even at an annual standard of
8 |ig/m ' Blacks will experience higher rates of premature mortality risk from PM2.5 exposure
than Whites.

3-160

-------
Table 3-19. Average national percent PM2.5 reduction in demographic populations aged 65
and over residing in the full set of 47 study areas and subset of 30 study areas controlled
by the annual standard.

Ethnicity & Race

% PM Reduction
from 12 pg/m3 to 11
(interpolated) |jg/m3

% PM Reduction from
12 pg/m3 to 10 pg/m3

% PM Reduction from

12 pg/m3 to 9
(extrapolated) |jg/m3

% PM Reduction
from 12 pg/m3 to 8
(extrapolated) |jg/m3

30 areas

47 areas

30 areas

White

Black

Hispanic

Asian

Native American

Table 3-20. Average national percent PM2.5 risk reduction in demographic populations
aged 65 and over residing in the full set of 47 study areas and subset of 30 study areas
controlled by the annual standard.

% Risk Reduction

Ethnicity & Race

from 12 pg/m3 to 11

from 12 pg/m3 to 10

from 12 pg/m3 to 9

from 12 pg/m3 to 8

(interpolated) |jg/m3

|jg/m3

(extrapolated) |jg/m3

30 areas

47 areas

30 areas

White

Black

Hispanic

Asian

Native American

While average exposure concentrations and risk estimates across demographic
populations can convey some insight regarding whether certain populations may be
disproportionately impacted, distributional information, while more complex, can provide a more
comprehensive understanding of the analytical results. As such, we compare both estimated
PM2.5 exposures and mortality risk rates per 100k individuals to the running sum of each
demographic population. To permit the direct comparison of demographic populations with
different absolute numbers, populations are expressed as a percentage in Figure 3-22 and Figure
3-23.76

In both Figure 3-22 and Figure 3-23, PM2.5 concentration information is on the left side
and mortality risk estimates are on the right side. Recent conditions (2015) information for both
exposure and risk can be found in Appendix C, section C.4, as well as sensitivity analyses

76 Information on the absolute number of all-cause premature mortality cases within each racial and ethnic
population demographic can be found in Appendix C Tables C-12 and C-13.

3-161

-------
investigating the impact of race-stratified concentration-response functions and baseline
incidence rates on the results. Cumulative distribution plots of PM2.5 concentrations and
population-normalized mortality risk reductions when shifting from the current to an alternative
annual standard are available in Figure 3-23. Figure 3-22 shows that under the hypothetical air
quality scenarios, disparities exist between Black and White populations with regards to both
PM2.5 exposures and PM2.5-attributable mortality risk rates under the current PM NAAQS.
Figure 3-23 shows that in absolute terms, the Black population is predicted to experience larger
reductions in both PM2.5 exposures and PM2.5-attributable mortality risk rates as the standard is
lowered. Table 3-19 and Table 3-20 show that minority populations are estimated to also
experience larger reductions in PM2.5 exposures and PIVh.s-attributable mortality risk in
relative/proportional terms. When considering the lowest alternative annual standard evaluated,
an alternative annual standard of 8 |ig/m3, disparities in exposure are virtually eliminated,
whereas disparities in mortality risk remain, due to the concentration-response relationship
identified from Di et al. (2017b).

3-162

-------
Study Modeling
Areas Scenario

¦6 100%-

areas

Just meeting
12/35 ^g/m3

Just meeting
10/30 Ljg/m3

Just meeting
12/35 Mg'm3 Jf

Interpolated
to 11 pg/rn3

Just meeting
10/30 Mg/m3

Extrapolated
to 9 (jg/m3

Extrapolated
to 8 [jg/nr

Ethnicity & Race

¦ White

¦ Black

¦ Hispanic
Asian

Native American

7 8 9 10 11 12
PM Concentration (pg/m3) *

13 14

200 400 600 800
Mortality Risk Rate (per 100k) *

1000

Figure 3-22. P1VI2.5 exposure concentrations and PMj.s-attributable mortality risk estimates
by demographic population when just meeting current or alternative PM2.5 standards.

3-163

-------
Study Modeling
Areas Scenario

47 12/35-10/30
areas pg/m3

12/35-11
30 (interpolated)
areas pg/m3

12/35-10/30
pg/m3

12/35-9

(extrapolated)

pg/m3

§ C 75%-

03 .2
Q- "c5

CD "5 50%-

S Q.

15 f?

I a 25%-

3 0%-
¦s 100%-

! c 75%-

s s

« -§ 50%

1 °

| 25%

" 0%-
•S 100%

§ = 75%-

03 J

% 3 50%

> CL

15 °

| ^ 25%

" 0%
"o 100%

§ c 75%-

25%-

"o 100%-

I c 75%

12/36-8 cl -s

(extrapolated)
-------
• Evaluating multiple methods for simulating air quality scenarios: The approach used to
adjust air quality (i.e., Pri-PM and Sec-PM adjustments) has some impact on overall
estimates of risk (e.g., Table 3-14). However, the adjustment approach has a larger impact on
the distribution of risk reductions, particularly for alternative annual levels of 9.0 and 8.0
Hg/m3 (Figure 3-19).

• Characterizing the 95th percentile confidence intervals associated with risk estimates: There
is considerable variation in the range of confidence intervals associated with the point
estimates generated for this analysis (Table 3-14), with some concentration-response
functions displaying substantially greater variability than others (e.g., short-term PM2.5
exposure and all-cause mortality based on effect estimates from Ito et al. (2013) versus long-
term PM2.5 exposure all-cause mortality estimates based on Turner et al., 2016. There are a
number of factors potentially responsible for the varying degrees of statistical precision in
effect estimates, including sample size, exposure measurement error, degree of control for
confounders/effect modifiers, and variability in PM2.5 concentrations evaluated in the original
epidemiologic study.

• Qualitative assessment of additional sources of uncertainty: Based in part on WHO (2008)
guidance and on guidance documents developed by the EPA (U.S. EPA, 2001, U.S. EPA,
2004), we also completed a qualitative characterization of sources of uncertainty including an
assessment of both the magnitude and direction of impact of those uncertainties on risk
estimates. The classification of the magnitude of impact for sources of uncertainty includes
three levels: (a) low (unlikely to produce a sufficient impact on risk estimates to affect their
interpretation), (b) medium (potential to have a sufficient impact to affect interpretation), and
(c) high (likely to have an impact sufficient to affect interpretation). For several of the
sources, we provide a classification between these levels (e.g., low-medium, medium-high)77
The below uncertainties, as well as various additional sources of uncertainty, are detailed in
the EstimatingPM2.5 and Ozone- Attributable Health Benefits TSD (U.S. EPA,

202 lb). Sources of uncertainty with at least a low classification as to the magnitude of
potential impacts include the following (from Appendix C, Table C-32):78

- Use of air quality modeling to adjust PM2.5 concentrations: The baseline and
adjusted air quality concentration fields were developed using modeling to fill
spatial and temporal gaps in monitoring and explore "what if' scenarios. State-of-
the-science modeling methods were used, but modeling-related biases and errors
introduce uncertainty into the PM2.5 concentration estimates. In addition, due to
the national scale of the assessment, scenarios are based on changing modeled
emissions of primary PM2.5 or NOx and SO2 from all anthropogenic sources
throughout the U.S. by fixed percentages. Although this approach tends to target

77 Additional information is available in Appendix C, section C.3.

78 We also identified several additional factors judged to have less than a medium classification of impact on the risk
estimates generate, including: (a) the temporal mismatch between ambient air quality data characterizing
exposure and mortality in long-term exposure-related epidemiology studies, (b) compositional and source
differences in PM, (c) exposure measurement error in epidemiology studies assessing the relationship between
mortality and exposure to ambient PM2 5, (d) lag structure in short-term expo sure-related mortality epidemiology
studies, and (e) assumed causal association between PM and mortality that supports modeling changes in risk
associated with future changes in ambient PM2 5. See Table C-32 in Appendix C for additional discussion of these
sources of uncertainty.

3-165

-------
key emission sources in each study area, it does not tailor emission changes to
specific sources. The two adjustment cases span a wide range of emission
conditions, but these cases are necessarily a subset of the full set of possible
emission scenarios that could be used to adjust PM2.5 concentrations to simulate
"just meeting" standards.

- Use of linear interpolation/extrapolation to adjust air quality: The use of
interpolation and extrapolation to simulate just meeting annual standards with
levels of 11.0, 9.0, and 8.0 |j,g/m3 does not fully capture potential non-linearities
associated with real-world changes in air quality.

- Potential confounding of the PIVh.s-mortality effect: Factors are considered
potential confounders if demonstrated in the scientific literature to be related to
the health effect and correlated with PM2.5. Omitting potential confounders from
analyses could either increase or decrease the magnitude of PM2.5 effect estimates
(e.g., Di et al., 2017b, supplemental Figure S2). Thus, not accounting for
confounders can introduce uncertainty into effect estimates and, consequently,
into the estimated impacts generated using those effect estimates. Confounders
vary according to study design, exposure duration, and health effect. For studies
of short-term exposures, confounders may include meteorology (e.g., temperature,
humidity), day of week, season, medication use, allergen exposure, and long-term
temporal trends. For studies of long-term exposures, confounders may include
socioeconomic status, race, age, medication use, smoking status, stress, noise, and
occupational exposures. While various approaches to control for potential
confounders have been adopted across the studies used in the risk assessment, and
across the broader body of PM2.5 epidemiologic studies assessed in the 2019 ISA,
no individual study adjusts for all potential confounders (U.S. EPA, 2019, Table
A-l).

- Potential for exposure error: Epidemiologic studies have employed a variety of
approaches to estimate population-level PM2.5 exposures (e.g., stationary monitors
and hybrid modeling approaches). These approaches are based on using measured
and/or predicted ambient PM2.5 concentrations as surrogates for population
exposures. As such, exposure estimates in epidemiologic studies are subject to
exposure error. The 2019 ISA notes that, while bias in either direction can occur,
exposure error tends to result in underestimation of health effects in
epidemiologic studies of PM exposure (U.S. EPA, 2019, section 3.5). Consistent
with this, Hart et al. (2015) reports that correction for PM2.5 exposure error using
personal exposure information results in a moderately larger effect estimate for
long-term PM2.5 exposure and mortality, though with wider confidence intervals.
Error in the underlying epidemiologic studies contributes to uncertainty in the risk
estimates based on concentration-response relationships in those studies. Beyond
the exposure error in concentration-response functions, the use of a different
approach to represent exposures in the risk assessment (i.e., 12 x 12 km gridded
surface based on modeling) could introduce additional error into risk estimates.

Shape of the concentration-response relationship at low ambient PM
concentrations: Interpreting the shapes of concentration-response relationships,
particularly at PM2.5 concentrations near the lower end of the air quality
distribution, can be complicated by relatively low data density in the lower

3-166

-------
concentration range, the possible influence of exposure measurement error, and
variability among individuals with respect to air pollution health effects. These
sources of variability and uncertainty tend to smooth and "linearize" population-
level concentration-response functions, and thus could obscure the existence of a
threshold or nonlinear relationship (U.S. EPA, 2015a, section 6.c).

Additional uncertainties are associated with the at-risk analysis. Importantly, the smaller
population within each demographic group reduces statistical power. As this risk and exposure
assessment focuses on urban areas, demographic groups that primarily reside in rural areas, such
as Native Americans, are underrepresented.

3.4.3 Conclusions of the Risk Assessment

Although limitations in the underlying data and approaches lead to some uncertainty
regarding estimates of PIVh.s-associated risk (summarized in section 3.4.1.7), the risk assessment
estimates that the current primary PM2.5 standards could allow a substantial number of PM2.5-
associated deaths in the U.S. For example, when air quality in the 47 study areas is adjusted to
simulate just meeting the current standards, the risk assessment estimates 40,600-45,100 long-
term PM2.5 exposure-related deaths in a single year, with confidence intervals ranging from
30,300-59,000 deaths (Table 3-14). Additionally, the at-risk assessment estimated that Black
populations may experience disproportionally higher exposures and risk under simulated air
quality conditions just meeting the current primary PM2.5 annual standard as compared to White
populations (section 3.4.2.4).79

Compared to the current annual standard, meeting a revised annual standard with a lower
level is estimated to reduce PIVh.s-associated health risks in the 30 annually-controlled study
areas by about 7-9% for a level of 11.0 |ig/m3, 15-19% for a level of 10.0 |ig/m3, 22-28% for a
level of 9.0 |ig/m3, and 30-37% for a level of 8.0 |ig/m3. (Table 3-17)80 Meeting a revised annual
standard with a lower level may also reduce exposure and risk in Black populations slightly more
so than in White populations in simulated scenarios just meeting alternative annual standards
(section 3.4.2.4).

Revising the level of the 24-hour standard to 30 |j,g/m3 is estimated to lower PM2.5-
associated risks across a more limited population and number of areas then revising the annual

79 Risk estimates in Black populations are largely due to race-specific concentration-response functions.

80 Importantly, as the magnitude of estimated risk reductions increases with lower alternative annual standards,
estimated risk reductions are associated with lower ambient PM2 5 concentrations. Lower PM2 5 concentrations
may less closely align with those observed in the epidemiologic study from which the concentration-response
function was obtained, contributing to uncertainty. Additional information on estimated ambient concentrations of
the original Medicare and ACS cohorts evaluated by Di et al., 2017b and Turner et al., 2016, respectively, can be
found in section 6.1.2.1 of the Estimating PM2.5 and Ozone- Attributable Health Benefits TSD (U.S. EPA, 2021b).

3-167

-------
standard (section 3.4.2.3). Risk reduction predictions are largely confined to areas located in the
western U.S., several of which are also likely to experience risk reductions upon meeting a
revised annual standard.

3.5 CASAC ADVICE AND PUBLIC COMMENTS

This section discusses the CASAC's advice regarding the adequacy of the public health
protection afforded by the current primary PM2.5 standards (Sheppard, 2022). The CASAC's
advice is documented in a letter sent to the EPA Administrator on the draft PA (Sheppard, 2022).
The range of views summarized here generally reflects differing judgments as to the relative
weight to place on various types of evidence, the risk-based information, and the associated
uncertainties, as well as differing judgments about the importance of various PIVh.s-related health
effects from a public health perspective.

In its comments on the draft PA, the CASAC stated that: "[ojverall the CASAC finds the
Draft PA to be well-written and appropriate for helping to 'bridge the gap' between the agency's
scientific assessments and quantitative technical analyses, and the judgments required of the
Administrator in determining whether it is appropriate to retain or revise the National Ambient
Air Quality Standards (NAAQS)" (Sheppard, 2022, p. 1 of consensus letter). The CASAC also
stated that the "[djraft PA adequately captures and appropriately characterizes the key aspects of
the evidence assessed and integrated in the 2019 ISA and Draft ISA Supplement of PIVh.s-related
health effects" (Sheppard, 2022, p. 2 of consensus letter). The CASAC also stated that "[t]he
interpretation of the risk assessment for the purpose of evaluating the adequacy of the current
primary PM2.5 annual standard is appropriate given the scientific findings presented" (Sheppard,
2022a p. 2 of consensus letter).

With regard to the adequacy of the current primary annual PM2.5 standard, "all CASAC
members agree that the current level of the annual standard is not sufficiently protective of
public health and should be lowered" (Sheppard, 2022, p. 2 of consensus letter). Additionally,
"the CASAC reached consensus that the indicator, form, and averaging time should be retained,
without revision" (Sheppard, 2022, p. 2 of consensus letter). With regard to the level of the
primary annual PM2.5 standard, the CASAC had differing recommendations for the best range for
an alternative level. The majority of the CASAC "judge[d] that an annual average in the range of
8-10 [j,g/m3" was most appropriate, while the minority of the CASAC members stated that "the
range of the alternative standard of 10-11 [j,g/m3 is more appropriate" (Sheppard, 2022, p. 16 of
consensus responses). The CASAC did highlight, however, that "the alternative standard level of
10 [j,g/m3 is within the range of acceptable alternative standards recommended by all CASAC
members, and that an annual standard below 12 [j,g/m3 is supported by a larger and coherent body
of evidence" (Sheppard, 2022, p. 16 of consensus responses).

3-168

-------
In reaching conclusions on a recommended range of 8-10 (J,g/m3 for the primary annual
PM2.5 standard, the majority of the CAS AC placed weight on various aspects of the available
scientific evidence and quantitative risk assessment information (Sheppard, 2022, p. 16 of
consensus responses). In particular, these members cited recent U.S.- and Canadian-based
epidemiologic studies that show positive associations between PM2.5 exposure and mortality with
study-reported means below 10 [j,g/m3. Further, these members also noted that the lower portions
of the air quality distribution (i.e., concentrations below the mean) provide additional
information to support associations between health effects and PM2.5 concentrations lower than
the long-term mean concentration. In addition, the CASAC members recognized that the
available evidence has not identified a threshold concentration, below which an association no
longer remains, pointing to the conclusion in the draft ISA Supplement that the "evidence
remains clear and consistent in supporting a no-threshold relationship, and in supporting a linear
relationship for PM2.5 concentrations >8 [j,g/m3" (U.S. EPA, 2021a, section 3.2.2.2.7). Finally,
these CASAC members placed weight on the at-risk assessment as providing support for
protection of at-risk demographic groups, including minority populations.

In reaching conclusions on a recommended range of 10-11 [j,g/m3 for the primary annual
PM2.5 standard, the minority of the CASAC emphasized that there were few key epidemiologic
studies that reported positive and statistically significant health effects associations for PM2.5 air
quality distributions with overall mean concentrations below 9.6 [j,g/m3 (Sheppard, 2022, p. 17 of
consensus responses). In so doing, the minority of the CASAC specifically noted the variability
in the relationship between study-reported means and area annual design values based on the
methods utilized in the studies, noting that design values are generally higher than area average
exposure levels. Further, the minority of the CASAC stated that "uncertainties related to
copollutants and confounders make it difficult to justify a recommendation below 10-11 [j,g/m3"
(Sheppard, 2022, p. 17 of consensus responses). Finally, the minority of the CASAC placed less
weight on the risk assessment results noting large uncertainties, including the approaches used
for adjusting air quality to simulate just meeting the current and alternative standards.

With regard to the current primary 24-hour PM2.5 standard, the CASAC did not reach
consensus regarding the adequacy of the public health protection provided by the current
standard. The majority of the CASAC members concluded "that the available evidence calls into
question the adequacy of the current 24-hour standard" (Sheppard, 2022, p. 3 of consensus
letter), while the minority of the CASAC members agreed with "the EPA's preliminary
conclusion [in the draft PA] to retain the current 24-hour PM2.5 standard without revision"
(Sheppard, 2022, p. 4 of consensus letter). The CASAC recommended that in future reviews, the
EPA also consider alternative forms for the primary 24-hour PM2.5 standard. Specifically, the
CASAC "suggests considering a rolling 24-hour average and examining alternatives to the 98th

3-169

-------
percentile of the 3-year average," pointing to concerns that the computing 24-hour average PM2.5
concentrations using the current midnight-to-midnight timeframe could potentially underestimate
the effects of high 24-hour exposures, especially in areas with wood-burning stoves and
wintertime stagnation (Sheppard, 2022, p. 18 of consensus responses).

The majority of the CAS AC favored revising the level of the primary 24-hour PM2.5
standard and suggested that a range of 25-30 [j,g/m3 would be adequately protective. In so doing,
the CASAC placed weight on the available epidemiologic evidence, including epidemiologic
studies that restricted analyses to 24-hour PM2.5 concentrations below 25 [j,g/m3. These members
also placed weight on results of controlled human exposure studies with exposures close to the
current standard, which they note provide support for the epidemiologic evidence to lower the
standard. These members noted the limitations in using controlled human exposure studies alone
in considering adequacy of the 24-hour standard, recognizing that controlled human exposure
studies preferentially recruit less susceptible individuals and have a typical exposure duration much
shorter than 24 hours. These members also placed "greater weight on the scientific evidence than
on the values estimated by the risk assessment," citing their concerns that the risk assessment
"may not adequately capture areas with wintertime stagnation and residential wood-burning
where the annual standard is less likely to be protective" (Sheppard, 2022, p. 17 of consensus
responses). Furthermore, these CASAC members "also are less confident that the annual
standard could adequately protect against health effects of short-term exposures" (Sheppard,
2022a p. 17 of consensus responses).

The minority of the CASAC agreed with the EPA's preliminary conclusion in the draft
PA to retain the current primary 24-hour PM2.5 standard, without revision. In so doing, these
members placed greater weight on the risk assessment, noting that the risk assessment accounts
for both the level and the form of the current standard and the way attainment with the standard
is determined. Further, they state that the "risk assessment indicates that the annual standard is
the controlling standard across most of the urban study areas evaluated and revising the level of
the 24-hour standard is estimated to have minimal impact on the PIVh.s-associated risks" and that,
because of his, "the annual standard can be used to limit both long- and short-term PM2.5
concentrations" (Sheppard, 2022, p. 18 of consensus responses). Further, these members place
more weight on the controlled human exposure studies, which show "effects at PM2.5
concentrations well above those typically measured in areas meeting the current standards"
which suggest that "the current standards are providing adequate protection against these
exposures" (Sheppard, 2022, p. 18 of consensus responses).

While the CASAC members expressed differing opinions on the appropriate revisions to
the current standards, they did "find that both primary standards, 24-hour and annual, are critical

3-170

-------
to protect public health given the evidence on detrimental health outcomes at both short-term and
long-term exposures including peak events" (Sheppard, 2022, p. 13 of consensus responses).

The CASAC also provided a number of recommendations and suggestions for additional
areas of research to inform future reviews of the primary PM2.5 NAAQS. One key area for
additional research is related to the need for advances in and the use of hybrid modeling
methods, including satellite data and products. They also suggest that such studies "should report
the hybrid model concentration distribution statistics (i.e., 10th, 25th, 50th, 75th, and 90th
percentiles), validation, and statistics comparing hybrid model concentrations with monitoring
concentrations and design values where available" (Sheppard, 2022, p. 14 of consensus
responses). They also support additional reporting of concentration distribution statistics (i.e.,
10th, 25th, 50th, 75th, and 90th percentiles), along with covariance between pollutants, in
epidemiologic studies in general, not just those that use hybrid modeling methods (Sheppard,
2022, p. 15 of consensus responses). Other areas identified by the CASAC for future research
include sources and chemical components of PM2.5 (including accountability studies), life course
epidemiologic and mechanistic studies, and health effects of UFP (Sheppard, 2022, pp. 14-15 of
consensus responses).

We also received a number of public comments on the adequacy of the primary PM2.5
standards. A number of commenters, primarily those from national public health, medical, and
environmental nongovernmental organizations (NGOs) and some state, local, or other
government public health or environmental agencies, supported revising the primary PM2.5
standards. Generally, these commenters supported revising the primary annual PM2.5 standard to
a level below 12 (j,g/m3, and as low as 8 (j,g/m3, and many also supported revising the level of the
primary 24-hour PM2.5 standard within the range of 25-30 (J,g/m3, as recommended by the
majority of the CASAC. Other comments, including those representing industries,
manufacturers, and industry and/or business organizations, as well as some state, local, or other
government public health or environmental agencies, supported retaining the current primary
PM2.5 standards, without revision. Commenters who supported either to retain or revise the
primary PM2.5 standards cited to various aspects of the available scientific evidence and the
results of the risk assessment in providing support for their comments.

3.6 KEY CONSIDERATIONS REGARDING THE ADEQUACY OF THE
PRIMARY PM2.5 STANDARDS

In considering the adequacy of the primary PM2.5 standards, the overarching question we
consider is:

3-171

-------
• Does the scientific evidence and risk-based information support or call into question
the adequacy of the protection afforded by the current primary PM2.5 standards?

To assist us in interpreting the scientific evidence and the results of recent quantitative
risk analyses to address this question, we have focused on a series of more specific questions, as
detailed in sections 3.6.1 and 3.6.2 below. In considering the scientific and technical information,
we consider both the information available at the time of the 2012 and 2020 reviews and
information available in this reconsideration, which have been critically assessed in the 2019 ISA
and the ISA Supplement. In so doing, a key consideration is whether the information in this
reconsideration alters our overall conclusions from the 2020 review regarding health effects
associated with PM2.5 in ambient air.

3.6.1 Evidence-based Considerations

In considering the evidence with regard to the overarching question posed above
regarding the adequacy of the current PM2.5 standards, we address a series of more specific
questions that focus on policy-relevant aspects of the evidence. These questions begin with
consideration of the available evidence on health effects associated with exposure to PM2.5.
(section 3.6.1.1). The subsequent questions consider identification of populations at-risk of
PM2.5-related health effects (section 3.6.1.2), and the exposure durations and levels of PM2.5
associated with health effects (section 3.6.1.3). Important uncertainties associated with the
evidence are considered in section 3.6.1.4.

3.6.1.1 Health Effects Associated with Exposure to PM2.5

In answering the overarching question above, we begin by considering the following
question:

• Is there newly available evidence that indicates the importance of certain particle
characteristics (i.e., components or size fractions) other than PM2.5 mass with regard
to concentrations in ambient air, and potential for human exposures and health
effects?

No newly available evidence has been identified in this reconsideration regarding particle
characteristics, such as components or size fractions, other than PM2.5 mass with regard to
concentrations in ambient air, and potential for health effects. While some studies evaluate the
health effects of particular sources of fine particles, or of particular fine particle components,
evidence from these studies does not identify any one source or component that is a better
predictor of health effects than PM2.5 mass (U.S. EPA, 2019, section 1.5.4). The 2019 ISA
specifically notes that "results of these studies confirm and further support the conclusion of the
2009 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

3-172

-------
strongly related with health effects than PM2.5 mass" (U.S. EPA, 2019, section 1.5.4). In
addition, the evidence for health effects following exposures specifically to the ultrafine fraction
of fine particles continues to be far more limited than the evidence for PM2.5 mass as a whole. As
discussed in the 2019 ISA, the lack of a consistent UFP definition in health studies and across
disciplines, together with the nonuniformity in the exposure metric examined, contribute to such
limitations (U.S. EPA, 2019, section 1.4.3). Thus, as was the case for previous reviews, the
evidence base for health effects of fine particles does not support consideration of other PM
characteristics, such as components, or size fractions. For these reasons, we continue to focus on
the health effects associated with PM2.5 mass.

• Does the available scientific evidence alter our conclusions regarding the nature of

health effects attributable to human exposure to PM2.5 from ambient air?

The scientific evidence, including that assessed in the 2019 ISA and ISA Supplement, is
consistent with the conclusion reached in the previous reviews regarding health effects and PM
exposures where a causal relationship was concluded. Specifically, as in prior reviews, it was
concluded that there is a "causal relationship" between short- and long-term PM2.5 exposures and
mortality and cardiovascular effects (U.S. EPA, 2019, sections 11.1, 11.2, 6.1, 6.2; U.S. EPA,
2022, sections 3.2.1, 3.2.2, 3.1.1, and 3.1.2). Further, a "likely to be causal relationship" was
concluded for short- and long-term PM2.5 exposures and respiratory effects (U.S. EPA, 2019,
sections 5.1 and 5.2). Additionally, conclusions reached in the 2019 ISA differ with regard to
cancer and nervous systems effects and long-term PM2.5 exposure, based on evidence assessed in
the 2019 ISA and it was concluded that there is a "likely to be causal relationship" (U.S. EPA,
2019, sections 10.2 and 8.2). The evidence base is concluded to be "suggestive of, but not
sufficient to infer, a causal relationship" between short- and long-term PM2.5 exposures and
metabolic effects (U.S. EPA, 2019, sections 7.1 and 7.2), reproduction and fertility (U.S. EPA,
2019, section 9.1.1), and pregnancy and birth outcomes (U.S. EPA, 2019, section 9.1.2). In
addition, effects associated with short-term exposure to UFP and cardiovascular (U.S. EPA,
2019, section 6.5), respiratory (U.S. EPA, 2019, section 5.5), and nervous system effects (U.S.
EPA, 2019, section 8.5), as well as long-term exposure to UFP and nervous system effects (U.S.
EPA, 2019, section 8.6) are concluded to be "suggestive of, but not sufficient to infer, a causal
relationship." As in the 2020 review, the strongest evidence, including with regard to quantitative
characterizations of relationships between PM2.5 exposure and effects, is for mortality and
cardiovascular effects.

3.6,1,2 Populations At-Risk of PM2.5-related Health Effects

Populations or lifestages can be at increased risk of an air pollutant-related health effect
due to one or more factors. These factors can be intrinsic, such as physiological factors that may

3-173

-------
influence the internal dose or toxicity of a pollutant, or extrinsic, such as sociodemographic, or
behavioral factors. The questions considered in this section address what the available evidence
indicates regarding which populations are particularly at risk of health effects related to exposure
to PM2.5 in ambient air.

• Does the current evidence alter our understanding of populations that are

particularly at-risk from PM2.5 exposures? Is there evidence that suggests additional
at-risk populations that should be given increased focus for this reconsideration?

The current evidence does not alter our understanding of which populations are
potentially at greater risk from health effects of PM2.5 exposures. As in previous reviews, the
2019 ISA continues to provide support that factors that may contribute to increased risk of
PM2.5-related health effects include lifestage (children and older adults), pre-existing diseases
(cardiovascular disease and respiratory disease), race/ethnicity, and socioeconomic status. Other
factors that have the potential to contribute to increased risk, but for which the evidence is less
clear, include obesity, diabetes, genetic factors, smoking status, sex, diet, and residential location
(U.S. EPA, 2019, chapter 12).

In addition to these population groups, the 2019 ISA and ISA Supplement note that there
is strong evidence for racial and ethnic differences in PM2.5 exposures and PIVh.s-related health
risk. There is strong evidence demonstrating that Black and Hispanic populations, in particular,
have higher PM2.5 exposures than non-Hispanic White populations (U.S. EPA, 2019, Figure 12-
2; U.S. EPA, 2022, Figure 3-38). Further, there is consistent evidence across multiple studies that
demonstrate increased risk of PIVh.s-related health effects, with the strongest evidence for health
risk disparities for mortality (U.S. EPA, 2019, section 12.5.4).

Studies assessed in the 2019 ISA and ISA Supplement also provide evidence of exposure
and health risk disparities based on SES. The evidence indicates that lower SES communities are
exposed to higher concentrations of PM2.5 compared to higher SES communities (U.S. EPA,
2019, section 12.5.3; U.S. EPA, 2022, section 3.3.3.1.1). Additionally, evidence supports the
conclusions that lower SES is associated with cause-specific mortality and certain health
endpoints (i.e., MI and CHF), but less so for all-cause or total (non-accidental) mortality (U.S.
EPA, 2019, section 12.5.3;) U.S. EPA, 2022, section 3.3.3.1).

3.6.1.3 Exposure Concentrations Associated with Health Effects

In answering the overarching question with regard to the adequacy of the primary PM2.5
standards, as described above, we next consider the scientific evidence and the support it
provides for the occurrence of adverse public health effects and the associated exposure
concentrations at which such effects occur. In so doing, we ask the following questions:

3-174

-------
• Does the current evidence alter our conclusions regarding the exposure duration and
concentrations associated with health effects? To what extent does the scientific
evidence indicate health effects attributable to exposures to PM2.5 concentrations
lower than previously reported and what are important uncertainties in that
evidence?

The evidence available in this reconsideration regarding PM2.5 exposures associated with
health effects affirms and strengthens the evidence available at the time of the 2020 review,
taking into account studies that have become available since that time. Consistent with the
evidence available in the 2020 review, and as assessed in the 2019 ISA and the ISA Supplement,
the strong evidence base of epidemiologic studies report associations between long- and short-
term PM2.5 exposures and a variety of outcomes, including mortality and cardiovascular effects.
Additionally, as detailed in section 3.3.1, animal toxicological studies and controlled human
exposure studies continue to provide support understanding the effects of exposure to PM2.5, and
support for biologically plausible mechanisms through which adverse human health outcomes
could occur. In addition, controlled human exposure studies have consistently reported that PM2.5
exposures lasting from less than one hour up to five hours can impact cardiovascular function
and provide some insight into how short-term exposure to PM2.5 may impact cardiovascular
function in ways that could lead to more serious outcomes.

The controlled human exposure studies, as discussed in detail in the 2019 ISA (U.S. EPA,
2019, section 6.1) and summarized above in section 3.3.3.1, have demonstrated effects on
cardiovascular function following PM2.5 exposures ranging from one to five hours, with the most
consistent evidence for impaired vascular function (U.S. EPA, 2019, section 6.1.13.2). In
addition, although less consistent, the 2019 ISA notes that studies examining PM2.5 exposures
also provide evidence for increased blood pressure (U.S. EPA, 2019, section 6.1.6.3), conduction
abnormalities/arrhythmia (U.S. EPA, 2019, section 6.1.4.3), changes in heart rate variability
(U.S. EPA, 2019, section 6.1.10.2), changes in hemostasis that could promote clot formation
(U.S. EPA, 2019, section 6.1.12.2), and increases in inflammatory cells and markers (U.S. EPA,
2019, section 6.1.11.2). The 2019 ISA concludes that, when taken as a whole, controlled human
exposure studies demonstrate that exposure to PM2.5 may impact cardiovascular function in ways
that could lead to more serious outcomes (U.S. EPA, 2019, section 6.1.16). Thus, such studies
can provide insight into the potential for specific PM2.5 exposures to result in physiological
changes that could increase the risk of more serious effects, though the health relevance of the
occurrence of these acute effects is less certain.

To provide some insight into what these studies may indicate regarding the primary PM2.5
standards, air quality analyses examine monitored 2-hour PM2.5 concentrations at sites meeting
the current primary PM2.5 standards (as described in section 2.3.2 and section A.3 of Appendix

3-175

-------
A).81 The 2-hour PM2.5 concentrations to which individuals were exposed in most of these
studies are well-above the ambient concentrations typically measured in locations meeting the
current primary standards. For example, at air quality monitoring sites meeting the current
primary PM2.5 standards (i.e., the 24-hour standard and the annual standard), the 2-hour
concentrations generally remain below 10 (J,g/m3, and virtually never exceed 30 |ig/m3. Two-hour
concentrations are higher at monitoring sites violating the current standards, but generally remain
below 16 [j,g/m3 and virtually never exceeding 80 [j,g/m3. Thus, while controlled human exposure
studies provide support for the biological mechanisms and plausibility of the serious
cardiovascular effects associated with ambient PM2.5 exposures in epidemiologic studies (U.S.
EPA, 2019, chapter 6), the exposures evaluated in most of these studies are well-above the
ambient concentrations typically measured in locations meeting the current primary standards,
and the results are variable across some of the controlled human exposure studies evaluated at
near ambient PM2.5 concentrations.

While controlled human exposure studies provide insight on the exposure concentrations
that directly elicit health effects in humans, uncertainty exists in translating the observations in
animal toxicology studies to potential adverse health effects in humans. The interpretation of the
animal toxicology studies with regard to the potential implications for human health is
complicated by the fact that the concentrations of PM2.5 in animal toxicologic studies are much
higher than those shown to elicit effects in human populations, and there are also significant
anatomical and physiological differences between animal models and humans. Most of the
animal toxicology studies have generally examined short-term exposures to PM2.5 concentrations
from 100 to >1,000 [j,g/m3 and long-term exposures to concentrations from 66 to >400 [j,g/m3
(e.g., see U.S. EPA, 2019, Table 1-2). Two exceptions are a study reporting impaired lung
development following long-term exposures (i.e., 24 hours per day for several months prenatally
and postnatally) to an average PM2.5 concentration of 16.8 [j,g/m3 (Mauad et al., 2008) and a
study reporting increased carcinogenic potential following long-term exposures (i.e., 2 months)
to an average PM2.5 concentration of 17.7 [j,g/m3 (Cangerana Pereira et al., 2011). These two
studies report serious effects following long-term exposures to PM2.5 concentrations close to the
ambient concentrations reported in some PM2.5 epidemiologic studies (U.S. EPA, 2019, Table 1-
2), though still above the ambient concentrations likely to occur in areas meeting the current
primary standards. Thus, as is the case with controlled human exposure studies, animal
toxicology studies support the plausibility of various adverse effects that have been linked to
ambient PM2.5 exposures (U.S. EPA, 2019).

81 In addition, 4-hour and 5-hour PM2 5 concentrations at monitoring sites meeting or violating the current primary
PM2 5 standards were also evaluated (as described in section 2.3.2 and section A.3 of Appendix A).

3-176

-------
Epidemiologic studies in the U.S. and Canada, assessed in the 2019 ISA and ISA
Supplement, continue to report positive and statistically significant associations between long-
and short-term exposure to PM2.5 and mortality and morbidity, including both new studies
evaluated in the ISA Supplement related to total mortality and cardiovascular mortality and
morbidity and studies that examined populations and lifestages that may be at comparatively
higher risk of experiencing a PIVh.s-related health effects (e.g., older adults). Such studies
employ various designs and examine a variety of health outcomes, geographic areas, and
approaches to controlling for confounding variables. With regard to controlling for potential
confounders in particular, key epidemiologic studies use a wide array of approaches. Time-series
studies control for potential confounders that vary over short time intervals (e.g., including
temperature, humidity, dew point temperature, and day of the week) while cohort studies control
for community- and/or individual-level confounders that vary spatially (e.g., including income,
race, age, socioeconomic status, smoking, body mass index, and annual weather variables such
as temperature and humidity) (Appendix B, Table B-4). Sensitivity analyses indicate that adding
covariates to control for potential confounders can either increase or decrease the magnitude of
PM2.5 effect estimates, depending on the covariate, and that none of the covariates examined can
fully explain the association with mortality (e.g., Di et al., 2017b, Figure S2 in Supplementary
Materials). Thus, while no individual study adjusts for all potential confounders, a broad range of
approaches have been adopted across studies to examine confounding, supporting the robustness
of reported associations.

Available studies additionally indicate that PM2.5 health effect associations are robust
across various approaches to estimating PM2.5 exposures and across various exposure windows.
This includes recent studies that estimate exposures using ground-based monitors alone and
studies that estimate exposures using data from multiple sources (e.g., satellites, land use
information, modeling), in addition to monitors. While none of these approaches eliminates the
potential for exposure error in epidemiologic studies, such error does not call into question the
fundamental findings of the broad body of PM2.5 epidemiologic evidence. In fact, the 2019 ISA
notes that while bias in either direction can occur, exposure error tends to lead to underestimation
of health effects in epidemiologic studies of PM exposure (U.S. EPA, 2019, section 3.5).
Consistent with this, a recent study reports that correction for PM2.5 exposure error using
personal exposure information results in a moderately larger effect estimate for long-term PM2.5
exposure and mortality (Hart et al., 2015). While most PM2.5 epidemiologic studies have not
employed similar corrections for exposure error, several studies report that restricting analyses to
populations in close proximity to a monitor (i.e., in order to reduce exposure error) result in
larger PM2.5 effect estimates (e.g., Willis et al., 2003; Kloog et al., 2013). The consistent
reporting of PM2.5 health effect associations across exposure estimation approaches, even in the

3-177

-------
face of exposure error, together with the larger effect estimates reported in some studies that
have attempted to reduce exposure error, provides further support for the robustness of
associations between PM2.5 exposures and mortality and morbidity.

Consistent findings from the broad body of epidemiologic studies are also supported by
an emerging body of studies employing alternative methods for confounder control to further
inform the causal nature of the relationship between long- or short-term term PM2.5 exposure and
mortality (U.S. EPA, 2019, sections 11.1.2.1, 11.2.2.4; U.S. EPA, 2022, sections 3.1.1.3, 3.1.2.3,
3.2.1.3, and 3.2.2.3). These studies, summarized above in Table 3-11, used a variety of statistical
methods to control for confounding bias and consistently report positive associations, which
support the positive and significant effects seen in cohort studies associated with short- and long-
term exposure to PM2.5 and mortality.

In addition to broadening our understanding of the health effects that can result from
exposures to PM2.5 and strengthening support for some key effects (e.g., nervous system effects,
cancer), recent epidemiologic studies strengthen support for health effect associations at
relatively low ambient PM2.5 concentrations. Studies that examine the shapes of concentration-
response functions over the full distribution of ambient PM2.5 concentrations have not identified
a threshold concentration, below which associations no longer exist (U.S. EPA, 2019, section
1.5.3; U.S. EPA, 2022, sections 2.2.3.1 and 2.2.3.2). While such analyses are complicated by the
relatively sparse data available at the lower end of the air quality distribution (U.S. EPA, 2019,
section 1.5.3), the evidence remains consistent in supporting a no-threshold relationship, and in
supporting a linear relationship for PM2.5 concentrations > 8 [j,g/m3. However, uncertainties
remain about the shape of the C-R curve at PM2.5 concentrations < 8 (j,g/m3, with some recent
studies providing evidence for either a sublinear, linear, or supralinear relationship at these lower
concentrations.

Consistent with previous reviews, we note that the use of information from epidemiologic
studies to inform conclusions on the primary PM2.5 standards is complicated by the fact that such
studies evaluate associations between distributions of ambient PM2.5 and health outcomes, and
do not identify the specific exposures that can lead to the reported effects. Rather, health effects
can occur over the entire distribution of ambient PM2.5 concentrations evaluated, and
epidemiologic studies do not identify a population-level threshold below which it can be
concluded with confidence that PM-associated health effects do not occur (U.S. EPA, 2019,
section 1.5.3). However, the study-reported ambient PM2.5 concentrations reflecting estimated
exposure in the middle portion of the PM2.5 air quality distribution, which corresponds to the
bulk of the underlying data, which provide the strongest support for reported health effect
associations and can inform our conclusions on the current and potential alternative standards. In
using this information to inform our conclusions, we recognize that the mean PM2.5

3-178

-------
concentrations reported by key epidemiologic studies differ in how mean concentrations were
calculated (Table 3-6, Table 3-7, Table 3-8, Table 3-9), as well as their interpretation in what
means represent in the context of the current standards. To frame our evaluation of study-
reported mean PM2.5 concentrations, we specifically consider the following question:

• How do the study-reported means from the key epidemiologic studies and the related
air quality analyses that compare study means to area design values inform our
consideration of the level of the current annual PM2.5 standard?

As discussed above, and consistent with the approaches in the 2012 and 2020 reviews (78
FR 3161, January 15, 2013; U.S. EPA, 2011, sections 2.1.3 and 2.3.4.1; 85 FR 82716-82717,
December 18, 2020; U.S. EPA, 2020, sections 3.1.2 and 3.2.3), in this reconsideration, we focus
on study-reported means (or medians) of daily and annual average PM2.5 concentrations over the
entire study period as indicators for the middle portions of the air quality distributions, over
which studies generally provide strong support for reported associations and for which our
confidence in the magnitude and significance of associations observed in the epidemiologic
studies is greatest (78 FR 3101, January 15, 2013). Based on the information above with regard
to the key U.S. and Canadian epidemiologic studies, in this reconsideration, we also focus on the
key U.S. epidemiologic studies that either use monitoring to estimate exposure or that use hybrid
modeling approaches and population weight the study-reported mean. For these studies, the
study-reported means are as follows:

• For key U.S. epidemiologic studies that used monitors to estimate PM2.5 exposures (Figure 3-
8), overall mean PM2.5 concentrations range between 9.9 |j,g/m3 82 to 16.5 |j,g/m3.

• For key U.S. epidemiologic studies that used hybrid model-predicted exposure (Figure 3-10),
the majority of these studies estimate mean PM2.5 exposures by averaging up from the grid
cell spatial resolution used in the modeling approach to the spatial resolution of health study
data (e.g., ZIP code or census tract). This incorporates an aspect of population weighting in
the calculation of the mean.

In studies that average up from the grid cell level to the ZIP code, postal code, or
census tract level, mean PM2.5 concentrations range from 9.8 |j,g/m3 to 12.2 |j,g/m3.
The one study that population-weighted the grid cell prior to averaging up to the
ZIP code or census tract level report mean PM2.5 concentrations of 9.3 |j,g/m3.
In assessing the range of reported exposure concentrations for which we have the
strongest support for adverse health effects occurring, we consider whether this evidence
supports or calls into question the adequacy of the current primary annual PM2.5 standard to
provide adequate protection against these exposure concentrations. This means, as in past
reviews, application of a decision framework based on assessing means reported in key

82 This is generally consistent with, but slightly below, the lowest study-reported mean PM2 5 concentration from
monitor-based studies available in the 2020 PA, which was 10.7 |ig'/m3 (U.S. EPA, 2020, Figure 3-7).

3-179

-------
epidemiologic studies must also consider how the study means were computed and how these
values compare to the annual standard metric (including the level, averaging time and form) and
the use of the monitor with the highest PM2.5 design value in an area for compliance. Based on
the air quality analyses in Chapter 2 and the discussion above, we note that the design values
associated with the study-reported means in these key U.S. based epidemiologic studies are only
somewhat higher: 10-20% for monitor-based studies and 15-18% higher for the studies that
include hybrid modeling approaches and utilize population weighting. Based on these results, we
can generally conclude that the study-reported mean concentrations in the studies are associated
with air quality conditions that would be achieved by meeting annual standard levels that are 10-
20% higher and 15-18%) higher than study-reported means for monitor-based studies and hybrid
modeling-based studies that use population weighting, respectively. Therefore, an annual
standard level that is no more than 10-20%> higher than the study-reported means in the monitor-
based studies (i.e., 9.9-16.5 |ag/m3), and no more than 15-18%) higher than the study-reported
means in the studies that include hybrid modeling approaches and utilize population weighting
(i.e., 9.3-12.2 |ig/m3), would generally maintain air quality exposures to below those associated
with the study-reported mean PM2.5 concentrations, exposures for which we have the strongest
support for adverse health effects occurring. This relationship is indicative of the fact that PM2.5
exposures in an area are represented by a distribution of concentrations across that area, with the
annual standard level at the design value monitor being associated with the highest annual
average exposure concentration for that area.

For the remainder of key U.S.-based epidemiologic studies that also use hybrid modeling
approaches to estimate exposure, these studies calculate the study-reported mean by averaging
from the grid cell spatial resolution across the entire study area, whether that be the nation or a
region of the country. While estimating exposures similar to the other studies that use hybrid
modeling approaches, these studies do not weight the estimated exposure concentrations based
on population density or location of health events in calculating the study-reported mean.

Because of this, these epidemiologic studies report some of the lowest mean values simply due to
the methodology used to calculate the study-reported means but not because exposures used in
the epidemiologic studies were necessarily lower than those in other studies using hybrid
modeling approaches.

• For these studies, the reported mean PM2.5 concentrations range from 8.1 [j,g/m3 to 11.9
[j,g/m3.

For this set of studies, we would expect the associated annual design values to be much
higher (i.e., 40-50% higher) than the study-reported means (i.e., 8.1-11.9 (j,g/m3). This larger
difference between design values and study-reported mean concentrations makes it more difficult
to consider how these studies can be used to determine the adequacy of the protection afforded

3-180

-------
by the current or potential alternative annual standards, compared to monitor-based studies and
hybrid model-based studies that use population weighting for which the difference between
design values and study-reported means is much smaller.

In addition to the key U.S.-based epidemiologic studies, there are also key
epidemiological studies from Canada. This set of studies includes those that estimate exposure
using monitors, as well as hybrid modeling approaches. In addition, all of the studies that use
hybrid modeling data also compute the study-reported means using some aspects of population
weighting. The overall study-reported mean PM2.5 concentrations from the key Canadian
epidemiologic studies are similar to, though somewhat lower than, those from the U.S. studies:

• Range of monitor-based mean PM2.5 concentrations: 6.9 |ig/m3 to 13.3 |ig/m3

• Range of mean PM2.5 concentrations in studies that use hybrid modeling (all of which
average up to postal codes and thus include some aspects of population weighting): 5.9
|ig/m3 to 9.8 |ig/m3

Similar to our examination of the key U.S. epidemiologic studies, we also consider how
the information from key Canadian epidemiologic studies can be used to inform our conclusions
regarding the adequacy of the protection afforded by the current primary annual PM2.5 against
the range of reported exposure concentrations reported in these studies. Consistent with approach
for considering the key U.S.-based epidemiologic studies, we also consider how the study means
were computed in the key Canadian-based epidemiologic studies and how these values compare
to the annual standard metric and the use of the monitor with the highest PM2.5 design value in an
area for compliance. Due to technical challenges and limitations, the air quality analyses in
Chapter 2 did not include an analysis for Canada similar to the analysis in the U.S. comparing
monitoring data and hybrid modeling results. We do note that the analysis using U.S. data found
that the U.S. design values associated with the study-reported means in the key U.S.
epidemiologic studies were generally 10-20% higher for monitor-based studies and 15-18%
higher for the key U.S. epidemiologic studies that include hybrid modeling approaches and
utilize population weighting. We note that the measured annual average PM2.5 concentrations in
U.S. and Canadian cities are fairly similar, while PM2.5 concentrations in rural areas tend to be
lower in Canada than the U.S. Despite the lack of an air quality analysis in Canada, it would not
be unreasonable to assume that the ratio between highest monitor and area average
concentrations (either monitor-based or hybrid model-based estimates) in Canada would be
higher because of the influence of the lower concentrations in rural areas, resulting in a higher
percentage difference than those estimated for the U.S. Given some of these uncertainties, we
recognize challenges in using the study-reported means from the Canadian studies to provide
specific quantitative insight into how the study-reported means in the Canadian studies would
compare to area design values in the U.S.

3-181

-------
In addition to study-reported mean concentrations and consistent with previous reviews
and with the CASAC's recommendation to include other approaches . .for the purpose of
informing the adequacy of the standards, including evaluation of the distribution of
concentrations reported in epidemiology studies, including the median concentration and 25th
percentile concentration, if available" (Sheppard, 2022, p. 8 of consensus responses), we also
consider the subset of key epidemiologic studies that reported PM2.5 concentrations
corresponding to the 25th and 10th percentiles of health data or exposure estimates in answering
the following question:

• How do the study-reported PM2.5 concentrations corresponding to the 25th and 10th
percentiles of health data or exposure estimates provide insight to inform our
consideration of the level of the current annual PM2.5 standard?

In the 2012 review, the 2011 PA noted the interrelatedness of the distributional statistics
and a range of one standard deviation around the mean which contains approximately 68% of
normally distributed data, in that one standard deviation below the mean falls between the 25th
and 10th percentiles (U.S. EPA, 2011 p. 2-71). Given this, the 2011 PA provided information, as
available for a subset of key epidemiologic studies, on the study-reported PM2.5 concentrations
corresponding to the 25th and 10th percentiles of health data or exposure estimates. In that review,
the Administrator placed some weight on studies that provided mean PM2.5 concentrations
around the 25th percentile of the distributions of deaths and cardiovascular-related
hospitalizations and judged the region around the 25th percentile as a reasonable part of the
distribution to guide the decision on the appropriate standard level (78 FR 3161, January 15,
2013). Given the potential for consideration of this information in this reconsideration with
regard to the adequacy of the standard level, we note that of the key epidemiologic studies
evaluated in the 2019 ISA and ISA Supplement, a subset of studies report PM2.5 concentrations
corresponding to the 25th and 10th percentiles of health data or exposure estimates to provide
insight into the concentrations that comprise the lower quartiles of the air quality distributions.

Consistent with previous reviews and as described in section 3.3.3.2.1 above, we
conclude that focusing on concentrations somewhat below the means (e.g., 25th and 10th
percentiles), when such information is available from epidemiologic studies, is a reasonable
approach for considering lower portions of the air quality distribution. We continue to recognize
that the health data are appreciably more sparse and our understanding of the magnitude and
significance of the associations correspondingly become more uncertain in the lower part of the
air quality distribution. We also note that health effects can occur over the entire distribution of
ambient PM2.5 concentrations evaluated, and epidemiologic studies do not identify a population-
level threshold below which it can be concluded with confidence that PM-associated health
effects do not occur (U.S. EPA, 2019, section 1.5.3). However, using values below the 10th

3-182

-------
percentile would lead to even greater uncertainties and diminished confidence in the magnitude
and significance of the associations.

In the key U.S. epidemiologic studies that report the 25th and 10th percentiles of health
events corresponding to mean PM2.5 concentrations (i.e., averaged over the study period for each
study city), we note:

• For the key studies that we can most clearly understand how the study-reported means relate
to the annual standard metric:

0 Monitor-based 25th percentiles of health events correspond to mean PM2.5

concentrations (i.e., averaged over the study period for each study city): at or above
11.5 |j,g/m3

0 Monitor-based 10th percentiles of health events correspond to mean PM2.5
concentrations: at or above 9.8 |j,g/m3

o PM2.5 concentrations corresponding to 25th percentiles of estimated exposures that use
hybrid modeling approaches and used population-weighted approaches to estimate
PM2.5 exposures range from 6.7 to 9.1 |j,g/m3

o PM2.5 concentrations corresponding to 10th percentiles of estimated exposures that use
hybrid modeling approaches and used population-weighted approaches to estimate
PM2.5 exposures range from 4.7 to 7.3 |j,g/m3

Additionally, for the other set of key U.S. studies that use hybrid modeling approaches,
but don't include population weighting in calculating the study-reported means and other air
quality distributional statistics:

o PM2.5 concentrations corresponding to 25th percentiles of estimated exposures for these
studies range from 4.6 to 9.2 |j,g/m3

In thinking about these values relative to an area annual design value, we emphasize that
the 25th and 10th percentiles provide information about the lower quartiles of the air quality
distributions, while the study-reported mean provides information about the average or typical
exposures, and the corresponding area annual design value provides the highest average annual
PM2.5 concentration being measured. In this way, all of these metrics (i.e., lower percentiles,
study mean, annual design value) have a relationship relative to the other, and each of these
metrics can be used to inform our consideration of the level of the current annual standard. We
also note that the air quality analyses in Chapter 2 focuses on mean PM2.5 concentrations and
does not include other PM2.5 concentrations in the lower portion of the air quality distribution.
Therefore, given the lack of additional information to inform our understanding of the
relationship between percentiles of the air quality distribution other than the mean and the annual
design value, any direct comparison of study-reported PM2.5 concentrations corresponding to
lower percentiles (e.g., 25th and/or 10th) and annual design values is more uncertain than such
comparisons with the mean.

3-183

-------
• Are there other types of epidemiologic studies or approaches that can additionally

inform our consideration of the adequacy of the current PM2.5 standards?

There are a number of U.S. and Canadian studies that examine health effect associations
in analyses with the highest exposures excluded (i.e., studies that restrict analyses below certain
PM2.5 concentrations), which report positive and statistically significant associations in analyses
restricted to annual average PM2.5 exposures at or below 12 |j,g/m3 and/or to daily exposures
below 35 |j,g/m3 (Table 3-10). These restricted analyses provide additional support for effects at
lower concentrations, exhibiting associations for mean concentrations presumably below the
mean concentrations for the main analyses. While mean PM2.5 concentrations for these restricted
analyses may not be reported in most studies, we can presume that the mean PM2.5
concentrations in the restricted analyses are less than the study-reported mean PM2.5
concentrations in the main analyses, which range from 8.1 |ig/m3 to 11.6 |ig/m3 in the U.S.-based
studies, and was 7.8 |ig/m3 for the one study in Canada. It is important to note that, even if the
studies had reported the mean PM2.5 concentrations for the restricted analysis, we would not
necessarily be able to use these means in a similar decision framework as was used in past
reviews (section 3.3.3.2.1), given uncertainties associated with identifying the relationship
between a calculated mean concentration that excludes specific daily or annual average
concentrations above a certain threshold and the design value used to determine compliance with
a standard (either the annual or 24-hour standard). Moreover, there is uncertainty in how studies
exclude concentrations (e.g., at what spatial resolution are concentrations being excluded), which
would make any comparisons of mean concentrations in restricted analyses difficult to compare
to design values.

Studies that restrict 24-hour average PM2.5 concentrations to values of less than 35 |ig/m3
often do not report the mean PM2.5 concentration for the restricted analysis, as noted above,
although the mean of the restricted analysis is presumably less than the mean PM2.5
concentration in the main analysis. However, in some studies, the majority of PM2.5
concentrations from the main study are already less than the restricted concentration (e.g., in Di
et al., 2017a, where of all case and control days, 93.6% had PM2.5 concentrations below 25
|ig/m3), which contributes to the uncertainty in how much lower a mean concentration in a
restricted study is compared to the mean in the main analysis. As a result, there are limitations in
how this information can be used in evaluating the adequacy of the current or potential
alternative levels of the 24-hour standard. Additionally, it is difficult to use the means, when
reported, from studies of restricted analyses to evaluate the level of protection afforded by the
current or potential alternative levels of the primary 24-hour PM2.5 standard because the
relationship between the study-reported mean concentration and the 98th percentile form of the

3-184

-------
24-hour standard is not well understood, in particular for a short-term standard designed to limit
exposures to peak PM2.5 concentrations.

Finally, accountability studies evaluate whether changes in air quality are associated with
improvements in public health and a number of recent studies are evaluated in the ISA
Supplement (summarized in Table 3-12 above). These studies exhibit positive and significant
associations, including some studies that report starting PM2.5 concentrations below 12.0 |ig/m3,
indicating that public health improvements may occur following PM2.5 reductions in areas that
already meet the current annual PM2.5 standard. For example, studies by Corrigan et al. (2018)
and Sanders et al. (2020b) both found improvements in mortality rates due to improvements in
air quality in both attainment and nonattainment areas following implementation of the 1997
primary annual PM2.5 NAAQS.83 Other recent studies additionally report that declines in ambient
PM2.5 concentrations over a period of years have been associated with decreases in mortality
rates and increases in life expectancy, improvements in respiratory development, and decreased
incidence of respiratory disease in children, further supporting the robustness of PM2.5 health
effect associations reported in the epidemiologic evidence.

3,6,1,4 Uncertainties in the Health Effects Evidence

A number of key uncertainties and limitations were identified in the previous review with respect
to health effects evidence, as described in the 2020 PA. This section considers the currently
available information, including that newly available in this reconsideration, with regard to such
areas of uncertainty.

• To what extent have previously identified uncertainties in the health effects evidence
been reduced and/or have new uncertainties emerged?

We continue to recognize uncertainties that persist from previous reviews. First, we note
uncertainties related to the susceptibility of different population groups for which evidence is not
as clear (e.g., based on differences in underlying factors such as obesity, smoking status and
residential location). For human exposures studies, there are uncertainties related to mixed
results seen at concentrations near ambient PM2.5 levels. It is also unclear how the results alone
and the importance of the effects observed in these studies, particularly in studies conducted at
near-ambient PM2.5 concentrations, should be interpreted with respect to adversity to public
health. With respect to animal toxicology studies, while these studies also help establish
biological plausibility, uncertainty exists in extrapolating the effects seen in animal toxicology
studies, and the PM2.5 concentrations that cause those effects to human populations.

83 We note that the studies by Corrigan et al. (2018) and Sanders et al. (2020b) report monitor-based average PM2 5
concentrations, and that these studies do not report design values.

3-185

-------
Uncertainties associated with the epidemiologic evidence (e.g., the potential for
copollutant confounding and exposure measurement error) remain, though new studies assessed
in the ISA Supplement employ statistical methods like alternative methods for confounder
control to more extensively account for confounders and are more robust to model
misspecification. In so doing, however, we note the strength in the epidemiologic evidence in its
support for determination of a causal relationship for mortality and cardiovascular effects as
summarized in section 3.3.1 above.

With regard to controlling for potential confounders in particular, key epidemiologic
studies use a wide array of approaches. Time-series studies control for potential confounders that
vary over short time intervals (e.g., including temperature, humidity, dew point temperature, and
day of the week), while cohort studies control for community- and/or individual-level
confounders that vary spatially (e.g., including income, race, age, socioeconomic status,
smoking, body mass index, and annual weather variables such as temperature and humidity)
(Appendix B, Table B-4). Sensitivity analyses indicate that adding covariates to control for
potential confounders can either increase or decrease the magnitude of PM2.5 effect estimates,
depending on the covariate, and that none of the covariates examined can fully explain the
association with mortality (e.g., Di et al., 2017b, Figure S2 in Supplementary Materials). Thus,
while no individual study adjusts for all potential confounders, a broad range of approaches have
been adopted across studies to examine confounding, supporting the robustness of reported
associations. Available studies additionally indicate that PM2.5 health effect associations are
robust across various approaches to estimating PM2.5 exposures and across various exposure
windows. This includes recent studies that estimate exposures using ground-based monitors
alone and studies that estimate exposures using data from multiple sources (e.g., satellites, land
use information, modeling), in addition to monitors. While none of these approaches eliminates
the potential for exposure error in epidemiologic studies, such error does not call into question
the fundamental findings of the broad body of PM2.5 epidemiologic evidence.

Additionally, studies that examine the shapes of concentration-response functions over
the full distribution of ambient PM2.5 concentrations have not identified a threshold
concentration, below which associations no longer exist (U.S. EPA, 2019, section 1.5.3, U.S.
EPA, 2022, sections 2.2.3.1 and 2.2.3.2). While such analyses are complicated by the relatively
sparse data available at the lower end of the air quality distribution (U.S. EPA, 2019, section
1.5.3), the evidence remains consistent in supporting a no-threshold relationship, and in
supporting a linear relationship for PM2.5 concentrations > 8 [j,g/m3. However, uncertainties
remain about the shape of the C-R curve at PM2.5 concentrations < 8 (j,g/m3, with some recent
studies providing evidence for either a sublinear, linear, or supralinear relationship at these lower
concentrations3.

3-186

-------
While studies using hybrid modeling methods have demonstrated reduced exposure
measurement error and uncertainty in the health effect estimates, these methodologies have
inherent limitations and uncertainties, as described in more detail in section 2.3.3.1.5 and above
in 3.3.4, and the performance of the modeling approaches depends on the availability of
monitoring data which varies by location. Factors likely contributing to poorer model
performance often coincide with relatively low ambient PM2.5 concentrations, in areas where
predicted exposures are at a greater distance to monitors, and under conditions where the
reliability and availability of key datasets (e.g., air quality modeling) are limited. Thus,
uncertainty in hybrid model predictions becomes an increasingly important consideration as
lower predicted concentrations are considered.

In addition, limitations and or uncertainties exist in the analysis (section 2.3.3.2.4)
evaluating the comparison of estimated PM2.5 concentrations using hybrid modeling surfaces and
their relationship to design values that should be considered. While design values in general are
higher than estimated PM2.5 concentrations using these two hybrid modeling approaches (DI2019
and HA2020), it is important to recognize that these are just two hybrid modeling approaches to
estimating PM2.5 concentrations and other models/approaches/spatial scales may result in
somewhat different PM2.5 concentrations and relationships with design values. The analysis
evaluating the relationship between surfaces by DI2019 and HA2020 and design values (section
2.3.3.2.4) estimates PM2.5 concentrations by CBSAs, but not every health study uses PM2.5
estimates at this spatial scale, and spatial scales for exposure estimates can vary by study. As an
example of this variation, in Di et al. (2017b), an annual average PM2.5 concentration was
assigned to a person at-risk of death according to the ZIP code of the person's residence. The
analysis completed in section 2.3.3.2.4 was a nationwide analysis and ratios between design
values and mean concentrations are based on national estimates. However, not all health studies
are national studies (i.e., some studies are completed in different regions of the country, like the
southeast or northeast) and ratios in different parts of the country could be higher or lower,
depending on factors like population, as well as the proportion of rural versus urban areas. This
analysis used specific air quality years (2000-2016) and the use of other air quality years could
result in higher or lower ratios.

Regardless of whether an epidemiologic study uses monitoring data or a hybrid modeling
approach when estimating PM2.5 exposures, one important challenge that persists is associated
with the interpretation of the study-reported mean PM2.5 concentrations and how they compare to
design values. This is particularly true given the variability that exists across the various
approaches to estimate exposure and to calculate the study-reported mean. These types of
challenges are also present in using information from Canadian studies to directly and

3-187

-------
quantitatively inform questions on the level of the annual standard given the difficulty of
interpreting what the Canadian study means represent relative to U.S. design values.

3.6.2 Risk-based Considerations

Our consideration of the scientific evidence available in this reconsideration, as at the
time of the 2020 review, is informed by results from a quantitative analysis of risk. The
overarching consideration in this section is whether the current risk information alters our overall
conclusions regarding health risk associated with exposure to PM2.5 in ambient air. As in our
consideration of the evidence in section 3.6.1 above, we have focused the discussion regarding
the risk information around key questions related to air quality conditions simulated to just meet
existing and alternative primary PM2.5 standards.

Prior to addressing the key risk questions, we provide a summary of important aspects of
the assessment, including the study areas, air quality scenarios, and risk metrics (section 3.6.2.1).
We then consider aspects of the questions beginning with the magnitude of risk estimated by
both the overall assessment and for certain at-risk populations, followed by the key uncertainties
associated with the quantitative analyses with regard to drawing conclusions as to the adequacy
of protection afforded by the current primary PM2.5 standards (section 3.6.2.2 and 3.6.2.3). We
also consider uncertainties associated with the risk assessment (section 3.6.2.4). Lastly, we
consider the risk estimates from the quantitative assessments with regard to the extent to which
such estimates may be judged to be important from a public health perspective (section 3.6.2.5).

3.6.2.1 Risk Assessment Analyses

In the risk assessment conducted for this reconsideration, described in detail in section
3.4 above and Appendix C, we have estimated PM2.5 health risks associated with air quality
conditions that just meet the current primary PM2.5 standards and potential alternative standard
levels. These analyses inform our understanding of the health risks for all-cause or nonaccidental
mortality associated with long- and short-term PM2.5 exposures. These analyses estimate
exposure and risk for populations in 47 urban study areas, as well as subsets of those study areas
depending on which of the primary PM2.5 standards is controlling in a given study area.

The 47 urban study areas were identified as they required relatively small adjustments
(<20%) to just meet the current primary PM2.5 standards and present a variety of circumstances
with regard to risk associated with long- and short-term exposures to PM2.5 in ambient air. This
set of study areas and the associated populations are intended to be informative to the EPA's
consideration of potential risks that may be associated with the air quality conditions that meet
the current and potential alternative primary PM2.5 standards. The 47 study areas include nearly
60 million people ages 30 years or older and illustrate the differences likely to occur across
various locations with such air quality as a result of area-specific differences in emissions,

3-188

-------
meteorological, and population characteristics. While the same conceptual air quality scenarios
are simulated in all study areas (i.e., conditions that just meet the existing or alternate standards),
source, meteorological and population characteristics in the study areas contribute to variability
in the estimated magnitude of risk across study areas.

As an initial matter, we note that, consistent with the overall approach for this
reconsideration, the risk assessment has a target scope that focuses on all-cause or nonaccidental
mortality associated with long- and short-term PM2.5 exposures (section 3.4.1.2). As noted in
section 3.6.1 above, the evidence assessed in the 2019 ISA and ISA Supplement support a causal
relationship between long- and short-term PM2.5 exposures and mortality. Concentration-
response functions used in the risk assessment are from large, multicity U.S. epidemiologic
studies that evaluate the relationship between PM2.5 exposures and mortality and were identified
using criteria that take into account factors such as study design, geographic coverage,
demographic populations, and health endpoints (U.S. EPA, 2021b, section 2.1).

In the risk assessment, air quality modeling was used to develop a PM2.5 concentration
field for 2015 (described in more detail in section 3.4.1.4 and Appendix C). The 2015 PM2.5
concentration field was adjusted to simulate just meeting the existing annual and 24-hour
standards of 12.0 |ig/m3 and 35 |ig/m3 and to just meeting potential alternative annual and 24-
hour standards of 10.0 |ig/m3 and 30 |ig/m3. The adjustments made to the PM2.5 concentration
field are based on assumptions. Changes in PM2.5, in reality, require specific information
regarding emissions changes, with concentration gradients of PM2.5 varying accordingly across
an area. The risk assessment used two adjustment approaches to serve as bounding scenarios for
the various ways an alternative standard may be met: (1) preferentially adjusting direct/primary
PM emissions, for which changes in PM2.5 tend to be more localized near the direct emissions
sources of PM (Pri-PM), and (2) preferentially adjusting SO2 and NOx precursor emissions to
simulate changes in secondarily formed PM2.5, for which reductions in PM2.5 tend to be more
evenly spread across a study area (Sec-PM). In addition to the air quality modeling approach,
linear interpolation and extrapolation were used to simulate just meeting alternative annual
standards with levels of 11.0 (interpolated between 12.0 and 10.0 |j,g/m3), 9.0 ng/m3, and 8.0
Hg/m3 (both extrapolated from 12.0 and 10.0 |ag/m3) in the subset of study areas controlled by
the annual standard.

Evidence strongly supports that different racial and ethnic groups, such as Black and
Hispanic populations, have higher PM2.5 exposures than White and non-Hispanic populations,
respectively, thus contributing to increased risk of PM-related effects. In addition to the risk
assessment described above, quantitative analyses for this reconsideration also assess long-term
PM2.5-attributable exposure and mortality risk, stratified by racial/ethnic demographics.
Consistent with the overall risk assessment approach, the specific epidemiologic studies and

3-189

-------
concentration-response functions used in the at-risk analyses were selected to take into account
factors such as study design, geographic coverage, demographic populations, and health
endpoints. Of the available studies, Di et al., 2017b was identified as best characterizing
populations potentially at increased risk of long-term exposure and all-cause mortality and
provides race- and ethnicity-stratified concentration-response functions for ages 65 and over
(section 3.4.1.6 and Appendix C). Risk is quantitatively assessed within racial and ethnic
minority populations of older adults in the full set of 47 areas and the subset of 30 areas
controlled by the annual PM2.5 standard under Pri-PM air quality simulations. This analysis,
when considered alongside estimates of risk across all populations in the 47 study areas, can help
to inform conclusions on the annual primary PM2.5 standards that would be requisite to protect
the public health of minority populations potentially at increased risk of long-term PM2.5-related
mortality effects.

3.6,2,2 Estimating Risk under the Current and Alternative Primary PM2.5 Standards

In this section, we summarize the risk estimates associated with air quality scenarios just
meeting the current primary PM2.5 standards and potential alterative standard levels.

• What are the estimated PM2.5-associated health risks for air quality just meeting the
current primary PM2.5 standards?

In considering the risk results, we focus first on estimates for the full set of 47 urban
study areas. The risk assessment estimates that the current primary PM2.5 standards could allow a
substantial number of deaths in the U.S., with the large majority of those deaths associated with
long-term PM2.5 exposures. For example, when air quality in the 47 study areas is adjusted to just
meet the current standards, the risk assessment estimates about 41,000 to 45,000 deaths from all-
cause mortality in a single year (i.e., for long-term exposures; confidence intervals range from
about 30,000 to 59,000) (section 3.4.2.1). For the 30 study areas84 where just meeting the current
standards is controlled by the annual standard,85 long-term PM2.5 exposures are estimated to be
associated with as many as 39,000 (confidence intervals range from about 26,000 to 51,000)
deaths from all-cause mortality in a single year (section 3.4.2.2). For the 11 study areas86 where

84 These 30 areas controlled by the annual standard under all scenarios evaluated include a population of
approximately 48 million adults aged 30-99, or about 75% of the population included in the full set of 47 areas.

85 For these areas, the annual standard is the "controlling standard" because when air quality is adjusted to simulate
just meeting the current or potential alternative annual standards, that air quality also would meet the 24-hour
standard being evaluated.

86 These 11 areas controlled by the 24-hour standard under all scenarios evaluated include a population of
approximately 10 million adults aged 30-99, or about 17% of the population included in the full set of 47 areas.

3-190

-------
just meeting the current standards is controlled by the daily standard,87 long-term PM2.5
exposures are estimated to be associated with as many as 2,600 (confidence intervals ranging
from 1,700 to 3,400) deaths in a single year (section 3.4.2.3). The risk assessment estimates far
fewer deaths in a single year for short-term PM2.5 exposures as compared to long-term PM2.5
exposures, across all of the study area subsets.

While the absolute numbers of estimated deaths vary across exposure durations,
populations, and concentration-response functions, the general magnitude of risk estimates
supports the potential for significant public health impacts in locations meeting the current
primary PM2.5 standards. This is particularly the case given that the large majority of PM2.5-
associated deaths for air quality just meeting the current standards are estimated at annual
average PM2.5 concentrations from about 10 to 12 |ig/m3. These annual average PM2.5
concentrations fall within the range of long-term average concentrations over which key
epidemiologic studies provide strong support for reported positive and statistically significant
health effect associations.

• To what extent are risks estimated to decline when air quality is adjusted to just
meet potential alternative standards with lower levels?

In the 47 urban study areas, when air quality is simulated to just meet alternative
standards, there are substantially larger risk reductions associated with lowering the annual
standard then with lowering the 24-hour standard. Risks are estimated to decrease by 13-17%
when air quality is adjusted to just meet an alternative annual standard with a level of 10.0 |j,g/m3
or by 1-2% when adjusted to just meet an alternative 24-hour standard with a level of 30 |j,g/m3
(section 3.4.2.1). The percentage decrease when just meeting an alternative annual standard with
a level of 10.0 |j,g/m3 corresponds to approximately 7,400 fewer deaths per year (confidence
intervals ranging from about 4,100 to 9,800) attributable to long-term PM2.5 exposures.

In the 30 study areas where just meeting the current and alternative standards is
controlled by the annual standard, air quality adjusted to meet alternative annual standards with
lower levels is associated with reductions in estimated all-cause mortality risk. These reductions
in risk for alternative annual levels are as follows: 7-9% reduction for an alternative annual level
of 11.0 |ig/m3, 15-19%) reduction for a level of 10.0 |ig/m3, 22-28%) reduction for a level of 9.0
|ig/m3, and 30-37%) reduction for a level of 8.0 |ig/m3 (section 3.4.2.2). For each of these
standards, most of the risk remaining is estimated at annual average PM2.5 concentrations that
fall somewhat below the alternative standard levels.

87 For these areas, the 24-hour standard is the controlling standard because when air quality is adjusted to simulate
just meeting the current or potential alternative 24-hour standards, that air quality also would meet the annual
standard being evaluated. Some areas classified as being controlled by the 24-hour standard also violate the
annual standard.

3-191

-------
3,6,2,3 At-Risk Analyses

As noted above, in addition to the risk assessment described in sections 3.4.1.1-3.4.1.5
and 3.4.2.1-3.4.2.3, risk was quantitatively assessed within racial and ethnic minority populations
of older adults in the full set of 47 areas and the subset of 30 areas controlled by the annual PM2.5
standard under all air quality simulations evaluated (sections 3.4.1.6 and 3.4.2.4).

• What is the magnitude of population risk in at-risk populations in areas simulated to
just meet the current primary PM2.5 standards? To what extent are risks estimated
to decline within each demographic group when air quality is adjusted to just meet
potential alternative annual standards with lower levels?

The at-risk analysis first compares the average estimated PM2.5 exposure concentrations
for each demographic population when just meeting the current and alternative annual PM2.5
standards. Across all simulated air quality for both the full set of 47 and the subset of 30 study
areas, Blacks experience the highest average PM2.5 concentrations of the demographic groups
analyzed. Native Americans experienced the lowest average PM2.5 concentrations, particularly in
the full set of 47 study areas. White, Hispanic, and Asian populations were exposed to similar
average PM2.5 concentrations. Additionally, as the levels of potential alternative annual PM2.5
standards decrease, there is comparatively less disproportionate exposure between demographic
populations (section 3.4.2.4).

Risk estimates can provide additional information beyond the exposure information to
inform our understanding of potentially disproportionate impacts, in this instance by including
demographic-specific information on baseline incidence and the relationship between exposure
and health effect. Across all air quality scenarios and demographic groups evaluated, Black
populations are associated with the largest PIVh.s-attributable mortality risk rate per 100,000
people, while White populations are associated with the smallest PM2.5-attributative mortality
risk rate (section 3.4.2.4, Figure 3-20). Generally, as the levels of potential alternative annual
PM2.5 standards decrease in the 30 areas controlled by the annual standard, the average reduction
in PM2.5 concentration and mortality risk rates increase across all demographic populations
(section 3.4.2.4, Figure 3-21).

In comparing the reductions in average national PM2.5 concentrations and risk rates
within each demographic population, we note that the average percent PM2.5 concentrations and
risk reductions are slightly greater in the Black population than in the White population for each
alternative standard evaluated (11.0 |ig/m3, 10.0 |ig/m3, 9.0 |ig/m3, and 8.0 |ig/m3), when shifting
from the current annual PM2.5 standard (12.0 |ig/m3) in the full set of 47 areas and the subset of
30 areas controlled by the annual standard. We further note that the difference in average percent
risk reductions increases slightly more in Blacks than in Whites as the level of the potential
alternative annual standard decreases (section 3.4.2.4, Table 3-19 and Table 3-20).

3-192

-------
3,6,2,4 Uncertainties

In this section, we consider uncertainties associated with the quantitative estimates of risk
in the overall risk assessment and from risk rates and exposure estimates in the at-risk analysis
(sections 3.4.2.5, 3.4.1.7, and 3.4.1.8). Variability and uncertainty associated with the risk
estimates are assessed using several quantitative and qualitative approaches, as described in more
detail in section C.3 of Appendix C. Generally, the quantitative uncertainty characterization
approaches include the following: (1) evaluating multiple concentration-response functions for
the same health endpoint; (2) evaluating multiple methods for simulating air quality scenarios;
and (3) characterizing the 95% confidence intervals associated with risk estimates. The
qualitative uncertainty characterization approach is based on WHO (2008) guidance and on
guidance documents developed by the EPA (U.S. EPA, 2001, U.S. EPA, 2004). This qualitative
approach includes an assessment of both the magnitude and direction of impact of those
uncertainties on risk estimates, including three levels of classification for the magnitude: low,
medium, and high.88

• What are the key uncertainties associated with the risk estimates and at-risk
analysis, including those of particular significance with regard to drawing
conclusions as to the adequacy of the protection afforded by the current primary
PM2.5 standards?

Based on the uncertainty characterization and associated analyses in the risk assessment
and consideration of associated policy implications, we recognize several areas of uncertainty as
particularly important in our consideration of the risk estimates, as was also the case in previous
reviews, and in the risk rates and exposure and risk reductions in the at-risk analysis.

With regard to the concentration-response relationships, we recognize that the degree to
which different concentration-response functions result in different risk estimates could reflect
differences in study design and/or populations evaluated, as well as other factors. We also note
uncertainty in the risk assessment associated with the interpretation of the shapes of
concentration-response relationships, particularly at PM2.5 concentrations near the lower end of
the air quality distribution. This interpretation is complicated by relatively low data density in the
lower concentration range, the possible influence of exposure measurement error, and variability
among individuals with respect to air pollution health effects. These sources of variability and
uncertainty tend to smooth and "linearize" population-level concentration-response functions,

88 The classification of the magnitude of impact for sources of uncertainty includes three levels: (a) low (unlikely to
produce a sufficient impact on risk estimates to affect their interpretation), (b) medium (potential to have a
sufficient impact to affect interpretation), and (c) high (likely to have an impact sufficient to affect interpretation)
For several of the sources, a classification was provided between these levels (e.g., low-medium, medium-high).
More information is available in Appendix C, section C.3.

3-193

-------
and thus could obscure the existence of a threshold or nonlinear relationship (U.S. EPA, 2015a,
section 6.c). As described in section 3.3.1, the 2019 ISA concludes and the ISA Supplement
provides further support that the majority of evidence of long-term PM2.5 exposure and mortality
supports a linear, no-threshold concentration-response relationship, though there is initial
evidence indicating that the slope of the concentration-response curve may be steeper at lower
concentrations for cardiovascular mortality (U.S. EPA, 2019, section 1.5.3.2; U.S. EPA, 2022,
section 3.2.2.2). The 2019 ISA and ISA Supplement note that there is less certainty in the shape
of the concentration-response curve at mean annual PM2.5 concentrations generally below 8
[j,g/m3 because data density is reduced below this concentration (U.S. EPA, 2019, section 11.2.4;
U.S. EPA, 2022, section 3.2.2.2.7). As described in more detail in section 3.4.2.5 above and
Appendix C, a portion of risk modeling in the risk assessment does include locations with annual
ambient PM2.5 concentrations adjusted to below 8 ug/m3, so there is the potential for significant
uncertainty being introduced into the risk assessment (particularly for that portion of risk
modeled at or below 8 ug/m3). With regard to short-term PM2.5 exposure and mortality, the 2019
ISA concludes and the ISA Supplement provides additional support that, while difficulties
remain in assessing the shape of the PM2.5-mortality concentration-response relationship and
studies have not conducted systematic evaluations of alternatives to linearity, recent studies
continue to provide evidence of a no-threshold linear relationship, with less confidence at
concentrations lower than 5 [j,g/m3 (U.S. EPA, 2022, section 3.2.1.2.6). However, we note that in
most instances in the risk assessment for this reconsideration, the concentration-response
function used had only a small impact on the risk estimates.

With regard to the method for simulating air quality scenarios, the approach used to
adjust air quality (i.e., adjusting primary PM emissions or secondary PM emission precursors)
had some impact on the overall risk estimates. We also note that there may be uncertainty
associated with the methods used to simulate air quality scenarios just meeting the current and
potential alternative primary PM2.5 standards. The model-based methods for simulating air
quality scenarios that just meet the current and alternative standards could contribute to
uncertainties associated with the PM2.5 concentration estimates used in the risk assessment and
at-risk analyses. While state-of-the-science modeling methods were used to fill in the spatial and
temporal gaps in monitoring data, model-related biases and errors can introduce uncertainties.
Additionally, the modeling scenarios are based on "across-the-board" changes in primary PM2.5
or NOx and SO2 emissions from all anthropogenic sources throughout the U.S. by fixed
percentages. While this approach tends to target the key sources in each area, emission changes
are not tailored to specific periods or sources. Furthermore, while the two adjustment approaches
that were applied span a wide range of emissions conditions, they represent a subset of the
possible emissions cases that could be used to adjust PM2.5 concentrations. In addition, when

3-194

-------
simulating air quality scenarios that just meet potential alternative annual PM2.5 standards using
linear extrapolation/interpolation, we recognize that this approach does not fully capture the
potential non-linearities associated with real-world changes in air quality. However, it is
important to note that the adjustment approach had a larger impact on the distribution of risk
reductions, particularly for potential alternative annual standard levels of 9.0 and 8.0 |j,g/m3.

It is important to note that the air quality adjustment approaches applied in the risk
assessment differ from the development and modeling of emission control strategies that would
occur in implementing a standard. In implementing a standard, an appropriately defined
nonattainment area would reduce emissions of primary PM and/or PM precursors selected
through analysis of site-specific conditions to meet a standard that is exceeded. In the risk
assessment, gridded concentration fields over CBSAs were adjusted to higher or lower
concentrations to correspond to just meet standards based on emission changes applied
throughout the U.S. Two emission adjustment cases (primary PM and NOx and SO2) were used
to provide concentration fields that span a wide range of realistic spatial patterns, but the air
quality modeling for the risk assessment is not designed to reflect emission changes that might
occur in implementing a standard. The Regulatory Impact Analysis (RIA) associated with
NAAQS revisions provides illustrative estimates of emission changes needed to meet potential
alternative standards and more closely reflects implementation considerations (U.S. EPA, 2013,
U.S. EPA, 2015b).

We further note that there is considerable variation in the range of confidence intervals
associated with the point estimates generated in the risk assessment, with some concentration-
response functions displaying greater variability than others. A number of factors could
potentially influence the varying degrees of statistical precision in effect estimates, including
sample size, exposure measurement error, degree of control for confounders/effect modifiers,
and variability in PM2.5 concentrations evaluated in the original epidemiologic study.

There may also be uncertainty associated with the potential confounding of the PM2.5-
mortality effect and the omission of potential confounders from analyses could either increase or
decrease the magnitude of PM2.5 effect estimates. Not accounting for confounders can introduce
uncertainty into the effect estimates, and thereby introduce uncertainty into the risk estimates that
are generated using those effect estimates. While various approaches to control for potential
confounders have been adopted across the epidemiologic studies assessed in the 2019 ISA and
ISA Supplement, and those used in the risk assessment, no individual study adjusts for all
potential confounders.

In addition to the uncertainty associated with the risk assessment estimates, additional
uncertainties are associated with the risk rates, exposure estimate, and risk reductions in the at-
risk analysis. As an initial matter, we note that this analysis is based on race- and ethnicity-

3-195

-------
stratified concentration-response functions only for ages 65 and over (Di et al., 2017b). The use
of one study in such an analysis introduces uncertainties and limitations in the broad applicability
of such results in the context of the national U.S. population across demographic groups and age
ranges. In addition, each minority demographic group analyzed in the study comprised a smaller
percentage of the full study population, which reduces analytical power. Finally, the risk and
exposure assessment focuses on urban areas. This means that demographic groups that
preferentially reside in rural areas, such as Native Americans, are underrepresented in this
analysis. Additionally, average exposure concentrations estimated for demographic groups with
substantial rural populations, such as Whites, may be overestimated in this urban analysis.

In summary, here we recognize several particularly important uncertainties that affect the
quantitative estimates of risk rates and exposure in the at-risk analysis and their interpretation in
the context of considering the current primary PM2.5 standards. These include uncertainties
related to the modeling and adjustment methods for simulating air quality scenarios; the potential
influence of confounders on the relationship between PM2.5 exposure and mortality; the
interpretation of the shapes of concentration-response functions, particularly at lower
concentrations; and limited availability of studies to inform the at-risk analysis.

3,6.2,5 Potential Public Health Implications

In considering the public health implications of the quantitative risk assessment and at-
risk analysis that may inform the Administrator's judgments in this area, this section discusses
the information pertaining to the following questions.

• To what extent are the estimates of risk important from a public health perspective?

What does the information available in this reconsideration indicate with regard to

the size of the at-risk populations?

Several factors are important to consideration of public health implications. These
include the magnitude or severity of the effects associated with the estimated exposures, as well
as their adversity at the individual and population scales. Other important considerations include
the size of the population estimated to experience such effects or to experience exposures
associated with such effects. Thus, the discussion here reflects consideration of the risk-based
evidence in the context of potential health implications in previous NAAQS decisions.

With regard to PM2.5 concentrations in ambient air, the public health implications and
potential public health impacts of interest in this reconsideration relate to those effects where a
causal relationship with PM2.5 exposure was concluded. These are mortality and cardiovascular
effects related to both long- and short-term exposures, as summarized in section 3.3.1 above.
Such effects, including more serious effects such as mortality, can be considered severe from a
public health perspective.

3-196

-------
In considering public health implications, it is important to consider impacts on
population groups of differing susceptibility. The size of the at-risk populations (children, older
adults, those with pre-existing cardiovascular or respiratory diseases) in the U.S. is substantial.
As summarized in section 3.3.2, more than 22% of the population are children (<18 years old;
approximately 73 million people) and about 16% are older adults (65+ years old; approximately
54 million people). For adults in the U.S. 18 years old and older, cardiovascular diseases are
most prevalent in adult populations over the age of 65, with 29% of this age group reporting
some type of heart disease (Table 3-3 above). Similarly, adults over the age of 65 also have a
greater prevalence of respiratory diseases, particularly COPD including chronic bronchitis and
emphysema, while the asthma prevalence is generally consistent across all adult age groups for
those 18 years or older (Table 3-3). It is important to note that for older adults, the increased risk
in this lifestage can likely be attributed to the gradual decline in physiological processes that
occurs with aging, and some overlap exists between populations considered to be at-risk because
of pre-existing disease and lifestage (U.S. EPA, 2019, p. 12-25).

Another factor that may contribute to differences PM2.5 exposures and PIVh.s-related
health risk is race/ethnicity. As described above in section 3.3.2 and in the 2019 ISA and ISA
Supplement, there is strong evidence demonstrating that Black and Hispanic populations, in
particular, have higher PM2.5 exposures and health risk disparities compared to non-Hispanic
White populations. In the U.S., more than 12% of the U.S. population (more than 40.5 million
people) are Blacks and more than 18% are Hispanics (more than 60 million people), while 60%
of the population (nearly 197 million people) are non-Hispanic Whites (Table 3-2). Black and
Hispanic individuals of all ages make up a substantial portion of the population.

In considering the public health implications of the risk estimates across the study areas,
we note the purpose for the study areas is to illustrate circumstances that may occur in areas that
just meet the current or potential alternative standards, and not to estimate risk associated with
conditions occuring in those specific locations currently. We note that some areas across the U.S.
have air quality for PM2.5 that is near or above the existing standards. Thus, the air quality and
exposure circumstances assessed in the study areas in the risk assessment are of particular
importance in considering whether the currently available information calls into question the
adequacy of the public health protection afforded by the current standards.

The risk estimates for the study areas assessed in this reconsideration reflect differences
in exposure circumstances among those areas and illustrate the exposures and risks that might be
expected to occur in other areas with such circumstances under air quality conditions that just
meet the current standards or the alternative standards assessed. Thus, the exposure and risk
estimates indicate the magnitude of exposure and risk that might be expected in many areas of
the U.S. with PM2.5 concentrations at or near the current or alternative standards. Although the

3-197

-------
methodologies and data used to estimate risks in this reconsideration differ in several ways from
what was used in the 2020 review, the findings and considerations summarized here present a
pattern of exposure and risk that is generally similar to that considered in the 2020 review, and
indicate a level of protection generally consistent with that described in the 2020 PA.

In summary, the considerations raised here are important to conclusions regarding the
public health significance of the risk assessment results. Specifically, we note that available
evidence and information suggests that both long- and short-term PM2.5 exposures are associated
with adverse health effects, including more severe effects such as mortality. In addition, we note
that such effects impact large segments of the U.S. population, including those populations that
may have other factors that influence risk (i.e., lifestage, pre-existing cardiovascular and
respiratory diseases, race/ethnicity), as well as disparities in PM2.5 exposures and health risks
based on race and ethnicity. Therefore, we recognize that the air quality allowed by the current
primary PM2.5 standards could be judged to be associated with significant public health risk. We
recognize that such conclusions also depend in part on public health policy judgments that will
weigh in the Administrator's decision in this reconsideration with regard to the adequacy of
protection afforded by the current standards. Such judgments that are common to NAAQS
decisions include those related to public health implications of effects of differing severity. Such
judgments also include those concerning the public health significance of effects at exposures for
which evidence is limited or lacking, such as effects at lower concentrations than those
demonstrated in the key epidemiologic studies and in those population groups for which
population-specific information, such as concentration-response functions, are not available from
the epidemiologic literature.

3.6.3 Conclusions on the Adequacy of the Current Primary PM2.5 Standards

This section describes our conclusions for the Administrator's consideration in this
reconsideration of the primary PM2.5 standards. These conclusions are based on considerations
described in the sections above, and in the discussion below regarding the scientific evidence (as
summarized in the 2019 ISA (U.S. EPA, 2019) and the ISA Supplement (U.S. EPA, 2022)), the
quantitative assessments of PIVh.s-associated health risks, and analyses of PM2.5 air quality.

3,6.3.1 Current Standards

In taking into consideration the discussions responding to specific questions above in this
chapter, this section addresses the following overarching policy question.

3-198

-------
• Does the currently available scientific evidence and risk-based information support
or call into question the adequacy of the public health protection afforded by the
current annual and 24-hour PM2.5 standards?

In considering this question, we recognize that, as is the case with NAAQS reviews in
general, the extent to which the current primary PM2.5 standards are judged to be adequate will
depend on a variety of factors, including science policy and public health policy judgments to be
made by the Administrator on the strength and uncertainties of the scientific evidence. The
factors relevant to judging the adequacy of the standards also include the interpretation of, and
decisions as to the weight to place on, different aspects of the results of the risk assessment for
the study areas included and the associated uncertainties. Thus, we recognize that the
Administrator's conclusions regarding the adequacy of the current standards will depend in part
on judgments regarding aspects of the evidence and risk estimates, and judgments about the
degree of protection that is requisite to protect public health with an adequate margin of safety.

Our response to the overarching question above takes into consideration the discussions
that address the specific policy-relevant questions in prior sections of this document (sections
3.3, 3.4, 3.6.1, and 3.6.2) and builds on the approach from previous reviews (summarized in
section 3.1 above). We focus first on consideration of the evidence, including that assessed in the
2019 ISA and the ISA Supplement, and the extent to which it alters key conclusions supporting
the current standards. We then turn to consideration of the quantitative estimates of risk
developed in this reconsideration, including associated uncertainties and limitations, and the
extent to which they indicate differing conclusions regarding the magnitude of risk, as well as
level of protection from adverse effects, associated with the current standards. We additionally
consider the key aspects of the evidence and risk estimates emphasized in establishing the
current standards, and the associated public health policy judgments and judgments about the
uncertainties inherent in the scientific evidence and quantitative analyses that are integral to
decisions on the adequacy of the current primary PM2.5 standards.

We first note that our approach recognizes that the current annual standard (based on
arithmetic mean concentrations) and 24-hour standard (based on 98th percentile concentrations),
together, are intended to provide public health protection against the full distribution of short-
and long-term PM2.5 exposures. In general, the annual standard is most effective at controlling
exposures to "typical" daily PM2.5 concentrations that are experienced over the year, while the
24-hour standard, with its 98th percentile form, is most effective at limiting peak daily or 24-
hour PM2.5 concentrations. In considering the combined effects of these standards, we recognize
that changes in PM2.5 air quality designed to meet an annual standard would likely result not only
in lower short- and long-term PM2.5 concentrations near the middle of the air quality distribution,
but also in fewer and lower short-term peak PM2.5 concentrations. Additionally, changes

3-199

-------
designed to meet a lower 24-hour standard, with a 98th percentile form, would most effectively
result in fewer and lower peak 24-hour PM2.5 concentrations, but also have an effect on lowering
the annual average PM2.5 concentrations. Thus, our focus in evaluating the current primary
standards is on the protection provided by the combination of the annual and 24-hour standards
against the distribution of both short- and long-term PM2.5 exposures.

As an initial matter, we note the longstanding body of health evidence supporting
relationships between PM2.5 exposures (short- and long-term) and mortality or serious morbidity
effects. The evidence available in this reconsideration (i.e., assessed in U.S. EPA, 2019 and U.S.
EPA, 2022) and summarized above in section 3.3.1 and section 3.6.1) reaffirms, and in some
cases strengthens, the conclusions from the 2009 ISA regarding the health effects of PM2.5
exposures (U.S. EPA, 2009). As noted above, epidemiologic studies conducted in North
America, Europe, or Asia demonstrate generally positive, and often statistically significant,
PM2.5 health effect associations. Such studies report associations between estimated PM2.5
exposures and non-accidental, cardiovascular, or respiratory mortality; cardiovascular or
respiratory hospitalizations or emergency room visits; and other mortality/morbidity outcomes
(e.g., lung cancer mortality or incidence, asthma development). Recent experimental evidence, as
well as evidence from panel studies, strengthens support for potential biological pathways
through which PM2.5 exposures could lead to the serious effects reported in many population-
level epidemiologic studies, including support for pathways that could lead to cardiovascular,
respiratory, nervous system, and cancer-related effects.

Epidemiologic studies in the U.S. report health effect associations with mortality and/or
morbidity across multiple cities and in diverse populations, including in studies examining
populations and lifestages that may be at higher risk of experiencing a PIVh.s-related health effect
(e.g., older adults, children). Further, these studies use a variety of statistical designs, and employ
a variety of methods to examine exposure measurement error as well as to control for
confounding effects, including more recent studies that employ alternative methods for
confounder control. Results of these analyses support the robustness of the reported associations.
Additional findings from an expanded body of studies that employ alternative methods for
confounder control and accountability methods further inform the causal nature of the
relationship between long- or short-term term PM2.5 exposure and mortality (U.S. EPA, 2019,
sections 11.1.2.1, 11.2.2.4; U.S. EPA, 2022, sections 3.1.1.3, 3.1.2.3, 3.2.1.3, and 3.2.2.3). These
studies, summarized above in Table 3-11 and Table 3-12, examine both short- and long-term
PM2.5 exposure and cardiovascular effects and mortality, and using 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. Moreover, recent epidemiologic studies strengthen support for health effect

3-200

-------
associations at relatively low ambient PM2.5 concentrations. Studies that examine the shapes of
concentration-response relationships over the full distribution of ambient PM2.5 concentrations
have not identified a threshold concentration, below which associations no longer exist (U.S.
EPA, 2019, section 1.5.3, U.S. EPA, 2022, sections 2.1.1.5.1 and 2.1.1.5.2). While such analyses
are complicated by the relatively sparse data available at the lower end of the air quality
distribution (U.S. EPA, 2019, section 1.5.3), several studies report positive and statistically
significant associations in additional analyses restricted to annual average PM2.5 exposures below
12 |j,g/m3 or to daily exposures below 35 |j,g/m3 as exhibited in Table 3-10.

These and other recent studies provide support for health effect associations at lower
ambient PM2.5 concentrations than in previous reviews. In this reconsideration, a large number of
key studies report positive and statistically significant associations for air quality distributions
with lower overall mean PM2.5 concentrations (i.e., Figure 3-8, Figure 3-9, Figure 3-10, Figure 3-
11). Consistent with the 2012 review, it is important to consider the manner in which PM2.5 mean
concentrations are estimated (e.g., monitored concentrations versus modeled concentrations) and
the method by which means are calculated and reported as the overall mean PM2.5 concentration
(e.g., averaging across all grid cells in an urban area versus population weighting). Additional
analyses, new in this PA though similar to those in the 2012 review, suggest that the area annual
design value is generally greater than the study mean by 10-20% (monitor-based studies), 15-
18% (hybrid modeling with population weighting) or 40-50% (hybrid modeling without
population weighting). We note this information relative to the overall mean PM2.5
concentrations in key U.S. epidemiologic studies which are: 9.9 [j,g/m3 to 16.5 [j,g/m3 for monitor-
based studies; 9.3 [j,g/m3 to 12.3 [j,g/m3 for studies that use hybrid modeling and apply population
weighting; and 8.1 [j,g/m3 to 11.9 [j,g/m3 for studies that use hybrid modeling and do not apply
population weighting. The study-reported mean concentrations in Canadian studies are more
difficult to compare to the area annual standard design value but are lower than those reported in
the U.S. studies for both monitor-based and hybrid model methods, ranging from 7.0 |j,g/m3 to
9.0 |j,g/m3in monitor-based studies, and 6.0 |j,g/m3 to 10.0 |j,g/m3 in model-based studies. These
mean values are consistent with the mean PM2.5 concentrations reported in studies available at
the time of the 2020 review (U.S. EPA, 2020, Figure 3-8).

In addition to the overall study means, we also consider concentrations somewhat below
the means (e.g., 25th and 10th percentiles), when such information is available from the
epidemiologic studies. This is consistent with approaches used in previous reviews which
focused on consideration of other air quality distribution statistics somewhat below the long-term
mean PM2.5 concentrations reported in epidemiologic studies. In so doing, we continue to note,
as in previous reviews, that in the lower part of the air quality distribution the health data are
appreciably more sparse and our understanding of the magnitude and significance of the

3-201

-------
associations correspondingly becomes more uncertain. For the key studies that we can most
clearly understand how the study-reported means relate to the annual standard metric monitor-
based 25th percentiles of health events correspond to mean PM2.5 concentrations are at or above
11.5 |j,g/m3 and monitor-based 10th percentiles of health events correspond to mean PM2.5
concentrations are at or above 9.8 |j,g/m3. PM2.5 concentrations corresponding to 25th percentiles
of estimated exposures that use hybrid modeling approaches and used population-weighted
approaches to estimate PM2.5 exposures range from 6.7 to 9.1 |j,g/m3 and PM2.5 concentrations
corresponding to 10th percentiles of estimated exposures that use hybrid modeling approaches
and used population-weighted approaches to estimate PM2.5 exposures range from 4.7 to 7.3
|j,g/m3.

In assessing the adequacy of the current standard, we examine a subset of epidemiologic
studies that provide supplementary information related to evaluating the adequacy of the current
standard, many of which are newly available in this reconsideration. The first set of
epidemiologic studies are studies that employ alternative methods for confounder control to more
extensively account for confounders and are more robust to model misspecification (Table 3-11).
These studies report positive and significant associations for a variety of health outcomes and
support the positive and significant associations in analyses identified as key epidemiologic
studies above. We also evaluate what the accountability studies may indicate with respect to
improvements in public health with improvements in air quality. In so doing, we take note of
three accountability studies (Sanders et al., 2020a, Corrigan et al., 2018, and Henneman et al.,
2019a) newly available in this reconsideration with starting concentrations at or below 12.0
|ig/m3 that indicate positive and significant associations with mortality and reductions in ambient
PM2.5 (Table 3-12). We further evaluate studies with analyses that restrict annual average or daily
PM2.5 concentrations to values below level the annual or daily PM2.5 standard, respectively
(Table 3-10). These restricted analyses indicate positive and significant associations, including
mean PM2.5 concentrations presumably below the mean reported PM2.5 in the main cohort, where
long-term mean PM2.5 concentrations range from 8.1 |ig/m3 to 11.6 |ig/m3, as well as effect
estimates that are generally greater in magnitude than effect estimates seen in main analyses.
Epidemiologic studies that restrict annual or daily PM2.5 concentrations provide support for
positive and statistically significant associations at lower mean PM2.5 concentrations, but
uncertainties exist in these analyses (section 3.3.3.2 and section 3.6.1 above), including
uncertainty in how studies exclude concentrations (e.g., at what spatial resolution are
concentrations being excluded), which would make any comparisons of concentrations in
restricted analyses difficult to compare to design values. Further, studies that restrict 24-hour
average PM2.5 concentrations to values of less than 35 |ig/m3 often do not report the means for
the restricted analyses, although the mean of the restricted analysis is presumably less than the

3-202

-------
mean in the main analysis. However, in some studies, the majority of PM2.5 concentrations from
the main study are already less than the restricted concentration (e.g., in Di et al. (2017a), where
of all case and control days, 93.6% had PM2.5 concentrations below 25 |ig/m3), which contributes
to the uncertainty in how much lower a mean concentration in a restricted study is compared to
the mean in the main study. As a result, there are limitations in how this information can be used
in evaluating the adequacy of the current or potential alternative levels of the standard.
Additionally, it is difficult to use the means, when reported, from studies of restricted analyses to
evaluate the level of protection afforded by the current or potential alternative levels of the
primary 24-hour PM2.5 standard because the relationship between the study-reported mean
concentration and the 98th percentile form of the 24-hour standard is not well understood, in
particular for a short-term standard designed to limit exposures to peak PM2.5 concentrations.

In addition to the epidemiologic evidence, we examine experimental studies, including
controlled human exposure studies and animal toxicological studies. As detailed in above in
section 3.3.3.1 and section 3.6.1.3, these studies provide support for the effects of exposure to
PM2.5, and support for biologically plausible mechanisms through which adverse human health
outcomes could occur. Exposures in controlled human exposure studies last from less than one
hour and up to five hours, and indicate that the most consistent evidence is associated with
cardiovascular effects, and more specifically, impaired vascular function. PM2.5 exposures
evaluated in most of these studies are well-above the ambient concentrations typically measured
in locations meeting the current primary standards. For example, at air quality monitoring sites
meeting the current primary PM2.5 standards (i.e., the 24-hour standard and the annual standard),
the 2-hour concentrations generally remain below 10 (J,g/m3, and rarely exceed 30 [j,g/m3. Two-
hour concentrations are higher at monitoring sites violating the current standards, but generally
remain below 16 [j,g/m3 and rarely exceed 80 [j,g/m3. In addition, as noted earlier in section
3.3.3.1, chronic vascular dysfunction can be judged to be a biomarker of an adverse health effect
from air pollution, but the health relevance of acute reductions in vascular function are less
certain (Thurston et al., 2017). Thus, while these studies are important in establishing biological
plausibility, it is unclear how the results alone and the importance of the effects observed in these
studies, particularly in studies conducted at near-ambient PM2.5 concentrations, should be
interpreted with respect to adversity to public health.

In addition to the evidence above, we also consider what the risk assessment indicates
with regard to the adequacy of the current primary PM2.5 standards. The risk assessment
estimates that the current primary PM2.5 standards could allow a substantial number of deaths in
the U.S., with the large majority of those deaths associated with long-term PM2.5 exposures. For
example, when air quality in the 47 study areas is adjusted to simulate just meeting the current
standards, the risk assessment estimates 40,600-45,100 long-term PM2.5 exposure-related deaths

3-203

-------
in a single year, with confidence intervals ranging from 30,300-59,000. While the absolute
numbers of estimated deaths vary across exposure durations, populations, and concentration-
response functions, the general magnitude of risk estimates supports the potential for significant
public health impacts in locations meeting the current primary PM2.5 standards. This is
particularly the case given that the large majority of PIVh.s-associated deaths for air quality just
meeting the current standards are estimated at annual average PM2.5 concentrations from about
10 to 12 |j,g/m3. These annual average PM2.5 concentrations fall well-within the range of long-
term average concentrations over which key epidemiologic studies provide strong support for
reported positive and statistically significant PM2.5 health effect associations.

Based on the information summarized above, and discussed in more detail in sections 3.3,
3.4, and 3.5 of this PA, we particularly note the following in reaching conclusions on the current
primary PM2.5 standards:

• There is a long-standing body of strong health evidence demonstrating relationships between
long- or short-term PM2.5 exposures and a variety of outcomes, including mortality and
serious morbidity effects. Studies assessed in the 2019 ISA and the ISA Supplement have
reduced key uncertainties and broadened our understanding of the health effects that can
result from exposures to PM2.5.

- Recent U.S. and Canadian epidemiologic studies provide support for generally
positive and statistically significant health effect associations across a broad range
of ambient PM2.5 concentrations, including for air quality distributions with
overall mean concentrations lower than in the previous reviews.

Controlled human exposure studies and animal toxicological studies provide
support for the effects of exposure to PM2.5, and support for biologically plausible
mechanisms through which adverse human health outcomes could occur.

- Epidemiologic studies that use alternative methods for confounder control have
expanded since the 2020 PA and further inform the causal nature of the
relationship between short- and long-term exposure to PM2.5 and mortality and
cardiovascular effects. These studies use a variety of statistical approaches that
attempt to more extensively account for confounders and are more robust to
model misspecification.

• Recent U.S. accountability studies provide support for improvements in public health,
including reductions in mortality in studies with starting PM2.5 concentrations at or below the
current primary PM2.5 annual standard. Some epidemiologic studies (Corrigan et al., 2018
and Sanders et al., 2020b) that employ accountability methods using monitored data evaluate
the effect of the implementation of the 1997 annual PM2.5 standard, finding evidence of
reductions in mortality in areas with starting PM2.5 concentrations at or below 12.0 |j,g/m3.

• Studies that restrict analyses to air quality below the current daily or annual PM2.5 standard
exhibit positive and significant associations, which are often greater in magnitude than main
analyses. Di et al. (2017b) and Dominici et al. (2019) report positive and statistically
significant associations that are greater in analyses restricted below 12.0 |ig/m3 and report
mean concentrations of 9.6 |ig/m3. In studies that restrict analyses to 24-hour average

3-204

-------
concentrations < 35.0 |ig/m3 or lower, mean PM2.5 concentrations are not reported, though
such means are presumably somewhat below those based on the overall cohort, which range
from 8.2 |ig/m3 to 11.6 |ig/m3, and effect estimates are generally greater than those in the
overall cohort. More specifically, one U.S. study by Shi et al. (2016) reports positive and
statistically significant associations in analyses restricted to relatively low annual or 24-hour
PM2.5 exposure estimates.

• Evidence assessed in the 2019 ISA and ISA Supplement continues to indicate a linear, no-
threshold concentration-response relationship for PM2.5 concentrations > 8 [j,g/m3. However,
uncertainties remain about the shape of the C-R curve at PM2.5 concentrations < 8 (J,g/m3,
with some recent studies providing evidence for either a sublinear, linear, or supralinear
relationship at these lower concentrations (U.S. EPA, 2019, section 11.2.4; U.S. EPA, 2022,
section 2.2.3.2).

• Exposures in controlled human exposure studies last from less than one hour and up to five
hours and indicate that the most consistent evidence is associated with cardiovascular effects,
and more specifically, impaired vascular function. Further, air quality analyses suggest that
the ambient concentrations in these studies typically do not occur in locations meeting the
current primary standards, thus suggesting that the current primary PM2.5 standards provide
protection against these "peak" concentrations.

• We note the decision framework used in previous reviews that places significant weight on
key epidemiologic studies and consider whether the mean concentrations in these studies
would be allowed in areas meeting the current primary standard.

Such a decision framework placed significant weight on epidemiologic studies
that assessed associations between PM2.5 exposure and health outcomes that were
most strongly supported by the body of scientific evidence and recognized there is
significantly greater confidence in the magnitude and significance of observed
associations for the part of the air quality distribution corresponding to where the
bulk of the health events in each study have been observed, generally at or around
the mean concentration.

Additional analyses, new in this PA though similar to analyses in the 2012
review, suggest that the area annual design value is greater than the study-
reported mean values by 10-20% (monitor-based studies), 15-18% (hybrid
modeling with population weighting) or 40-50% (hybrid modeling without
population weighting).

- Focusing on the key epidemiologic studies available in this reconsideration, the
overall mean PM2.5 concentrations in key U.S. epidemiologic studies are as
follows:

o Range of monitor-based mean PM2.5 concentrations is from 9.9 [j,g/m3
to 16.5 [j,g/m3 (range in 2020 PA: 10.7 [j,g/m3 to 16.5 (J,g/m3)

o Range of mean PM2.5 concentrations in studies that use hybrid
modeling and apply population weighting: 9.3 [j,g/m3 to 12.3 [j,g/m3

o Range of mean PM2.5 concentrations in studies that use hybrid
modeling and do not apply population weighting: 8.1 [j,g/m3 to 11.9
[j,g/m3

3-205

-------
Though the Canadian studies are more difficult to utilize for comparison to the
annual design value used to determine compliance in the U.S., the overall mean
PM2.5 concentrations in key Canadian epidemiologic studies are within the range,
though somewhat lower than those from the U.S. studies, and are as follows:

o Range of monitor-based mean PM2.5 concentrations is from 6.9 |ig/m3
to 13.3 |ig/m3

o Range of mean PM2.5 concentrations in studies that use hybrid
modeling (all of which average up to postal codes and thus include
some aspects of population weighting) is 5.9 |ig/m3 to 9.8 |ig/m3

- Past decision frameworks also placed some weight on considering the annual
standard level relative to the 25th and 10th percentile of exposure concentrations
or health events while also noting that epidemiologic studies provide more limited
support for health effect associations based on air quality distributions at these
lower PM2.5 concentration percentiles.

o In key U.S. epidemiologic studies that use monitors to estimate PM2.5
exposures, 25th percentiles of health events correspond to mean PM2.5
concentrations (i.e., averaged over the study period for each study city)
at or above 11.5 |ig/m3 and 10th percentiles of health events correspond
to mean PM2.5 concentrations at or above 9.8 |ig/m3

• Of the key U.S. epidemiologic studies that use hybrid modeling approaches to estimate long-
term PM2.5 exposures and do not apply population weighting, the ambient PM2.5
concentrations corresponding to 25th percentiles of estimated exposures range from 4.6 |ig/m3
to 9.2 |ig/m3, while in studies that do apply population weighting, 25th percentiles range from
6.7 |ig/m3 to 9.1 |ig/m3. In the two studies (each apply population weighting) with
information available on the 10th percentile of health events, the ambient PM2.5
concentrations corresponding to the 10th percentile are 4.7 |ig/m3 and 7.3 |ig/m3. The risk
assessment estimates that the current primary PM2.5 standards could allow a substantial
number of PM2.5-associated deaths in the U.S. The large majority of these estimated deaths
are associated with the annual average PM2.5 concentrations near (and above in some cases)
the average concentrations in key epidemiologic studies reporting positive and statistically
significant health effect associations. Further, the risk assessment estimated that Black
populations may experience disproportionally higher exposures and risk under simulated air
quality conditions just meeting the current primary PM2.5 annual standard as compared to
White populations.

• The CASAC also provided their recommendations and advice on the adequacy of the current
primary PM2.5 standards (described in more detail in section 3.5 above).

As an initial matter, the CASAC agreed that both primary standards - 24-hour and
annual - are "critical to protect public health given the evidence on detrimental
health outcomes at both short-term and long-term exposures including peak
events" (Sheppard, 2022, p. 13 of consensus responses).

- With regard to the primary annual PM2.5 standard, all CASAC members agreed
that the current level of the primary annual PM2.5 standard is not sufficiently
protective of public health and should be lowered. They also reached consensus
that the indicator, form, and averaging time for the primary annual PM2.5 standard

3-206

-------
should be retained. The majority of the CAS AC members recommended revising
the level of the primary annual PM2.5 standard within the range of 8-10 |ig/m3,
while the minority of the CASAC members recommended revision with the range
of 10-11 |ig/m3.

- With regard to the primary 24-hour PM2.5 standard, the CASAC did not reach
consensus on the adequacy of the current standard. The majority of the CASAC
members found that the available evidence called into question the adequacy of
the current standard and favored lowering the level of the standard to within the
range of 25-30 |ig/m3. The minority of the CASAC agreed with the preliminary
conclusion in the draft PA that the current primary 24-hour PM2.5 standard is
adequate and should be retained, without revision.

When taken together, we reach the conclusion that the available scientific evidence, air quality

analyses, and the risk assessment, as summarized above, can reasonably be viewed as calling

into question the adequacy of the public health protection afforded by the combination of the

current annual and 24-hour primary PM2.5 standards. In particular, we note the information and

analyses new to this reconsideration (and discussed in detail above) in reaching this conclusion.

We also note that this is consistent with the advice and recommendations from the CASAC, as

well as a number of public comments submitted on the draft PA.

3.6.3.2 Potential Alternative Standards

In this section, we consider the potential alternative primary PM2.5 standards that could be
supported by the evidence and quantitative information available in this reconsideration. These
considerations are framed by the following overarching policy-relevant question, posed at the
beginning of this chapter:

• What is the range of potential alternative standards that could be supported by the
available scientific evidence and risk-based information to increase public health
protection against short- and long-term fine particle exposures?

In answering this question, we consider each of the elements of the annual and 24-hour PM2.5
standards: indicator, averaging time, form, and level. The sections below discuss our
consideration of these elements, and our conclusions that (1) it is appropriate to consider revising
the level of the current annual standard, in conjunction with retaining the current indicator,
averaging time, and form of that standard, to increase public health protection against fine
particle exposures and (2) depending on the decision made on the annual standard, consideration
could be given to either retaining or revising the level of the 24-hour PM2.5 standard.

3.6.3.2.1 Indicator

In initially setting standards for fine particles in 1997, the EPA concluded it was
appropriate to control fine particles as a group, rather than singling out any particular component
or class of fine particles. The Agency noted that community health studies had found significant

3-207

-------
health effect associations using various indicators of fine particles, and that health effects in a
large number of areas had significant mass contributions from differing components or sources
of fine particles. In addition, a number of toxicological and controlled human exposure studies
had reported health effects following exposures to high concentrations of numerous fine particle
components (62 FR 38667, July 18, 1997). In establishing a size-based indicator in 1997 to
distinguish fine particles from particles in the coarse mode, the EPA noted that the available
epidemiologic studies of fine particles were based largely on PM2.5 mass. The selection of a 2.5
|j,m size cut additionally reflected the regulatory importance of defining an indicator that would
more completely capture fine particles under all conditions likely to be encountered across the
U.S. and the monitoring technology that was generally available (62 FR 38666 to 38668, July 18,
1997).

Since the 1997 review, studies that evaluate fine particle-related health effects continue to
provide strong support for such effects using PM2.5 mass as the metric for fine particle exposures.
Subsequent reviews have recognized the strength of this evidence, concluding that it has
continued to support a PM2.5 mass-based indicator for a standard meant to protect against fine
particle exposures. In the 2012 review, some studies had additionally examined health effects of
exposures to particular sources or components of fine particles, or to the ultrafine fraction of fine
particles. Based on limitations in such studies, together with the continued strong support for
effects of PM2.5 exposures, the Agency retained PM2.5 mass as the indicator for fine particles and
did not supplement the PM2.5 standards with standards based on particle composition or on the
ultrafine fraction (78 FR 3123, January 15, 2013).

As in the 2012 review, studies assessed the 2019 ISA continue to provide strong support
for health effects following long- and short-term PM2.5 exposures (U.S. EPA, 2019). While some
studies evaluate the health effects of particular sources of fine particles, or of particular fine
particle components, evidence from these studies does not identify any one source or component
that is a better predictor of health effects than PM2.5 mass (U.S. EPA, 2019, section 1.5.4). As
summarized in section 3.6.1 above, the 2019 ISA the evidence confirms and further supports that
many PM2.5 components and sources are associated with health effects, and does not indicate that
any one source or component is consistently more strongly related with health effects than PM2.5
mass (U.S. EPA, 2019, section 1.5.4). Further, the evidence for health effects following
exposures specifically to the ultrafine fraction of fine particles continues to be far more limited
than the evidence for PM2.5 mass, and the varying definitions of UFP, as well as differences in
approaches to administering and measuring UFP, contribute to such limitations (U.S. EPA, 2019,
section 1.4.3). In its advice on the adequacy of the current primary PM2.5 standards, the CASAC
reached consensus that indicator should be retained, without revision (Sheppard, 2022, p. 2 of
consensus letter). Thus, for reasons similar to those discussed in the 2020 review (85 FR 82715,

3-208

-------
December 18, 2020) and consistent with the advice from the CAS AC, we reach the conclusion
that the available information continues to support the PM2.5 mass-based indicator and remains
too limited to support a distinct standard for any specific PM2.5 component or group of
components, and too limited to support a distinct standard for the ultrafine fraction.

3.6.3.2.2 Averaging Time

In 1997, the EPA initially set an annual PM2.5 standard to protect against health effects
associated with both long- and short-term PM2.5 exposures, and a 24-hour standard to supplement
the protection afforded by the annual standard (62 FR 38667 to 38668, July 18, 1997). In
subsequent reviews, the EPA retained both annual and 24-hour averaging times, largely
reflecting the strong evidence for health effects associated with annual and daily PM2.5 exposure
estimates (71 FR 61164, October 17, 2006; 78 FR 3123 to 3124, January 15, 2013; 85 FR 82715,
December 18, 2020).

In this reconsideration, epidemiologic and controlled human exposure studies have
examined a variety of PM2.5 exposure durations. Epidemiologic studies continue to provide
strong support for health effects associated with both long- and short-term PM2.5 exposures based
on annual (or multiyear) and 24-hour PM2.5 averaging periods, respectively.

With regard to short-term exposures in particular, a smaller number of epidemiologic
studies examine associations between sub-daily PM2.5 exposures and respiratory effects,
cardiovascular effects, or mortality. Compared to 24-hour PM2.5 exposure estimates, associations
with sub-daily estimates are less consistent and, in some cases, smaller in magnitude (U.S. EPA,
2019, section 1.5.2.1). In addition, studies of sub-daily exposures typically examine subclinical
effects, rather than the more serious population-level effects that have been reported to be
associated with 24-hour exposures (e.g., mortality, hospitalizations). Taken together, the 2019
ISA concludes that epidemiologic studies do not indicate sub-daily averaging periods are more
closely associated with health effects than the 24-hour average exposure metric (U.S. EPA, 2019,
section 1.5.2.1).

Additionally, while recent controlled human exposure studies provide consistent evidence
for cardiovascular effects following PM2.5 exposures for less than 24 hours (i.e., < 30 minutes to
5 hours), exposure concentrations in these studies are well-above the ambient concentrations
typically measured in locations meeting the current standards (section 3.3.3.1). Thus, these
studies also do not suggest the need for additional protection against sub-daily PM2.5 exposures,
beyond that provided by the current primary standards.

Drawing from the evidence assessed in the 2019 ISA, and the observations noted above,
we reach the conclusion that the available evidence continues to provide strong support for
consideration of retaining the current annual and 24-hour averaging times. The available

3-209

-------
evidence suggests that PM2.5 standards with these averaging times, when coupled with
appropriate forms and levels, can protect against the range of long- and short-term PM2.5
exposures that have been associated with health effects. In its advice on the adequacy of the
current primary PM2.5 standards, the CASAC reached consensus that averaging times for the
standards should be retained, without revision (Sheppard, 2022, p. 2 of consensus letter). Thus,
as in the 2020 review (78 FR 82715, December 18, 2020) and consistent with the advice from
the CASAC, we reach the conclusion that the currently available evidence does not support
considering alternatives to the annual and 24-hour averaging times for standards meant to protect
against long- and short-term PM2.5 exposures.

3.6.3.2.3 Form

The form of a standard defines the air quality statistic that is to be compared to the level
in determining whether an area attains that standard. As in other recent reviews, our foremost
consideration in reaching conclusions on form is the adequacy of the public health protection
provided by the combination of the form and the other elements of the standard.

As noted above, in 1997 the EPA initially set an annual PM2.5 standard to protect against
health effects associated with both long- and short-term PM2.5 exposures and a 24-hour standard
to provide supplemental protection, particularly against the short-term exposures to "peak" PM2.5
concentrations that can occur in some areas (62 FR 38667 to 38668, July 18, 1997). The EPA
established the form of the annual PM2.5 standard as an annual arithmetic mean, averaged over 3
years, from single or multiple community-oriented monitors. That is, the level of the annual
standard was to be compared to measurements made at each community-oriented monitoring site
or, if specific criteria were met, measurements from multiple community-oriented monitoring
sites could be averaged together (i.e., spatial averaging) (62 FR 38671 to 38672, July 18, 1997).
In the 1997 review, the EPA also established the form of the 24-hour PM2.5 standard as the 98th
percentile of 24-hour concentrations at each monitor within an area (i.e., no spatial averaging),
averaged over three years (62 FR at 38671 to 38674, July 18, 1997). In the 2006 review, the EPA
retained these standard forms but tightened the criteria for using spatial averaging with the
annual standard (71 FR 61117, October 17, 2006).89

In the 2012 review, the EPA's consideration of the form of the annual PM2.5 standard
again included a focus on the issue of spatial averaging. An analysis of air quality and population
demographic information indicated that the highest PM2.5 concentrations in a given area tended
to be measured at monitors in locations where the surrounding populations were more likely to

89 Specifically, the Administrator revised spatial averaging criteria such that "(1) [t]he annual mean concentration at
each site shall be within 10 percent of the spatially averaged annual mean, and (2) the daily values for each
monitoring site pair shall yield a correlation coefficient of at least 0.9 for each calendar quarter (71 FR 61167,
October 17, 2006).

3-210

-------
live below the poverty line and to include larger percentages of racial and ethnic minorities (U.S.
EPA, 2011, p. 2-60). Based on this analysis, the 2011 PA concluded that spatial averaging could
result in disproportionate impacts in minority populations and populations with lower SES. The
Administrator concluded that public health would not be protected with an adequate margin of
safety in all locations, as required by law, if disproportionately higher PM2.5 concentrations in
low income and minority communities were averaged together with lower concentrations
measured at other sites in a large urban area. Therefore, she concluded that the form of the
annual PM2.5 standard should be revised to eliminate spatial averaging provisions (78 FR 3124,
January 15, 2013).

In the 2012 review, the EPA also considered the form of the 24-hour PM2.5 standard. The
Agency recognized that the existing 98th percentile form for the 24-hour standard was originally
selected to provide a balance between limiting the occurrence of peak 24-hour PM2.5
concentrations and identifying a stable target for risk management programs. Updated air quality
analyses in the 2012 review provided additional support for the increased stability of the 98th
percentile PM2.5 concentration, compared to the 99th percentile (U.S. EPA, 2011, Figure 2-2, p.
2-62). Thus, the Administrator concluded that it was appropriate to retain the 98th percentile form
for the 24-hour PM2.5 standard (78 FR 3127, January 15, 2013).

In light of the absence of new information in this reconsideration to support consideration
of alternative forms for the primary PM2.5 standards, we conclude that the information available
in previous reviews continues to provide support for the current forms of the annual and 24-hour
PM2.5 standards. As discussed above (section 3.3.1), epidemiologic studies continue to provide
strong support for health effect associations with both long-term (e.g., annual or multi-year) and
short-term (e.g., mostly 24-hour) PM2.5 exposures. These studies provide the strongest support
for such associations for the part of the air quality distribution corresponding to the bulk of the
underlying data, typically around the overall mean concentrations reported (section 3.3.3.2.1).
The form of the current annual standard (i.e., arithmetic mean, averaged over three years)
remains appropriate for targeting "typical" daily and annual exposures around these means of the
PM2.5 air quality distribution. In addition, controlled human exposure studies provide evidence
for health effects following single short-term PM2.5 exposures near the peak concentrations
measured in the ambient air (section 3.3.3.1). Thus, the evidence also supports retaining a
standard focused on providing supplemental protection against short-term peak exposures. The
information available in this reconsideration continues to support the decision to use a 98th
percentile form for a 24-hour standard that is meant to provide a balance between limiting the
occurrence of such peak 24-hour PM2.5 concentrations and identifying a stable target for risk
management programs. In its advice on the adequacy of the current standards, the CASAC
reached consensus that the form of the annual standard should be retained, without revision

3-211

-------
(Sheppard, 2022, p. 2 of consensus letter). While the CASAC provided recommendations
regarding the adequacy of the current 24-hour standard conditional on the current form (i.e., 98th
percentile, averaged over three years), they recommended that in future reviews, the EPA also
consider alternative forms for the primary 24-hour PM2.5 standard (Sheppard, 2022, p. 18 of
consensus responses). Thus, when the information summarized above is taken together and
consistent with the advice from the CASAC, we reach the conclusion that it is appropriate to
consider retaining the forms of the current annual and 24-hour PM2.5 standards, in conjunction
with a revised level of the annual standard as discussed below.

3.6.3.2.4 Level

With regard to level, we specifically address the following policy-relevant question:

• For primary PM2.5 standards defined in terms of the current averaging times and
forms, what potential alternative levels are appropriate to consider in order to
increase public health protection against long- and short-term exposures to PM2.5 in
ambient air?

In answering this question, we consider key epidemiologic studies that evaluate associations
between PM2.5 air quality distributions and mortality or morbidity, controlled human exposure
studies examining effects following short-term PM2.5 exposures, air quality analyses that help to
place these studies into a policy-relevant context, the risk assessment estimates of PM2.5-
associated mortality under various alternative standard scenarios, the advice and
recommendations from the CASAC, and public comments received on the draft PA.

Consideration of the evidence and analyses, as summarized in this chapter, informs our
evaluation of the public health protection that could be provided by alternative annual and 24-
hour standards with revised levels. There are various ways to combine an annual standard (based
on arithmetic mean concentrations) and a 24-hour standard (based on 98th percentile
concentrations), to achieve an appropriate degree of public health protection. In particular, we
recognize that changes in PM2.5 air quality designed to meet an annual standard would likely
result not only in lower short- and long-term PM2.5 concentrations near the middle of the air
quality distribution (i.e., around the mean of the distribution), but also in fewer and lower short-
term peak PM2.5 concentrations. Additionally, changes designed to meet a 24-hour standard, with
a 98th percentile form, would result not only in fewer and lower peak 24-hour PM2.5
concentrations, but also in lower average PM2.5 concentrations.

However, while either standard could be viewed as providing some measure of protection
against both average exposures and peak exposures, the 24-hour and annual standards are not
expected to be equally effective at limiting both types of exposures. Specifically, the 24-hour
standard (with its 98th percentile form) is more directly tied to short-term peak PM2.5
concentrations, and thus more likely to appropriately limit exposures to such concentrations, than

3-212

-------
the more typical concentrations that make up the middle portion of the air quality distribution.
Therefore, compared to a standard that is directly tied to the middle of the air quality distribution,
the 24-hour standard is less likely to appropriately limit the "typical" daily and annual exposures
that are most strongly associated with the health effects observed in epidemiologic studies. In
contrast, the annual standard, with its form based on the arithmetic mean concentration, is more
likely to effectively limit the PM2.5 concentrations that comprise the middle portion of the air
quality distribution, affording protection against the daily and annual PM2.5 exposures that
strongly support associations with the most serious PIVh.s-related effects in epidemiologic studies
(e.g., mortality, hospitalizations).

For these reasons, we focus on alternative levels of the annual PM2.5 standard as the
principle means of providing increased public health protection against the bulk of the
distribution of short- and long-term PM2.5 exposures, and thus protecting against the exposures
that provide strong support for associations with mortality and morbidity in key epidemiologic
studies. We additionally consider the 24-hour standard, with its 98th percentile form, primarily as
a means of providing supplemental protection against the short-term exposures to peak PM2.5
concentrations that can occur in some areas (e.g., those with strong contributions from local or
seasonal sources), even when overall mean PM2.5 concentrations remain relatively low.

To inform our consideration of potential alternative annual and 24-hour standard levels,
we specifically note the key observations in section 3.6.3.1 (rather than repeating them here) and
note more specifically, related to those observations that:

Mean PM2.5 Concentrations in Key Epidemiologic Studies and Relationships between Mean
PM2.5 Concentrations and Annual Design Values

• The relationship between the mean PM2.5 concentrations and the area design value continues
to be an important consideration in evaluating the adequacy of the current or potential
alternative annual standard levels. In a given area, the area design value is the monitor in an
area with the highest PM2.5 concentrations and is used to determine compliance with the
standard. The highest PM2.5 concentrations spatially distributed in the area would generally
occur at or near the area design value monitor and the distribution of PM2.5 concentrations
would generally be lower in other locations and monitors in that area. That is, study-reported
means for an area's air quality conditions will generally be below the area design value
monitor, as our analyses show, since the means are generally based on composite monitors or
averages of estimated concentrations. As such, setting a standard level that requires the
design value monitor to meet study-reported means will generally result in lower
concentrations of PM2.5 across the entire area, such that even those people living near an area
design value monitor (where PM concentrations are generally highest) will be exposed to
levels below the air quality conditions reported in the epidemiologic studies.

• Areas meeting a particular annual PM2.5 standard would be expected to have average PM2.5
concentrations (i.e., averaged across the area and over time) somewhat below the level of that

3-213

-------
standard (which is measured at the peak monitor). This is supported by analyses of
monitoring data in CBSAs across the U.S., which show that maximum annual PM2.5 design
values are often 10% to 20% higher than long-term mean PM2.5 concentrations in an area
(section 2.3.3.1, Figure 2-28; Table 2-3). Additional analyses also support differences
between annual PM2.5 design values and long-term mean PM2.5 concentrations in hybrid
modeling studies, with the extent of the difference depending on the methods used to
estimate mean PM2.5 concentrations. These analyses suggest that the area annual design
values are generally higher than the study mean by 15-18% (hybrid modeling with
population weighting) or 40-50% higher (hybrid modeling without population weighting)
(section 2.3.3.2.4, Table 2-7).

• Most key U.S. epidemiologic studies indicate consistently positive and statistically
significant health effect associations based on air quality distributions with overall mean
PM2.5 concentrations at or above 9.3 |ig/m3 (9.9 |ig/m3 based on U.S. studies that use
monitors to estimate PM2.5 exposures). Other key epidemiologic studies (which do not
incorporate population weighting into their calculation of the study mean) report mean PM2.5
concentrations to be as low as 8.1 |ig/m3 with the air quality analyses suggesting that areas
included in these studies would have corresponding area annual design values generally 40-
50%) higher than the study-reported mean concentrations.

• Though the mean PM2.5 concentrations from Canadian studies are more difficult to directly
compare to the annual design value used to determine compliance in the U.S., the overall
mean PM2.5 concentrations in key Canadian epidemiologic studies are close to, though
somewhat lower than, those from the U.S. studies. The range of monitor-based mean PM2.5
concentrations is from 6.9 |ig/m3 to 13.3 |ig/m3 while the range of mean PM2.5 concentrations
in studies that use hybrid modeling (all of which average up to postal codes and thus include
some aspects of population weighting) is 5.9 |ig/m3 to 9.8 |ig/m3.

• Epidemiologic studies provide more limited support for health effect associations based on
air quality distributions at lower PM2.5 percentile concentrations. In assessing the 25th
percentile of data, PM2.5 concentrations in key U.S. epidemiologic studies that use hybrid
modeling methods and do not apply some aspects of population weighting range from 4.6
|ig/m3 to 9.2 |ig/m3, while those that apply some aspects of population weighting range from
6.7 |ig/m3 to 9.1 |ig/m3. In U.S. studies that use monitored values have 25th percentiles
ranging from 11.5 |ig/m3 to just below 13.0 |ig/m3. In Canada two monitored studies report
25th percentile concentrations around 6.5 |ig/m3, while hybrid modeled studies in Canada, all
of which average up to postal codes and thus include some aspects of population weighting,
report 25th percentile concentrations around 8.0 |ig/m3 in two studies, and 4.3 |ig/m3 in one
study.

Scientific Evidence and Associated Uncertainties Supporting Associations at Lower

Concentrations

• Recent evidence further demonstrates that associations with mortality remain robust in
copollutants analyses (U.S. EPA, 2019, section 11.2.3), and that associations persist in
analyses restricted to long-term exposures below 12 |j,g/m3 (Di et al., 2017b) or 10 |j,g/m3
(Shi et al., 2016) (i.e., indicating that risks are not disproportionately driven by the upper
portions of the air quality distribution).

3-214

-------
• Studies that examine the shapes of concentration-response functions over the full distribution
of ambient PM2.5 concentrations have not identified a threshold concentration, below which
associations no longer exist (U.S. EPA, 2019, section 1.5.3, U.S. EPA, 2022, section 2.2.3.1
and 2.2.3.2). While such analyses are complicated by the relatively sparse data available at
the lower end of the air quality distribution (U.S. EPA, 2019, section 1.5.3), the evidence
remains consistent in supporting a no-threshold relationship, and in supporting a linear
relationship for PM2.5 concentrations > 8 [j,g/m3. However, uncertainties remain about the
shape of the C-R curve at PM2.5 concentrations < 8 (j,g/m3, with some recent studies
providing evidence for either a sublinear, linear, or supralinear relationship at these lower
concentrations (U.S. EPA, 2022, section 3.2.2.2).

• While there is no specific point in the air quality distribution of any epidemiologic study that
represents a "bright line" at and above which effects have been observed and below which
effects have not been observed, there is significantly greater confidence in the magnitude and
significance of observed associations for the part of the air quality distribution corresponding
to where the bulk of the health events in each study have been observed, generally at or
around the mean concentration, with more limited support for health effect associations
based on air quality distributions at lower PM2.5 percentile concentrations.

• Controlled human exposure studies demonstrate consistent evidence of effects at higher
concentrations (e.g., > 120 |ig/m3) and provide support for biological plausibility for more
serious effects (e.g., hospital admissions) (U.S. EPA, 2019, Figure 6-1).

Scientific Evidence on Short-term PM2.5 Exposures and Health Effects

• While controlled human exposure studies support the plausibility of the serious
cardiovascular effects that have been linked with ambient PM2.5 exposures (U.S. EPA, 2019,
chapter 6), the PM2.5 exposure concentrations evaluated in most of these studies are well-
above the ambient concentrations typically measured in locations meeting the current
primary standards (and thus well-above those likely to be measured in locations that would
meet revised standards with lower annual or 24-hour levels) (Figure 2-19, Figure A-2, Figure
A-3).

PM2.5-AssociatedRisk Estimates

• The risk assessment estimates that, compared to the current standards, potential alternative
annual standards with levels from 11.0 down to 8.0 |j,g/m3 could reduce PIVh.s-associated
mortality broadly across the United States. Meeting a revised annual standard with a lower
level is estimated to reduce PIVh.s-associated health risks in the 30 annually-controlled study
areas by about 7-9% for a level of 11.0 |ig/m3, 15-19% for a level of 10.0 |ig/m3, 22-28% for
a level of 9.0 |ig/m3, and 30-37% for a level of 8.0 |ig/m3, compared to the current annual
standard.

• Revising the level of the 24-hour standard to 30 |j,g/m3 is estimated to lower PIVh.s-associated
risks across a more limited population and number of areas than revising the annual standard
(section 3.4.2.3). Risk reduction predictions are largely confined to areas located in the
western U.S., several of which are also likely to experience risk reductions upon meeting a
revised annual standard.

3-215

-------
• The at-risk assessment estimated that Black populations may experience disproportionally
higher exposures and risk under simulated air quality conditions just meeting the current
primary PM2.5 annual standard as compared to White populations. Meeting a revised annual
standard with a lower level may also proportionally reduce exposure and risk in Black
populations slightly more so than in White populations in simulated scenarios just meeting
alternative annual standards.

• Uncertainties in risk estimates (e.g., in the size of risk estimates) can result from a number of
factors, including assumptions about the shape of the C-R relationship with mortality at low
ambient PM concentrations, the potential for confounding and/or exposure measurement
error in the underlying epidemiologic studies, and the methods used to adjust PM2.5 air
quality. In considering such uncertainties, we recognize that the risk estimates can help to
place the evidence for specific effects into a broader public health context, but should be
considered along with the inherent uncertainties and limitations of such analyses when
informing judgments about the potential for additional public health protection associated
with PM2.5 exposure and related health effects.

At-risk Populations and Public Health Implications

• The 2019 ISA cites extensive evidence indicating that "both the general population as well as
specific populations and lifestages are at risk for PIVh.s-related health effects" (U.S. EPA,
2019, p. 12-1).

• Factors that may contribute to increased risk of PIVh.s-related health effects include lifestage
(children and older adults), pre-existing diseases (cardiovascular disease and respiratory
disease), race/ethnicity, and socioeconomic status. There is also strong evidence for racial
and ethnic differences in PM2.5 exposures and PM2.5-related health risks.

• At-risk populations make up a substantial portion of the U.S. population (section 3.3.2
above):

- Lifestage: children (age <18) - 22%; older adults (age 65+) - 16%

- Race/ethnicity: non-Hispanic Black - 12%; Hispanic - 18%

• Prevalence of pre-existing diseases vary by lifestage and race/ethnicity (section 3.3.2 above):

Cardiovascular diseases: Older adults (age 65+) have a higher prevalence than
younger adults (age <64); American Indians or Alaska Natives have a higher
prevalence of hypertension and stroke; Black populations have the highest
prevalence of hypertension and stroke.

- Respiratory diseases: Older adults have a higher prevalence of emphysema than
younger adults; adults age 44+ have a higher prevalence of chronic bronchitis;
prevalence of chronic bronchitis and emphysema is generally similar across
racial/ethnic groups; asthma prevalence is generally similar across age groups, but
is higher among Black and American Indian or Alaska Native populations.

The information summarized in these key observations could support various decisions on

the levels of the annual and 24-hour PM2.5 standards, depending on the weight given to different

3-216

-------
aspects of the evidence, air quality and risk information, including its uncertainties. In this PA
we seek to provide as broad an array of policy options as is supportable by the available evidence
and quantitative information, recognizing that the selection of a specific approach to reaching
final decisions on the primary PM2.5 standards will reflect the judgments of the Administrator as
to what weight to place on the various types of evidence and information, and on associated
uncertainties. Potential approaches to considering support for particular alternative annual and
24-hour standard levels are discussed below.

Alternative Annual Standard Levels

As discussed above, the degree to which particular alternative annual standard levels
below 12.0 |ig/m3 are supported will depend on the weight placed on various aspects of the
scientific evidence, air quality and risk information, and its associated uncertainties. In selecting
a particular level from 10.0 |ig/m3 to < 12.0 |ig/m3, consideration of the evidence could take into
account individual study characteristics such as study design and statistical approaches, precision
of reported associations, study size and location, and uncertainties in the study itself or in our
analyses of study area air quality. For example, a level below 12 |ig/m3 and as low as about 10.0
|ig/m3 could be supported to the extent weight is placed on the following:

• Setting a standard expected to maintain the PM2.5 air quality distributions below those
present in most key epidemiologic studies, recognizing the general relationships
demonstrated in the air quality analyses between study mean calculation and the annual
standard and noting the values of the study-reported means as listed below:

The monitor-based key epidemiologic studies report mean PM2.5 concentrations
from 9.9 [j,g/m3 to 16.5 |ig/m3;

The key epidemiologic studies that incorporate hybrid modeling and population-
weight study mean PM2.5 concentrations report means from 9.3 [j,g/m3 to 12.2
[j,g/m3.

• Noting the challenges in drawing a direct comparison between the Canadian study means and
the annual design value metric used for compliance in the U.S., but also noting that the
study-reported means from the Canadian studies are similar, though somewhat lower, than
those in the U.S.

• Setting a standard level within the starting range of the mean PM2.5 concentrations evaluated
in accountability studies, recognizing that some of the studies that report public health
improvements with improvements to air quality have starting concentrations that range
between 10.0 |ig/m3 to 12.0 |ig/m3 (Table 3-12).

• Setting a standard estimated to reduce PIVh.s-associated health risks, such that a substantial
portion of the risk reduction that would be accomplished is attributable to annual average
PM2.5 concentrations within the range of overall means for which key epidemiologic studies
indicate consistently positive and statistically significant health effect associations (> about 8
Hg/m3) while also noting important uncertainties inherent in the risk assessment as described

3-217

-------
in detail in sections 3.4.1.7 and 3.4.1.8. Further, the at-risk analyses indicate that the average
percent reduction in PM2.5 concentrations and risk are slightly greater in the Black population
than in the White population for each alternative standard evaluated (11.0 |ig/m3 and 10.0
|ig/m3), when shifting from the current annual PM2.5 standard (12.0 |ig/m3) in the full set of
47 areas and the subset of 30 areas controlled by the annual standard (section 3.4).

• Noting a number of uncertainties associated with the scientific evidence and risk information
including: (1) there are few key epidemiologic studies (and only one key U.S. study) that
report positive and statistically significant health effect associations for PM25 air quality
distributions with overall mean concentrations below 9.6 |ag/m\ and areas meeting a standard
with a level of 10.0 ng/m3 would generally be expected to have lower long-term mean
PM25 concentrations (and potentially around 8.0 ng/m3 in some areas) (section 3.3.3.2.1); (2)
there is increasing uncertainty in PM25 exposure estimates in some of the largest key studies
at lower ambient concentrations (i.e., those that use hybrid model predictions to estimate
exposures), given the more limited information available to develop and validate model
predictions (sections 2.3.3 and 3.3.3.2.1); and (3) there is increasing uncertainty in
quantitative estimates of PM25-associated mortality risk for standard levels below 10.0 |ag/m\
given that a substantial proportion of the risk reductions estimated for lower standard
levels occur at annual average PM25 concentrations below 8 |ag/m\ and thus below the lower
end of the range of overall mean PM25 concentrations in key epidemiologic studies that
consistently report positive and statistically significant associations (section 3.4.1.7).

• Setting the level of the annual standard within the range of 10.0 |ig/m3 to < 12 |ig/m3 would
be consistent with the advice of the minority of CAS AC members. A level of 10.0 |ig/m3 for
the annual standard is within the range of alternative standards recommended by all CAS AC
members, and all members of the CASAC agree that an annual standard level < 12.0 |ig/m3 is
supported by a large and coherent body of evidence.

In contrast, an annual standard with a level below 10.0 |ig/m3 and as low as 8.0 |ig/m3,
could be supported to the extent greater weight is placed on the potential public health
improvements that could result from additional reductions in ambient PM2.5 concentrations (i.e.,
beyond those achieved by a standard with a level of 10.0 |ig/m3) and less weight is placed on the
limitations in the evidence that contribute to greater uncertainty at lower concentrations. For
example, a level below 10.0 |ig/m3 could be supported to the extent greater weight is placed on
the following:

• Setting the annual standard at or below most or all of the study-reported means, including
means of hybrid modeling studies that did not use population-weighted approaches, such that
the standard would be expected to maintain the PM2.5 air quality distributions further below
those present in most key epidemiologic studies and noting that the relationships between
study mean calculation and the annual standard in the PA analyses are approximations and
less weight should be placed on them and the mathematical approach used to calculate the
mean.

• Results of the key Canadian epidemiologic studies, which report mean PM2.5 concentrations
that are lower than those reported in U.S. studies and for which the PM2.5 concentrations
generally range from 7.0 |ig/m3 to 13.3 |ig/m3 (monitor-based) and 6.0 |ig/m3 to 10.0 |ig/m3

3-218

-------
(hybrid model-based and all of which apply some aspects of population weighting) (section
3.3.3.2.1);

• Consideration of the air quality distribution below the mean for which key epidemiologic
studies have reported associations with health effects. The ambient PM2.5 concentrations
around the 25th percentile of underlying data, which range from 11.5 |ig/m3 to 12.9 |ig/m3 in
U.S. monitor-based studies, from 6.5 |ig/m3 to 6.8 |ig/m3 in Canadian monitor-based studies.
In key U.S. epidemiologic studies that use hybrid modeling methods and do not apply some
aspects of population weighting range from 4.6 |ig/m3 to 9.2 |ig/m3, while those that apply
some aspects of population weighting range from 6.7 |ig/m3 to 9.1 |ig/m3 while hybrid
modeled studies in Canada, all of which average up to postal codes and thus include some
aspects of population weighting, report 25th percentile concentrations around 8.0 |ig/m3 in
two studies, and 4.3 |ig/m3 in one study (section 3.3.3.2.1);

• Noting studies that examined the shapes of concentration-response functions over the full
distribution of ambient PM2.5 concentrations and concluded that the evidence remains
consistent in supporting a no-threshold relationship, and in supporting a linear relationship
for PM2.5 concentrations > 8 [j,g/m3. However, uncertainties remain about the shape of the C-
R curve at PM2.5 concentrations < 8 (J,g/m3, with some recent studies providing evidence for
either a sublinear, linear, or supralinear relationship at these lower concentrations

• The potential for continued public health improvements with improvements in air quality
below the lowest starting concentration evaluated in accountability studies, which was
approximately 10.0 |ig/m3 (Table 3-12);

• Studies that restrict analyses to air quality associated with levels below the current annual
standard and report positive and significant associations, often with effect estimates that are
greater in magnitude than those reported in the main analysis. Although the mean of the
restricted analyses are generally not reported, in one key U.S. epidemiologic study, the mean
concentration when restricting annual average PM2.5 concentrations to below 12.0 |ig/m3 was
presumably lower than the overall mean concentration of 8.1 |ig/m3 reported in the main
analysis (Shi et al., 2016) (Table 3-10);

• The potential public health importance of the additional reductions in PIVh.s-associated health
risks estimated for a level below 10.0 |ig/m3 and the potential for continued improvements
below the lowest level examined in the risk assessment (8.0 |j,g/m3). Further, the at-risk
analyses indicate that the average percent reduction in PM2.5 concentrations and risk are
slightly greater in the Black population than in the White population for each alternative
standard evaluated (9.0 |ig/m3 and 8.0 |ig/m3), when shifting from the current annual PM2.5
standard (12.0 |ig/m3) in the full set of 47 areas and the subset of 30 areas controlled by the
annual standard (section 3.4).

• Setting a standard within the range of 8.0 |ig/m3 to 10.0 |ig/m3 would be within the range
supported by the majority of CASAC members.

Alternative 24-Hour Standard Levels

We additionally evaluate the degree to which the evidence supports considering potential

alternative levels for the 24-hour PM2.5 standard, in conjunction with the current 98th percentile

3-219

-------
form of that standard. With respect to current and recent air quality relationships, we note that
the risk assessment indicates that the annual standard is the controlling standard across most of
the urban study areas evaluated and revising the level of the 24-hour standard to 30 |ig/m3 would
be estimated to lower PIVh.s-associated risks, compared to the current standards, largely in a few
study areas located in the western U.S. (several of which are also likely to experience risk
reductions upon meeting a revised annual standard). Additionally, recent air quality analyses
indicate that almost all CBSAs with maximum annual PM2.5 design values at or below 12.0
|ig/m3 also have maximum 24-hour PM2.5 design values below 35 |j,g/m3 (and below 30 |j,g/m3 in
most areas) (chapter 2, Figure 2-18). The exceptions are a few CBSAs in the western U.S.

As in previous reviews, we recognize that the annual standard would generally be the
controlling standard across much of the U.S., except for certain areas where there are high
seasonal emissions (e.g., wood smoke) and conducive meteorology (e.g., temperature inversions)
or where there are more unique source-oriented influences (e.g., near manufacturing sources). In
such areas, the 24-hour standard is the generally controlling standard, though the number of these
areas in the U.S. is small. Thus, as was the approach in multiple recent reviews, we focus on the
annual standard as the principle means of limiting both long- and short-term PM2.5
concentrations, recognizing that the 24-hour standard, with its 98th percentile form, would
provide supplemental protection against short-term peak exposures, particularly for areas with
high peak-to-mean ratios (e.g., areas with strong seasonal sources). Compared to the annual
standard, we recognize that the 24-hour standard is less likely to appropriately limit the more
typical PM2.5 exposures (i.e., corresponding to the middle portion of the air quality distribution)
that are most strongly associated with the health effects observed in epidemiologic studies. Thus,
as in previous reviews (78 FR 3161-3162, January 15, 2013; 85 FR 82715, December 18, 2020),
we focus on the 24-hour standard as a means of providing supplemental protection against the
short-term exposures to "peak" PM2.5 concentrations, such as can occur in areas with strong
contributions from local or seasonal sources.

Taking into account this approach, an important consideration is whether additional
protection is needed against short-term exposures to peak PM2.5 concentrations in areas meeting
both the current 24-hour standard and the current, or a revised, annual standard. To the extent
that the evidence indicates that such exposures can lead to adverse health effects, it would be
appropriate to consider alternative levels for the 24-hour standard. In considering this issue, we
evaluate the evidence from key health studies. With regard to these studies, we particularly note
the following:

• Controlled human exposure studies provide evidence for health effects following single,
short-term PM2.5 exposures to concentrations that typically correspond to upper end of the
PM2.5 air quality distribution in the U.S. (i.e., "peak" concentrations). In the studies evaluated

3-220

-------
at near ambient PM2.5 concentrations, results are mixed, but they do report statistically
significant effects on one or more indicators of cardiovascular function following 2-hour
exposures to PM2.5 concentrations at and above 120 [j,g/m3 (at and above 149 [j,g/m3 for
vascular impairment, the effect shown to be most consistent across studies).

• Animal toxicologic studies provide evidence of effects related to short-term exposures to
PM2.5 at concentrations ranging from 100 to > 1,000 [j,g/m3 and providing further evidence to
support the biological mechanisms and plausibility of various adverse effects associated with
short-term exposures.

The body of epidemiologic evidence provides limited support forjudging adequacy of the
level of the 24-hour standard. As discussed in detail above (section 3.3.3.2.1), epidemiologic
studies provide the strongest support for reported health effect associations for the part of the air
quality distribution corresponding to the bulk of the underlying data (i.e., estimated exposures
and/or health events), often around the overall mean concentrations evaluated rather than near
the upper end of the distribution. Additionally, the magnitudes of the associations in restricted
analyses are similar to or larger than the magnitudes of the associations based on the full cohorts
(Table 3-10), suggesting that, at a minimum, short-term exposures to peak PM2.5 concentrations
are not disproportionately responsible for reported health effect associations. However, while
restricted analyses report positive and significant associations when excluding PM2.5
concentrations >35 [j,g/m3 uncertainties also exist, including the fact that mean concentrations
are not reported in these studies, as well as the difficulties in using a mean concentration to
inform the level of a standard with a 98th percentile form. Based on the evidence above, we
assessed the protection provided by the current standards against the concentrations seen in the
controlled human exposure studies. The air quality analyses included in this PA show that 2-hour
ambient concentrations of PM2.5 at monitoring sites meeting the current standards almost never
exceed 30 [j,g/m3 (Figure 2-19). In fact, even the extreme upper end of the distribution of 2-hour
PM2.5 concentrations (i.e., 99.9th percentile of 2-hour concentrations at these sites is 62 [j,g/m3
during the warm season) at sites meeting the current standards remain well-below the PM2.5
exposure concentrations consistently shown to elicit effects (e.g., 120 (j,g/m3). While sites
meeting the current standards remain well below concentrations shown to elicit effects, we also
note some caution in placing too much weigh on the need to provide protection against any of
the exposures observed in the controlled human exposure studies given that it is unclear how the
results alone and the importance of the effects observed in these studies, particularly in the
studies conducted at near-ambient PM2.5 concentrations, should be interpreted with respect to
adversity to public health. Further, these air quality analyses are not used to establish whether a
no-effect threshold exists, but rather are used to inform whether concentrations observed to elicit
effects in controlled human exposure studies (e.g., concentrations > 120 [j,g/m3 for 2-, 4-, and 5-
hour exposures) occur in ambient air.

3-221

-------
When the information summarized above is considered in the context of the 24-hour
standard, we reach the conclusion that, in conjunction with a lower annual standard level
intended to increase protection against average short- and long-term PM2.5 exposures across the
U.S., the evidence does not support the need for additional protection against short-term
exposures to peak PM2.5 concentrations. In particular, while the epidemiologic studies do support
the need to consider increasing protection against the typical daily and annual PM2.5 exposures
that provide strong support for reported health effect associations, these studies do not provide
the same support for a need for increasing protection against short-term, peak exposures. Further,
the epidemiologic studies do not indicate that the reported health effect associations in these
studies are strongly influenced by exposures to the peak concentrations in the air quality
distribution. Also, while animal toxicologic studies provide evidence to support the biological
mechanisms and plausibility of various adverse effects associated with short-term exposures,
they provide limited support forjudging adequacy of the level of the 24-hour standard.

Controlled human exposure studies support the occurrence of effects following single short-term
exposures to PM2.5 concentrations that correspond to the peak of the air quality distribution,
though these concentrations are well above those typically measured in areas meeting the current
standards. Generally, this would suggest that, in areas meeting the current standards,
concentrations observed in the controlled human exposure studies are rarely occuring in ambient
air. As such, the available evidence supports the need for the current 24-hour standard to protect
against peak concentrations but does not clearly support the need for a lower level of that
standard. Thus, in the context of a 24-hour standard that is meant to provide supplemental
protection (i.e., beyond that provided by the annual standard alone) against short-term exposures
to peak PM2.5 concentrations, the evidence supports consideration of retaining the current 24-
hour standard with its level of 35 |j,g/m3.

However, we also recognize that a different policy approach than that described above
could be applied to considering the level of the 24-hour standard. For example, consideration
could be given to lower 24-hour standard levels in order to increase protection across the U.S.
against the broader PM2 5 air quality distribution. If such an approach is evaluated in this
reconsideration, consideration of 24-hour standard levels as low as 30 |j,g/m3 could be supported
(either alone or in conjunction with a lower annual standard level). The risk assessment estimates
that a level of 30 |j,g/m3 would increase protection compared to the current standards, though
only in a small number of study areas largely confined to the western U.S. (section 3.4.2).
Additionally, more weight could be placed on restricted analyses that evaluate short-term
exposures that indicate positive and significant associations in studies that restrict daily
concentrations less than 35 |j,g/m3 (see Table 3-10). However, as described in more detail in
sections 3.3.3 and 3.6.1 above, there are a number of limitations associated with such studies,

3-222

-------
including how studies exclude concentrations (e.g., at what spatial resolution are concentrations
being excluded) and a lack of reporting of the mean concentrations for the restricted analysis.
Furthermore, when mean concentrations are reported for the restricted analysis, it is difficult to
understand how these means relate to the level of the 24-hour standard, given that the standard
has a 98th percentile form, rather than a daily average.

If this alternative approach to revising the primary PM2.5 standards is adopted, the
uncertainty inherent in using the 24-hour standard to increase protection against the broad
distribution of PM2.5 air quality should be carefully considered. Specifically, the degree of
protection provided by any particular 24-hour standard against the typical PM2 5 exposures
corresponding to the middle portion of the air quality distribution will vary across locations and
over time, depending on the relationship between those typical concentrations and the short-term
peak PM25 concentrations that are directly targeted by the 24-hour standard (i.e., with its 98th
percentile form). Thus, lowering the level of the 24-hour standard is likely to have a more
variable impact on public health than lowering the level of the annual standard. Depending on
the 24-hour standard level set, some areas could experience reductions that are greater than
warranted, based on the evidence, while others could experience reductions that are less than
warranted. Therefore, the rationale supporting this approach would need to recognize and
account for the uncertainty inherent in using 24-hour standard, with a 98th percentile form, to
increase protection against the broad distribution of PM2.5 air quality.

3.7 AREAS FOR FUTURE RESEARCH AND DATA COLLECTION

In this section, we identify key areas for additional research and data collection for fine
particles, based on the uncertainties and limitations that remain in the evidence and technical
information. Additional research in these areas could reduce uncertainties and limitations in
future reviews of the primary PM2.5 standards. Important areas for future research include the
following:

• Further elucidating the physiological pathways through which exposures to the PM2.5
concentrations present in the ambient air across much of the U.S., specifically the
progression of one subclinical event to another ultimately resulting in the mortality and
morbidity effects reported in many epidemiologic studies. This could include the following:

Controlled human exposure studies that examine exposures at near-ambient PM2.5
concentrations (e.g., Wyatt et al. (2020a); longer exposure periods (e.g., 5-hour as
in Hemmingsen et al. (2015b)), or repeated exposures to concentrations typically
measured in the ambient air across the U.S.

Studies that evaluate the health impacts of decreasing PM2.5 exposures (e.g., due
to changes in policies or behavior, shifts in important emissions sources, or
targeted interventions).

3-223

-------
• Additional research into alternative methods for confounder control90 in epidemiologic
studies to evaluate the causal nature of relationships between PM2.5 exposure and mortality or
morbidity.

• Additional research into accountability or quasi-experimental epidemiologic studies with
starting PM2.5 concentrations below 12.0 |ig/m3,

• Improving our understanding of the PM2.5 concentration-response relationships including the
shapes of concentration-response functions and the uncertainties around estimated functions
for various health outcomes and populations (e.g., older adults, people with pre-existing
diseases, children).

• Understanding of the potential for particle characteristics, other than size-fractionated mass,
to influence PM toxicity (e.g., composition, oxidative potential, etc.) and the PM health
effect associations observed in epidemiologic studies.

• Understanding the potential for combinations of components that may be associated with
specific sources to influence PM toxicity and PM-related health effect associations observed
in epidemiologic studies. In addition, accountability studies that evaluate information on
health effects associated with changes in PM composition, including changes in PM
composition associated with groups of components and their associated sources.

• Improving our understanding of the uncertainties inherent in the various approaches used to
estimate PM2.5 exposures in epidemiologic studies, including how those uncertainties may
vary across space and time, and over the PM2.5 air quality distribution. Approaches to
incorporating these uncertainties into quantitative estimates of PM2.5 concentration-response
relationships should also be explored.

• Additional health research on UFPs, with a focus on consistently defining UFPs across
studies and across disciplines (i.e., animal, controlled human exposure, and epidemiologic
studies), on using consistent exposure approaches in experimental studies, and on improving
exposure characterizations in epidemiologic studies. Also, further examine the potential for
translocation of UFPs from the respiratory tract into other compartments (i.e., blood) and
organs (e.g., heart, brain), with particular emphasis on studies conducted in humans.

• Additional work to measure UFP emissions, understand UFP sources, and the composition of
UFPs, using comparable methods to measure emissions from various types of sources (e.g.,
mobile sources, fires, etc.).

• Further evaluate the potential for some groups to be at higher risk of PM2.5-related effects
than the general population and the potential for PM2.5 exposures to contribute to the
development of underlying conditions that may then confer higher risk of PM2.5-related
effects. This includes additional exploration of disparities in exposure and health risks in
communities with environmental justice concerns, with a specific emphasis on minority
populations and tribal communities. More refined epidemiologic analyses in these
populations could further inform both qualitative and quantitative assessments of PM2.5-
related health risks.

90 These studies are often called causal inference or causal modeling studies in the scientific literature.

3-224

-------
Additionally, refined epidemiologic analyses that investigate the relationship
between exposure disparities and health risk disparities across different
demographic groups, including communities with environmental justice concerns,
with a specific emphasis on minority populations and tribal communities.

• Exploration of the relationship between short- and long-term PM2.5 exposures in the
exacerbation and progression of disease and how they interact with one another. Research
efforts such as this could elucidate the potential for long-term PM exposures to contribute to
the development and progression of atherosclerosis in adults and/or asthma in children, as
well as the triggering of a myocardial infarction and asthma exacerbation. It could also
include research to understand the potential role of PM exposures, both short- and long term,
in developmental outcomes (e.g., neurodevelopmental effects, reproductive and birth
outcomes).

• Research to further evaluate the combination of factors that contribute to differences in risk
estimates between cities, potentially including differences in exposures, demographics,
particle characteristics.

• Research to improve our understanding of variability in PM2.5 exposures within and across
various populations (e.g., defined by lifestage, pre-existing condition, etc.),, as well as the
temporal and spatial variability in ambient PM2.5 that is not captured by existing ambient
monitors. Epidemiologic and toxicologic studies examining health effects of air pollution
exposure during pregnancy and associations with adverse pregnancy outcomes (e.g., fetal
growth restriction, preterm birth) as well as outcomes in childhood (e.g., neurodevelopment
and lung development) should also include examination of potential mechanisms mediating
and modifying such effects.

• Future research further examining copollutant confounding and the role of varying
spatiotemporal scales of copollutant variability relative to the resolution of PM2.5, and the
possible role of sampling variability in the interpretation of small changes in coefficient
estimates when comparing estimates with and without copollutant adjustment. Future
epidemiologic research evaluating the relationship between short-term exposure and long-
term exposure and health, including how the different exposure periods interact with one
another and their association with different health outcomes.

In addition to research and data collection, additional information that could be reported

in epidemiologic studies may help to reduce uncertainties and limitations in future reviews of the

primary PM2.5 standards. This information includes:

• Descriptive statistics of PM2.5 concentrations that are used in epidemiologic studies to
evaluate associations between PM2.5 and health effects (e.g., minimum, maximum, 10th
percentile, 25th percentile, mean, median, 75th percentile).

In addition, and specifically in epidemiologic studies of short-term exposure,
descriptive statistics of PM2.5 concentrations at individual percentiles from the
95th percentile to the 99th percentile, as well as the number of days of
concentrations and/or health events within each of these percentiles.

• More detailed information on the methods used to calculate the mean PM2.5 concentrations
that are reported in the study (e.g., whether population weighting was applied, how the PM2.5

3-225

-------
concentrations estimated from hybrid modeling are averaged prior to being assigned to health
events).

- Noting whether the mean PM2.5 concentration reported is the concentration across
the area evaluated or if the mean PM2.5 concentration reported is based only PM2.5
concentrations used in analyses to assess the association between health outcomes
and PM2.5 (i.e., is the reported mean the estimated PM2.5 concentration across the
study area or is the reported mean the mean concentration assigned to study
individuals).

- Providing population-weighted means and other descriptive statistics.

• In analyses restrict PM2.5 concentrations below specific concentrations (e.g., below annual
averages of 12.0 |ig/m3 or below daily averages of 35 |ig/m3) reporting the mean PM2.5
concentrations, as well as other descriptive statistics in the restricted analysis could be
helpful.

• In analyses restricting PM2.5 concentrations below specific concentrations (e.g., below annual
averages of 12.0 |ig/m3 or below daily averages of 35 |ig/m3), a more detailed description of
how these analyses were completed (i.e., are individual grid cells removed, are specific ZIP
codes removed) and an explanation of limitations or uncertainties in the analyses would be
useful.

3-226

-------
REFERENCES

ATS (2000). What Constitutes an Adverse Health Effect of Air Pollution? American Journal of
Respiratory and Critical Care Medicine 161(2): 665-673.

Awad, YA, Di, Q, Wang, Y, Choirat, C, Coull, BA, Zanobetti, A and 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(4): e054.

Bai, L, Shin, S, Burnett, RT, Kwong, JC, Hystad, P, van Donkelaar, A, Goldberg, MS, Lavigne,
E, Copes, R, Martin, RV, Kopp, A and 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. Environment
International 132: 105004.

Baxter, LK, Crooks, JL and Sacks, JD (2017). Influence of exposure differences on city-to-city
heterogeneity in PM2.5-mortality associations in US cities. Environmental Health: A
Global Access Science Source 16(1): 1.

Behbod, B, Urch, B, Speck, M, Scott, JA, Liu, L, Poon, R, Coull, B, Schwartz, J, Koutrakis, P,
Silverman, F and Gold, DR (2013). Endotoxin in concentrated coarse and fine ambient
particles induces acute systemic inflammation in controlled human exposures.
Occupational and Environmental Medicine 70(11): 761-767.

Bell, ML, Ebisu, K, Leaderer, BP, Gent, JF, Lee, HJ, Koutrakis, P, Wang, Y, Dominici, F and
Peng, RD (2014). Associations of PM2.5 constituents and sources with hospital
admissions: analysis of four counties in Connecticut and Massachusetts (USA) for
persons > 65 years of age. Environmental Health Perspectives 122(2): 138-144.

Bell, ML, Ebisu, K, Peng, RD, Walker, J, Samet, JM, Zeger, SL and Dominic, F (2008).

Seasonal and regional short-term effects of fine particles on hospital admissions in 202
U.S. counties, 1999-2005. American Journal of Epidemiology 168(11): 1301-1310.

Bell, ML, Son, JY, Peng, RD, Wang, Y and Dominici, F (2015). Brief report: Ambient PM2.5 and
risk of hospital admissions: do risks differ for men and women? Epidemiology 26(4):
575-579.

Bellavia, A, Urch, B, Speck, M, Brook, RD, Scott, JA, Albetti, B, Behbod, B, North, M, Valeri,
L, Bertazzi, PA, Silverman, F, Gold, D and Baccarelli, AA (2013). DNA
hypomethylation, ambient particulate matter, and increased blood pressure: Findings
from controlled human exposure experiments. Journal of the American Heart Association
2(3): e000212.

Bennett, JE, Tamura-Wicks, H, Parks, RM, Burnett, RT, Pope, CA, Bechle, MJ, Marshall, JD,
Danaei, G and Ezzati, M (2019). Particulate matter air pollution and national and county

3-227

-------
life expectancy loss in the USA: A spatiotemporal analysis. PLoS Medicine 16(7):
el002856.

Berhane, K, Chang, CC, McConnell, R, Gauderman, WJ, Avol, E, Rapapport, E, Urman, R,

Lurmann, F and Gilliland, F (2016). Association of changes in air quality with bronchitic
symptoms in children in California, 1993-2012. JAMA: Journal of the American Medical
Association 315(14): 1491-1501.

Bevan, GH, Freedman, DA, Lee, EK, Rajagopalan, S and Al-Kindi, SG (2021). Association
between ambient air pollution and county-level cardiovascular mortality in the United
States by social deprivation index. American Heart Journal 235: 125-131.

Bourdrel, T, Annesi-Maesano, I, Alahmad, B, Maesano, CN and Bind, MA (2021). The impact
of outdoor air pollution on COVID-19: a review of evidence from in vitro, animal, and
human studies. European Respiratory Review 30(159).

Brauner, EV, M0ller, P, Barregard, L, Dragsted, LO, Glasius, M, Wahlin, P, Vinzents, P,
Raaschou-Nielsen, O and Loft, S (2008). Exposure to ambient concentrations of
particulate air pollution does not influence vascular function or inflammatory pathways in
young healthy individuals. Particle and Fibre Toxicology 5:13.

Bravo, MA, Ebisu, K, Dominici, F, Wang, Y, Peng, RD and Bell, ML (2017). Airborne fine
particles and risk of hospital admissions for understudied populations: Effects by
urbanicity and short-term cumulative exposures in 708 U.S. counties. Environmental
Health Perspectives 125(4): 594-601.

Brook, RD, Urch, B, Dvonch, JT, Bard, RL, Speck, M, Keeler, G, Morishita, M, Marsik, FJ,
Kamal, AS, Kaciroti, N, Harkema, J, Corey, P, Silverman, F, Gold, DR, Wellenius, G,
Mittleman, MA, Rajagopalan, S and Brook, JR (2009). Insights into the mechanisms and
mediators of the effects of air pollution exposure on blood pressure and vascular function
in healthy humans. Hypertension 54(3): 659-667.

Burnett, RT and Goldberg, MS (2003). Size-fractionated particulate mass and daily mortality in
eight Canadian cities. Health Effects Institute Special Report: Revised Analyses of Time-
Series Studies of Air Pollution and Health. Boston, MA. Available at:
https://www.healtheffects.org/svstem/files/TimeSeries.pdf.

Burnett, RT, Stieb, D, Brook, JR, Cakmak, S, Dales, R, Raizenne, M, Vincent, R and Dann, T
(2004). Associations between short-term changes in nitrogen dioxide and mortality in
Canadian cities. Archives of Environmental and Occupational Health 59(5): 228-236.

Cakmak, S, Hebbern, C, Pinault, L, Lavigne, E, Vanos, J, Crouse, DL and Tjepkema, 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. Environment International 111: 200-211.

3-228

-------
Cangerana Pereira, FA, Lemos, M, Mauad, T, de Assuncao, JV and Nascimento Saldiva, PH

(2011). Urban, traffic-related particles and lung tumors in urethane treated mice. Clinics
66(6): 1051-1054.

Chan, E. (2019). Email and attachment from Elizabeth Chan, EPA, Question re: Request for
additional PM2.5 information from Di et al 2017. Available in PM NAAQS Review
Docket (EPA-HQ-OAR-2015-0072). December 3, 2019. Office of Air Quality Planning
and Standards Research Triangle Park, NC.

Chen, H, Burnett, RT, Copes, R, Kwong, JC, Villeneuve, PJ, Goldberg, MS, Brook, RD, van
Donkelaar, A, Jerrett, M, Martin, RV, Brook, JR, Kopp, A and Tu, JV (2016). Ambient
fine particulate matter and mortality among survivors of myocardial infarction:
population-based cohort study. Environmental Health Perspectives 124(9): 1421-1428.

Chen, H, Zhang, ZL, van Donkelaar, A, Bai, L, Martin, RV, Lavigne, E, Kwong, JC and Burnett,
RT (2020). Understanding the Joint Impacts of Fine Particulate Matter Concentration and
Composition on the Incidence and Mortality of Cardiovascular Disease: A Component-
Adjusted Approach. Environmental Science and Technology 54(7): 4388-4399.

Christidis, T, Erickson, AC, Pappin, AJ, Crouse, DL, Pinault, LL, Weichenthal, SA, Brook, JR,
van Donkelaar, A, Hystad, P, Martin, RV, Tjepkema, M, Burnett, RT and Brauer, M

(2019). Low concentrations of fine particle air pollution and mortality in the Canadian
Community Health Survey cohort. Environmental Health: A Global Access Science
Source 18(1): 84.

Correia, AW, Pope, CA, III, Dockery, DW, Wang, Y, Ezzati, M and Dominici, F (2013). Effect
of air pollution control on life expectancy in the United States: an analysis of 545 U.S.
counties for the period from 2000 to 2007. Epidemiology 24(1): 23-31.

Corrigan, AE, Becker, MM, Neas, LM, Cascio, WE and Rappold, AG (2018). Fine particulate
matters: The impact of air quality standards on cardiovascular mortality. Environmental
Research 161: 364-369.

Cox, LA. (2019). Letter from Louis Anthony Cox, Jr., Chair, Clean Air Scientific Advisory
Committee, to Administrator Andrew R. Wheeler. Re: CASAC Review of the EPA's
Policy Assessment for the Review of the National Ambient Air Quality Standards for
Particulate Matter (External Review Draft - September 2019). December 16, 2019. EPA-
CASAC-20-001. Office of the Administrator, Science Advisory Board U.S. EPA HQ,
Washington DC. Available at:

https://vosemite.epa.gov/sab/sabproduct.nsf/264cbl227d55e02c85257402007446a4/E2F
6C7173 72016128525 84D20069DFB l/$File/EP A-C AS AC-20-001 .pdf.

Crouse, DL, Erickson, AC, Christidis, T, Pinault, L, van Donkelaar, A, Li, C, Meng, J, Martin,
RV, Tjepkema, M, Hystad, P, Burnett, R, Pappin, A, Brauer, M and Weichenthal, S

(2020). Evaluating the Sensitivity of PM2.5-Mortality Associations to the Spatial and
Temporal Scale of Exposure Assessment. Epidemiology 31(2): 168-176.

3-229

-------
Crouse, DL, Peters, PA, Hystad, P, Brook, JR, van Donkelaar, A, Martin, RV, Villeneuve, PJ,
Jerrett, M, Goldberg, MS, Pope, CA, Brauer, M, Brook, RD, Robichaud, A, Menard, R
and Burnett, RT (2015). Ambient PM 2.5, 03, and N02 exposures and associations with
mortality over 16 years of follow-up in the Canadian Census Health and Environment
Cohort (CanCHEC). Environmental Health Perspectives 123(11): 1180-1186.

Crouse, DL, Peters, PA, van Donkelaar, A, Goldberg, MS, Villeneuve, PJ, Brion, O, Khan, S,
Atari, DO, Jerrett, M, Pope, CA, Brauer, M, Brook, JR, Martin, RV, Stieb, D and
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. Environmental Health Perspectives 120(5): 708-714.

Dai, L, Zanobetti, A, Koutrakis, P and Schwartz, JD (2014). Associations of fine particulate
matter species with mortality in the United States: a multicity time-series analysis.
Environmental Health Perspectives 122(8): 837-842.

deSouza, P, Braun, D, Parks, RM, Schwartz, J, Dominici, F and Kioumourtzoglou, MA (2021).
Nationwide Study of Short-term Exposure to Fine Particulate Matter and Cardiovascular
Hospitalizations Among Medicaid Enrollees. Epidemiology 32(1): 6-13.

Devlin, RB, Ghio, AJ, Kehrl, H, Sanders, G and Cascio, W (2003). Elderly humans exposed to
concentrated air pollution particles have decreased heart rate variability. European
Respiratory Journal 40: 76S-80S.

Di, Q, Amini, H, Shi, L, Kloog, I, Silvern, R, Kelly, J, Sabath, MB, Choirat, C, Koutrakis, P and
Lyapustin, A (2019a). An ensemble-based model of PM2.5 concentration across the
contiguous United States with high spatiotemporal resolution. Environment International
130: 104909.

Di, Q, Amini, H, Shi, L, Kloog, I, Silvern, R, Kelly, J, Sabath, MB, Choirat, C, Koutrakis, P,

Lyapustin, A, Wang, Y, Mickley, LJ and Schwartz, J (2019b). An ensemble-based model
of PM2.5 concentration across the contiguous United States with high spatiotemporal
resolution. Environment International 130: 104909.

Di, Q, Dai, L, Wang, Y, Zanobetti, A, Choirat, C, Schwartz, JD and Dominici, F (2017a).

Association of short-term exposure to air pollution with mortality in older adults. JAMA:
Journal of the American Medical Association 318(24): 2446-2456.

Di, Q, Wang, Y, Zanobetti, A, Wang, Y, Koutrakis, P, Choirat, C, Dominici, F and Schwartz, JD
(2017b). Air pollution and mortality in the Medicare population. New England Journal of
Medicine 376(26): 2513-2522.

Dominici, F, Peng, RD, Bell, ML, Pham, L, McDermott, A, Zeger, SL and Samet, JL (2006).
Fine particulate air pollution and hospital admission for cardiovascular and respiratory
diseases. JAMA: Journal of the American Medical Association 295(10): 1127-1134.

Dominici, F, Schwartz, J, Di, Q, Braun, D, Choirat, C and Zanobetti, A (2019). Assessing

adverse health effects of long-term exposure to low levels of ambient air pollution: Phase

3-230

-------
1. Health Effects Institute. Boston, MA. Available at:
https://www.healtheffects.org/svstem/files/dominici-rr-200-report.pdf.

Elliott, EG, Laden, F, James, P, Rimm, EB, Rexrode, KM and 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. Environmental Health
Perspectives 128(12): 127012.

Erickson, AC, Brauer, M, Christidis, T, Pinault, L, Crouse, DL, van Donkelaar, A, Weichenthal,
S, Pappin, A, Tjepkema, M, Martin, RV, Brook, JR, Hystad, P and Burnett, RT (2019).
Evaluation of a method to indirectly adjust for unmeasured covariates in the association
between fine particulate matter and mortality. Environmental Research 175: 108-116.

Erickson, AC, Christidis, T, Pappin, A, Brook, JR, Crouse, DL, Hystad, P, Li, C, Martin, RV,
Meng, J, Pinault, L, von Donkelaar, A, Weichenthal, S, Tjepkema, M, Burnett, RT and
Brauer, M (2020). Disease assimilation: The mortality impacts of fine particulate matter
on immigrants to Canada. Health Reports 31(3): 14-26.

Eum, K, Suh, HH, Pun, V and Manjourides, 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(2):
e009.

Fan, M and 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. Environmental Health: A
Global Access Science Source 19(1): 28.

Fisher, JA, Puett, RC, Laden, F, Wellenius, GA, Sapkota, A, Liao, D, Yanosky, JD, Carter-

Pokras, O, He, X and Hart, JE (2019). Case-crossover analysis of short-term particulate
matter exposures and stroke in the health professionals follow-up study. Environment
International 124: 153-160.

Frank, N. (2012). Memorandum to PM NAAQS Review Docket (EPA-HQ-OAR-2007-0492)
regarding the Differences between maximum and composite monitor annual PM2.5
design values by CBSA. Dec 14, 2012. Docket ID No. EPA-HQ-OAR-2007-0492. Office
of Air Quality Planning and Standards Research Triangle Park, NC. Available at:
https://www.regulations.gov/document/EPA-HO-OAR-2007-0492-10Q99.

Franklin, M, Koutrakis, P and Schwartz, J (2008). The role of particle composition on the
association between PM2.5 and mortality. Epidemiology 19(5): 680-689.

Franklin, M, Zeka, A and Schwartz, J (2007). Association between PM2.5 and all-cause and
specific-cause mortality in 27 US communities. Journal of Exposure Science and
Environmental Epidemiology 17(3): 279-287.

Gauderman, WJ, Urman, R, Avol, E, Berhane, K, McConnell, R, Rappaport, E, Chang, R,
Lurmann, F and Gilliland, F (2015). Association of improved air quality with lung
development in children. New England Journal of Medicine 372(10): 905-913.

3-231

-------
Gharibvand, L, Shavlik, D, Ghamsary, M, Beeson, WL, Soret, S, Knutsen, R and Knutsen, SF
(2016). The association between ambient fine particulate air pollution and lung cancer
incidence: Results from the AHSMOG-2 study. Environmental Health Perspectives
125(3): 378-384.

Ghio, AJ, Hall, A, Bassett, MA, Cascio, WE and Devlin, RB (2003). Exposure to concentrated
ambient air particles alters hematologic indices in humans. Inhalation Toxicology 15(14):
1465-1478.

Ghio, AJ, Kim, C and Devlin, RB (2000). Concentrated ambient air particles induce mild

pulmonary inflammation in healthy human volunteers. American Journal of Respiratory
and Critical Care Medicine 162(3): 981-988.

Gong, H, Jr., Linn, WS, Sioutas, C, Terrell, SL, Clark, KW, Anderson, KR and Terrell, LL
(2003). Controlled exposures of healthy and asthmatic volunteers to concentrated
ambient fine particles in Los Angeles. Inhalation Toxicology 15(4): 305-325.

Gong, H, Jr., Linn, WS, Terrell, SL, Anderson, KR, Clark, KW, Sioutas, C, Cascio, WE, Alexis,
N and Devlin, RB (2004). Exposures of elderly volunteers with and without chronic
obstructive pulmonary disease (COPD) to concentrated ambient fine particulate pollution.
Inhalation Toxicology 16(11-12): 731-744.

Greven, S, Dominici, F and Zeger, S (2011). An Approach to the Estimation of Chronic Air

Pollution Effects Using Spatio-Temporal Information. Journal of the American Statistical
Association 106(494): 396-406.

Hart, JE, Garshick, E, Dockery, DW, Smith, TJ, Ryan, L and Laden, F (2011). Long-term

ambient multi-pollutant exposures and mortality. American Journal of Respiratory and
Critical Care Medicine 183(1): 73-78.

Hart, JE, Liao, X, Hong, B, Puett, RC, Yanosky, JD, Suh, H, Kioumourtzoglou, MA,

Spiegelman, D and Laden, F (2015). 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. Environmental Health: A Global Access Science Source 14: 38.

Hassett-Sipple, B, Schmidt, M and Raj an, P. (2010). Memorandum to PM NAAQS Review

Docket (EPA-HQ-OAR-2007-0492). Analysis of PM2.5 (Particulate Matter Smaller than
2.5 Micrometers in Diameter). Mar 30, 2010. Docket ID No. EPA-HQ-OAR-2007-0492.
Office of Air Quality Planning and Standards Research Triangle Park, NC. Available at:
https://www.regulations.gov/document/EPA-HQ-OAR-2007-0492-0Q77.

Hayes, RB, Lim, C, Zhang, YL, Cromar, K, Shao, YZ, Reynolds, HR, Silverman, DT, Jones,

RR, Park, Y, Jerrett, M, Ahn, J and Thurston, GD (2020). PM2.5 air pollution and cause-
specific cardiovascular disease mortality. International Journal of Epidemiology 49(1):
25-35.

Hazucha, MJ, Bromberg, PA, Lay, JC, Bennett, W, Zeman, K, Alexis, NE, Kehrl, H, Rappold,
A, Cascio, WE and Devlin, RB (2013). Pulmonary responses in current smokers and ex-

3-232

-------
smokers following a two hour exposure at rest to clean air and fine ambient air particles.
Particle and Fibre Toxicology 10(1): 58.

Hemmingsen, JG, Jantzen, K, Miller, P and 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(5): 635-642.

Hemmingsen, JG, Rissler, J, Lykkesfeldt, J, Sallsten, G, Kristiansen, J, P, PM and 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. Particle
and Fibre Toxicology 12(1): 6.

Henneman, LR, Choirat, C and Zigler, CM (2019a). Accountability assessment of health

improvements in the United States associated with reduced coal emissions between 2005
and 2012. Epidemiology (Cambridge, Mass.) 30(4): 477.

Henneman, LRF, Choirat, C and Zigler, ACM (2019b). Accountability assessment of health

improvements in the United States associated with reduced coal emissions between 2005
and 2012. Epidemiology 30(4): 477-485.

Higbee, JD, Lefler, JS, Burnett, RT, Ezzati, M, Marshall, JD, Kim, SY, Bechle, M, Robinson,
AL and Pope, CA (2020). Estimating long-term pollution exposure effects through
inverse probability weighting methods with Cox proportional hazards models.
Environmental Epidemiology 4(2): e085.

Honda, T, Eliot, MN, Eaton, CB, Whitsel, E, Stewart, JD, Mu, L, Suh, H, Szpiro, A, Kaufman,
JD, Vedal, S and Wellenius, GA (2017). Long-term exposure to residential ambient fine
and coarse particulate matter and incident hypertension in post-menopausal women.
Environment International 105: 79-85.

Huang, YC, Rappold, AG, Graff, DW, Ghio, AJ and Devlin, RB (2012). Synergistic effects of

exposure to concentrated ambient fine pollution particles and nitrogen dioxide in humans.
Inhalation Toxicology 24(12): 790-797.

Hystad, P, Demers, PA, Johnson, KC, Brook, J, van Donkelaar, A, Lamsal, L, Martin, R and
Brauer, M (2012). Spatiotemporal air pollution exposure assessment for a Canadian
population-based lung cancer case-control study. Environmental Health: A Global Access
Science Source 11: 22.

Hystad, P, Demers, PA, Johnson, KC, Carpiano, RM and Brauer, M (2013). Long-term

residential exposure to air pollution and lung cancer risk. Epidemiology 24(5): 762-772.

Ito, K, Ross, Z, Zhou, J, Nadas, A, Lippmann, M and Thurston, GD (2013). National Particle
Component Toxicity (NPACT) initiative: Study 3. Time-series analysis of mortality,
hospitalizations, and ambient PM2.5 and its components. 177. Health Effects Institute.
Boston, MA. Available at: https://www.healtheffects.org/svstem/files/RR177-Lippmann-
Studv3-AppendixG O.pdf.

3-233

-------
Jenkins, S. (2016). Email from Scott Jenkins to Robert Devlin. Re: Average PM2.5 concentration
in Devlin et al. (2003). Available in PM NAAQS Review Docket (EPA-HQ-OAR-2015-
0072-0021). March 2, 2016. Office of Air Quality Planning and Standards Research
Triangle Park, NC. Available at: https://www.regulations.gov/document/EPA-HQ-QAR-
2015-0072-0021.

Kioumourtzoglou, MA, Schwartz, J, James, P, Dominici, F and Zanobetti, A (2016). PM2.5 and
mortality in 207 us cities: Modification by temperature and city characteristics.
Epidemiology 27(2): 221-227.

Kloog, I, Coull, BA, Zanobetti, A, Koutrakis, P and Schwartz, JD (2012). Acute and chronic
effects of particles on hospital admissions in New-England. PLoS ONE 7(4): e34664.

Kloog, I, Nordio, F, Zanobetti, A, Coull, BA, Koutrakis, P and Schwartz, JD (2014). Short term
effects of particle exposure on hospital admissions in the Mid-Atlantic states: a
population estimate. PLoS ONE 9(2): e88578.

Kloog, I, Ridgway, B, Koutrakis, P, Coull, BA and Schwartz, JD (2013). Long- and short-term
exposure to PM2.5 and mortality: Using novel exposure models. Epidemiology 24(4):
555-561.

Krall, JR, Anderson, GB, Domini ci, F, Bell, ML and Peng, RD (2013). Short-term exposure to
particulate matter constituents and mortality in a national study of U.S. urban
communities. Environmental Health Perspectives 121(10): 1148-1153.

Krewski, D, Jerrett, M, Burnett, RT, Ma, R, Hughes, E, Shi, Y, Turner, MC, Pope, CA, III,
Thurston, G, Calle, EE, Thun, MJ, Beckerman, B, Deluca, P, Finkelstein, N, Ito, K,
Moore, DK, Newbold, KB, Ramsay, T, Ross, Z, Shin, H and Tempalski, B (2009).
Extended follow-up and spatial analysis of the American Cancer Society study linking
particulate air pollution and mortality. ISSN 1041-5505, HEI Research Report 140.
Health Effects Institute. Boston, MA. Available at:
https://www.healtheffects.org/svstem/files/Krewskil40Statement.pdf.

Kusha, M, Masse, S, Farid, T, Urch, B, Silverman, FS, Brook, RD, Gold, DR, Mangat, I, Speck,
M, Nair, K, Poku, K, Meyer, C, Mittleman, MA, Wellenius, GA and Nanthakumar, K
(2012). Controlled exposure study of air pollution and t wave alternans in volunteers
without cardiovascular disease. Environmental Health Perspectives 120(8): 1157-1161.

Laden, F, Schwartz, J, Speizer, FE and Dockery, DW (2006). Reduction in fine particulate air
pollution and mortality: extended follow-up of the Harvard Six Cities study. American
Journal of Respiratory and Critical Care Medicine 173(6): 667-672.

Lavigne, E, Burnett, RT and Weichenthal, S (2018). Association of short-term exposure to fine
particulate air pollution and mortality: effect modification by oxidant gases. Scientific
Reports 8(1): 16097.

3-234

-------
Lee, M, Koutrakis, P, Coull, B, Kloog, I and Schwartz, J (2015). Acute effect of fine particulate
matter on mortality in three Southeastern states from 2007-2011. Journal of Exposure
Science and Environmental Epidemiology 26(2): 173-179.

Lefler, JS, Higbee, JD, Burnett, RT, Ezzati, M, Coleman, NC, Mann, DD, Marshall, JD, Bechle,
M, Wang, Y, Robinson, AL and Arden Pope, C (2019). Air pollution and mortality in a
large, representative U.S. cohort: multiple-pollutant analyses, and spatial and temporal
decompositions. Environmental Health: A Global Access Science Source 18(1): 101.

Lepeule, J, Laden, F, Dockery, D and Schwartz, J (2012). Chronic exposure to fine particles and
mortality: an extended follow-up of the Harvard Six Cities study from 1974 to 2009.
Environmental Health Perspectives 120(7): 965-970.

Lipfert, FW and Wyzga, RE (2020). Environmental predictors of survival in a cohort of U.S.
military veterans: A multi-level spatio-temporal analysis stratified by race.

Environmental Research 183(Elsevier): 108842.

Lippmann, M, Chen, LC, Gordon, T, Ito, K and Thurston, GD (2013). National Particle

Component Toxicity (NPACT) Initiative: Integrated epidemiologic and toxicologic
studies of the health effects of particulate matter components: Investigators' Report. 177.
Health Effects Institute. Boston, MA.

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, Iniguez, C, Forsberg, B, Astrom, C, Ragettli, MS, Guo, YL, Chen,
BY, Bell, ML, Wright, CY, Scovronick, N, Garland, RM, Milojevic, A, Kysely, J, Urban,
A, Orru, H, Indermitte, E, Jaakkola, JJK, Ryti, NRI, Katsouyanni, K, Analitis, A,
Zanobetti, A, Schwartz, J, Chen, J, Wu, T, Cohen, A, Gasparrini, A and Kan, H (2019).
Ambient Particulate Air Pollution and Daily Mortality in 652 Cities. New England
Journal of Medicine 381(8): 705-715.

Liu, L, Urch, B, Poon, R, Szyszkowicz, M, Speck, M, Gold, DR, Wheeler, AJ, Scott, JA, Brook,
JR, Thorne, PS and Silverman, FS (2015). Effects of ambient coarse, fine, and ultrafine
particles and their biological constituents on systemic biomarkers: A controlled human
exposure study. Environmental Health Perspectives 123(6): 534-540.

Loop, MS, McClure, LA, Levitan, EB, Al-Hamdan, MZ, Crosson, WL and Safford, MM (2018).
Fine particulate matter and incident coronary heart disease in the REGARDS cohort.
American Heart Journal 197: 94-102.

Lucking, AJ, Lundback, M, Barath, SL, Mills, NL, Sidhu, MK, Langrish, JP, Boon, NA,

Pourazar, J, Badimon, JJ, Gerlofs-Nijland, ME, Cassee, FR, Boman, C, Donaldson, K,
Sandstrom, T, Newby, DE and Blomberg, A (2011). Particle traps prevent adverse
vascular and prothrombotic effects of diesel engine exhaust inhalation in men.

Circulation 123(16): 1721-1728.

3-235

-------
Mauad, T, Rivero, DH, de Oliveira, RC, Lichtenfels, AJ, Guimaraes, ET, de Andre, PA,

Kasahara, DI, Bueno, HM and Saldiva, PH (2008). Chronic exposure to ambient levels of
urban particles affects mouse lung development. American Journal of Respiratory and
Critical Care Medicine 178(7): 721-728.

McClure, LA, Loop, MS, Crosson, W, Kleindorfer, D, Kissela, B and Al-Hamdan, M (2017).

Fine Particulate Matter (PM2.5) and the Risk of Stroke in the REGARDS Cohort. Journal
of Stroke & Cerebrovascular Diseases 26(8): 1739-1744.

McConnell, R, Islam, T, Shankardass, K, Jerrett, M, Lurmann, F, Gilliland, F, Gauderman, J,

Avol, E, Kiinzli, N, Yao, L, Peters, J and Berhane, K (2010). Childhood incident asthma
and traffic-related air pollution at home and school. Environmental Health Perspectives
118(7): 1021-1026.

Miller, KA, Siscovick, DS, Sheppard, L, Shepherd, K, Sullivan, JH, Anderson, GL and
Kaufman, JD (2007). Long-term exposure to air pollution and incidence of
cardiovascular events in women. New England Journal of Medicine 356(5): 447-458.

NHLBI (2017). "NHLBI fact book, fiscal year 2012: Disease statistics." Retrieved August 23,
2017, from https://www.nhlbi.nih.gov/about/documents/factbook/2012/chapter4.

Ostro, B, Malig, B, Hasheminassab, S, Berger, K, Chang, E and Sioutas, C (2016). Associations
of source-specific fine particulate matter with emergency department visits in California.
American Journal of Epidemiology 184(6): 450-459.

Pappin, AJ, Christidis, T, Pinault, LL, Crouse, DL, Brook, JR, Erickson, A, Hystad, P, Li, C,
Martin, RV, Meng, J, Weichenthal, S, van Donkelaar, A, Tjepkema, M, Brauer, M and
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. Environmental Health Perspectives 127(10): 107008.

Parker, JD, Kravets, N and Vaidyanathan, 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(16): 1688-1697.

Peng, R, Bell, M, Geyh, A, McDermott, A, Zeger, S, Samet, J and Dominici, F (2009).

Emergency admissions for cardiovascular and respiratory diseases and the chemical
composition of fine particle air pollution. Environmental Health Perspectives 117(6):
957-963.

Peterson, GCL, Hogrefe, C, Corrigan, AE, Neas, LM, Mathur, R and Rappold, AG (2020).

Impact of Reductions in Emissions from Major Source Sectors on Fine Particulate
Matter-Related Cardiovascular Mortality. Environmental Health Perspectives 128(1):
17005.

Pinault, L, Brauer, M, Crouse, DL, Weichenthal, S, Erickson, A, van Donkelaar, A, Martin, RV,
Charbonneau, S, Hystad, P, Brook, JR, Tjepkema, M, Christidis, T, Menard, R,
Robichaud, A and Burnett, RT (2018). Diabetes Status and Susceptibility to the Effects of

3-236

-------
PM2.5 Exposure on Cardiovascular Mortality in a National Canadian Cohort.
Epidemiology 29(6): 784-794.

Pinault, L, Tjepkema, M, Crouse, DL, Weichenthal, S, van Donkelaar, A, Martin, RV, Brauer,
M, Chen, H and Burnett, RT (2016). Risk estimates of mortality attributed to low
concentrations of ambient fine particulate matter in the Canadian community health
survey cohort. Environmental Health: A Global Access Science Source 15(1): 18.

Pope, CA, III,, I, Burnett, RT, Thurston, GD, Thun, MJ, Calle, EE, Krewski, D and Godleski, JJ
(2004). Cardiovascular mortality and long-term exposure to particulate air pollution:
epidemiological evidence of general pathophysiological pathways of disease. Circulation
109(1): 71-77.

Pope, CA, III, Ezzati, M and Dockery, DW (2009). Fine-particulate air pollution and life

expectancy in the United States. New England Journal of Medicine 360(4): 376-386.

Pope, CA, Lefler, JS, Ezzati, M, Higbee, JD, Marshall, JD, Kim, SY, Bechle, M, Gilliat, KS,

Vernon, SE, Robinson, AL and Burnett, RT (2019). Mortality Risk and Fine Particulate
Air Pollution in a Large, Representative Cohort of U.S. Adults. Environmental Health
Perspectives 127(7): 77007.

Pope, CA, Turner, MC, Burnett, R, Jerrett, M, Gapstur, SM, Diver, WR, Krewski, D and Brook,
RD (2015). Relationships between fine particulate air pollution, cardiometabolic
disorders and cardiovascular mortality. Circulation Research 116(1): 108-U258.

Puett, RC, Hart, JE, Suh, H, Mittleman, M and Laden, F (2011). Particulate matter exposures,
mortality and cardiovascular disease in the health professionals follow-up study.
Environmental Health Perspectives 119(8): 1130-1135.

Puett, RC, Hart, JE, Yanosky, JD, Paciorek, C, Schwartz, J, Suh, H, Speizer, FE and Laden, F
(2009). Chronic fine and coarse particulate exposure, mortality, and coronary heart
disease in the Nurses' Health Study. Environmental Health Perspectives 117(11): 1697-
1701.

Puett, RC, Hart, JE, Yanosky, JD, Spiegelman, D, Wang, M, Fisher, JA, Hong, B and Laden, F
(2014). Particulate matter air pollution exposure, distance to road, and incident lung
cancer in the Nurses' Health Study cohort. Environmental Health Perspectives 122(9):
926-932.

Pun, VC, Kazemiparkouhi, F, Manjourides, J and Suh, HH (2017). Long-term PM2.5 exposures
and respiratory, cancer and cardiovascular mortality in American older adults. American
Journal of Epidemiology 186(8): 961-969.

Qiu, X, Wei, Y, Wang, Y, Di, Q, Sofer, T, Awad, YA and 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. Environmental Research 182: 109095.

3-237

-------
Raaschou-Nielsen, O, Andersen, ZJ, Beelen, R, Samoli, E, Stafoggia, M, Weinmayr, G,

Hoffmann, B, Fischer, P, Nieuwenhuijsen, MJ, Brunekreef, B, Xun, WW, Katsouyanni,
K, Dimakopoulou, K, Sommar, J, Forsberg, B, Modig, L, Oudin, A, Oftedal, B,

Schwarze, PE, Nafstad, P, De Faire, U, Pedersen, NL, Ostenson, CG, Fratiglioni, L,
Penell, J, Korek, M, Pershagen, G, Eriksen, KT, S0rensen, M, Tj0nneland, A, Ellermann,
T, Eeftens, M, Peeters, PH, Meliefste, K, Wang, M, Bueno-De-mesquita, B, Key, TJ, De
Hoogh, K, Concin, H, Nagel, G, Vilier, A, Grioni, S, Krogh, V, Tsai, MY, Ricceri, F,
Sacerdote, C, Galassi, C, Migliore, E, Ranzi, A, Cesaroni, G, Badaloni, C, Forastiere, F,
Tamayo, I, Amiano, P, Dorronsoro, M, Trichopoulou, A, Bamia, C, Vineis, P and Hoek,
G (2013). Air pollution and lung cancer incidence in 17 European cohorts: Prospective
analyses from the European Study of Cohorts for Air Pollution Effects (ESCAPE). The
Lancet Oncology 14(9): 813-822.

Ramanathan, G, Yin, F, Speck, M, Tseng, CH, Brook, JR, Silverman, F, Urch, B, Brook, RD and
Araujo, JA (2016). Effects of urban fine particulate matter and ozone on HDL
functionality. Particle and Fibre Toxicology 13(1): 26.

Sacks, JD, Ito, K, Wilson, WE and Neas, LM (2012). Impact of covariate models on the

assessment of the air pollution-mortality association in a single- and multipollutant
context. American Journal of Epidemiology 176(7): 622-634.

Sanders, NJ, Barreca, AI and Neidell, MJ (2020a). Estimating causal effects of particulate matter
regulation on mortality. Epidemiology (Cambridge, Mass.) 31(2): 160.

Sanders, NJ, Barreca, AI and Neidell, MJ (2020b). Estimating Causal Effects of Particulate
Matter Regulation on Mortality. Epidemiology 31(2): 160-167.

Schwartz, J, Austin, E, Bind, MA, Zanobetti, A and Koutrakis, P (2015). Estimating causal
associations of fine particles with daily deaths in Boston. American Journal of
Epidemiology 182(7): 644-650.

Schwartz, J, Bind, MA and Koutrakis, P (2017). Estimating causal effects of local air pollution
on daily deaths: Effect of low levels. Environmental Health Perspectives 125(1): 23-29.

Schwartz, J, Fong, K and Zanobetti, A (2018a). A National Multicity Analysis of the Causal
Effect of Local Pollution, [Formula: see text], and [Formula: see text] on Mortality.
Environmental Health Perspectives 126(8): 87004.

Schwartz, J, Wei, Y, Yitshak-Sade, M, Di, Q, Dominici, F and Zanobetti, A (2021). A national
difference in differences analysis of the effect of PM2.5 on annual death rates.
Environmental Research 194: 110649.

Schwartz, JD, Wang, Y, Kloog, I, Yitshak-Sade, M, Dominici, F and Zanobetti, A (2018b).

Estimating the Effects of PM2.5 on Life Expectancy Using Causal Modeling Methods.
Environmental Health Perspectives 126(12): 127002.

Sheppard, EA. (2022). Letter from Elizabeth A. (Lianne) Sheppard, Chair, Clean Air Scientific
Advisory Committee, to Administrator Michael S. Regan. Re: CASAC Review of the

3-238

-------
EPA's Policy Assessment for the Review of the National Ambient Air Quality Standards
for Particulate Matter (External Review Draft - October 2021). March 18, 2022. EPA-
CASAC-22-002. Office of the Administrator, Science Advisory Board U.S. EPA HQ,
Washington DC. Available at:

https://casac.epa.gov/ords/sab/f?p=105:18:10792850355838:::RP.18:P18 ID:2607#report

Shi, L, Zanobetti, A, Kloog, I, Coull, BA, Koutrakis, P, Melly, SJ and Schwartz, JD (2016).
Low-concentration PM2.5 and mortality: estimating acute and chronic effects in a
population-based study. Environmental Health Perspectives 124(1): 46-52.

Shin, HH, Gogna, P, Maquiling, A, Parajuli, RP, Haque, L and Burr, B (2021). Comparison of
hospitalization and mortality associated with short-term exposure to ambient ozone and
PM2.5 in Canada. Chemosphere 265: 128683.

Shin, S, Burnett, RT, Kwong, JC, Hystad, P, van Donkelaar, A, Brook, JR, Goldberg, MS, Tu, K,
Copes, R, Martin, R, Lia, Y, Kopp, A and Chen, H (2019). Ambient Air Pollution and the
Risk of Atrial Fibrillation and Stroke: A Population-Based Cohort Study. Environmental
Health Perspectives 127(8): 87009.

Sivagangabalan, G, Spears, D, Masse, S, Urch, B, Brook, RD, Silverman, F, Gold, DR, Lukic,
KZ, Speck, M, Kusha, M, Farid, T, Poku, K, Shi, E, Floras, J and Nanthakumar, K
(2011). The effect of air pollution on spatial dispersion of myocardial repolarization in
healthy human volunteers. Journal of the American College of Cardiology 57(2): 198-
206.

Stieb, DM, Szyszkowicz, M, Rowe, BH and Leech, JA (2009). Air pollution and emergency

department visits for cardiac and respiratory conditions: A multi-city time-series analysis.
Environmental Health: A Global Access Science Source 8(25): 25.

Sun, S, Stewart, JD, Eliot, MN, Yanosky, JD, Liao, D, Tinker, LF, Eaton, CB, Whitsel, EA and
Wellenius, GA (2019). Short-term exposure to air pollution and incidence of stroke in the
Women's Health Initiative. Environment International 132: 105065.

Szyszkowicz, M (2009). Air pollution and ED visits for chest pain. American Journal of
Emergency Medicine 27(2): 165-168.

Thurston, GD, Ahn, J, Cromar, KR, Shao, Y, Reynolds, HR, Jerrett, M, Lim, CC, Shanley, R,
Park, Y and Hayes, RB (2016). Ambient particulate matter air pollution exposure and
mortality in the NIH-AARP Diet and Health Cohort. Environmental Health Perspectives
124(4): 484-490.

Thurston, GD, Ito, K, Lall, R, Burnett, RT, Turner, MC, Krewski, D, Shi, Y, Jerrett, M, Gapstur,
SM, Diver, WR and Pope, CA, III (2013). NPACT Study 4. Mortality and long-term
exposure to PM2.5 and its components in the American Cancer Society's Cancer
Prevention Study II Cohort. Research Report 177. Health Effects Institute. Boston, MA.

3-239

-------
Thurston, GD, Kipen, H, Annesi-Maesano, I, Balmes, J, Brook, RD, Cromar, K, De Matteis, S,
Forastiere, F, Forsberg, B, Frampton, MW, Grigg, J, Heederik, D, Kelly, FJ, Kuenzli, N,
Laumbach, R, Peters, A, Rajagopalan, ST, Rich, D, Ritz, B, Samet, JM, Sandstrom, T,
Sigsgaard, T, Sunyer, J and Brunekreef, B (2017). A joint ERS/ATS policy statement:
what constitutes an adverse health effect of air pollution? An analytical framework.
European Respiratory Journal 49(1): 1600419.

Tong, H, Rappold, AG, Caughey, M, Hinderliter, AL, Bassett, M, Montilla, T, Case, MW,

Berntsen, J, Bromberg, PA, Cascio, WE, Diaz-Sanchez, D, Devlin, RB and Samet, JM
(2015). Dietary supplementation with olive oil or fish oil and vascular effects of
concentrated ambient particulate matter exposure in human volunteers. Environmental
Health Perspectives 123(11): 1173-1179.

Turner, MC, Jerrett, M, Pope, A, III, Krewski, D, Gapstur, SM, Diver, WR, Beckerman, BS,

Marshall, JD, Su, J, Crouse, DL and Burnett, RT (2016). Long-term ozone exposure and
mortality in a large prospective study. American Journal of Respiratory and Critical Care
Medicine 193(10): 1134-1142.

U.S. EPA (2001). Risk Assessment Guidance for Superfund: Volume III - Part A, Process for
Conducting Probabilistic Risk Assessment. Office of Emergency and Remedial
Response. Washington, D.C. U.S. EPA. EPA 540-R-02-002. Available at:
https://www.epa.gov/sites/production/files/2015-Q9/documents/rags3adt complete.pdf.

U.S. EPA (2004). An Examination of EPA Risk Assessment Principles and Practices. Office of
Research and Development. Washington, D.C. U.S. EPA. EPA/100/B-04/001. Available
at: https://semspub.epa.gov/work/10/500006305.pdf.

https://nepis.epa. gov/Exe/ZvPURL.cgi?Dockev=P 100AUMY.txt.

U.S. EPA (2013). Regulatory Impact Analysis for the Final Revisions to the National Ambient
Air Quality Standards for Particulate Matter. U.S. Environmental Protection Agency,
Office of Air Quality Planning and Standards. Research Triangle Park, NC. U.S. EPA.
EPA-452/R-12-005. February 2013. Available at:

3-240

-------
https://www.epa.gOv/sites/default/files/2020-07/documents/naaqs-pm ria final 2012-
12.pdf.

U.S. EPA (2015a). Preamble to the integrated science assessments. U.S. Environmental
Protection Agency, Office of Research and Development, National Center for
Environmental Assessment, RTP Division. Research Triangle Park, NC. U.S. EPA.
EPA/600/R-15/067. November 2015. Available at:
https://cfpub.epa. gov/ncea/isa/recordisplav.cfm?deid=310244.

U.S. EPA (2015b). Regulatory Impact Analysis of the Final Revisions to the National Ambient
Air Quality Standards for Ground-Level Ozone. U.S. Environmental Protection Agency,
Office of Air Quality Planning and Standards. Research Triangle Park, NC. U.S. EPA.
EPA-452/R-15-007. September 2015. Available at:

https://www.epa.gov/sites/default/files/2016-02/documents/20151001ria.pdf.

U.S. EPA (2016). Integrated review plan for the national ambient air quality standards for

U.S. EPA (2020). Policy Assessment for the Review of the National Ambient Air Quality

Standards for Particulate Matter. Office of Air Quality Planning and Standards, Health
and Environmental Impacts Division. Research Triangle Park, NC. U.S. EPA. EPA-
452/R-20-002. January 2020. Available at: https://www.epa.gov/sites/default/files/202Q-
01/documents/final policy assessment for the review of the pm naaqs 01-2020.pdf.

U.S. EPA (2021a). Supplement to the 2019 Integrated Science Assessment for Particulate Matter
(External Review Draft). U.S. Environmental Protection Agency, Office of Research and
Development, Center for Public Health and Environmental Assessment. Research
Triangle Park, NC. U.S. EPA. EPA/600/R-21/198. December 2019. Available at:
https://www.epa.gov/naaqs/particulate-matter-pm-standards-integrated-science-
assessments-current-review.

U.S. EPA (2021b). Technical Support Document (TSD) for the Final Revised Cross-State Air
Pollution Rule Update for the 2008 Ozone Season NAAQS Estimating PM2.5- and
Ozone-Attributable Health Benefits. U.S. Environmental Protection Agency, Office of
Air Quality Planning and Standards. Research Triangle Park, NC. U.S. EPA. EPA-HQ-
OAR-2020-0272. March 2021. Available at:

https://www.epa.gov/sites/default/files/2021-Q3/documents/estimating pm2.5-
and ozone-attributable health benefits tsd march 2021.pdf.

3-241

-------
U.S. EPA (2022). Supplement to the 2019 Integrated Science Assessment for Particulate Matter
(Final Report). U.S. Environmental Protection Agency, Office of Research and
Development, Center for Public Health and Environmental Assessment. Research
Triangle Park, NC. U.S. EPA. EPA/600/R 22/028. May 2022. Available at:
https://www.epa.gov/naaqs/particulate-matter-pm-standards-integrated-science-
assessments-current-review.

Urch, B, Speck, M, Corey, P, Wasserstein, D, Manno, M, Lukic, KZ, Brook, JR, Liu, L, Coull,
B, Schwartz, J, Gold, DR and Silverman, F (2010). Concentrated ambient fine particles
and not ozone induce a systemic interleukin-6 response in humans. Inhalation Toxicology
22(3): 210-218.

USDA (2018). Rural America at a Glance, 2018 Edition. U.S. Department of Agriculture,

Economic Research Serivice. Washington, DC. USDA. Economic Information Bulletin
200. November 2018. Available at:

https://www.ers.usda.gov/webdocs/publications/90556/eib-200.pdf.

Vieira, JL, Guimaraes, GV, de Andre, PA, Cruz, FD, Nascimento Saldiva, PH and Bocchi, EA
(2016a). Respiratory filter reduces the cardiovascular effects associated with diesel
exhaust exposure a randomized, prospective, double-blind, controlled study of heart
failure: the FILTER-HF trial. JACC: Heart Failure 4(1): 55-64.

Vieira, JL, Guimaraes, GV, de Andre, PA, Nascimento Saldiva, PH and Bocchi, EA (2016b).
Effects of reducing exposure to air pollution on submaximal cardiopulmonary test in
patients with heart failure: Analysis of the randomized, double-blind and controlled
FILTER-HF trial. International Journal of Cardiology 215: 92-97.

Villeneuve, PJ, Weichenthal, SA, Crouse, D, Miller, AB, To, T, Martin, RV, van Donkelaar, A,
Wall, C and Burnett, RT (2015). Long-term exposure to fine particulate matter air
pollution and mortality among Canadian women. Epidemiology 26(4): 536-545.

Wang, B, Eum, KD, Kazemiparkouhi, F, Li, C, Manjourides, J, Pavlu, V and 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.
Environmental Health: A Global Access Science Source 19(1): 20.

Wang, Y, Shi, L, Lee, M, Liu, P, Di, Q, Zanobetti, A and Schwartz, JD (2017). Long-term
exposure to PM2.5 and mortality among older adults in the Southeastern US.
Epidemiology 28(2): 207-214.

Ward-Caviness, CK, Weaver, AM, Buranosky, M, Pfaff, ER, Neas, LM, Devlin, RB, Schwartz,
J, Di, Q, Cascio, WE and Diaz-Sanchez, D (2020). Associations between long-term fine
particulate matter exposure and mortality in heart failure patients. Journal of the
American Heart Association 9(6): e012517.

Wei, Y, Wang, Y, Di, Q, Choirat, C, Wang, Y, Koutrakis, P, Zanobetti, A, Dominici, F and
Schwartz, JD (2019). Short term exposure to fine particulate matter and hospital

3-242

-------
admission risks and costs in the Medicare population: time stratified, case crossover
study. BMJ (Clinical Research Edition) 367: 16258.

Weichenthal, S, Crouse, DL, Pinault, L, Godri-Pollitt, K, Lavigne, E, Evans, G, van Donkelaar,
A, Martin, RV and Burnett, RT (2016a). Oxidative burden of fine particulate air pollution
and risk of cause-specific mortality in the Canadian Census Health and Environment
Cohort (CanCHEC). Environmental Research 146: 92-99.

Weichenthal, S, Hatzopoulou, M and Goldberg, MS (2014). Exposure to traffic-related air

pollution during physical activity and acute changes in blood pressure, autonomic and
micro-vascular function in women: A cross-over study. Particle and Fibre Toxicology
11(1): 70.

Weichenthal, S, Lavigne, E, Evans, G, Pollitt, K and Burnett, RT (2016b). Ambient PM2.5 and
risk of emergency room visits for myocardial infarction: Impact of regional PM2.5
oxidative potential: A case-crossover study. Environmental Health 15: 46.

Weichenthal, S, Lavigne, E, Evans, GJ, Godri Pollitt, KJ and Burnett, RT (2016c). PM2.5 and
emergency room visits for respiratory illness: effect modification by oxidative potential.
American Journal of Respiratory and Critical Care Medicine 194(5): 577-586.

WHO (2008). Uncertainty and Data Quality in Exposure Assessment. The Internaltional
Programme on Chemical Safety. Geneva. WHO.

https://www.who.int/ipcs/publications/methods/harmonization/exposure assessment.pdf.

Willis, A, Jerrett, M, Burnett, RT and Krewski, D (2003). The association between sulfate air
pollution and mortality at the county scale: an exploration of the impact of scale on a
long-term exposure study. Journal of Toxicology and Environmental Health, Part A:
Current Issues 66(16-19): 1605-1624.

Wu, X, Braun, D, Kioumourtzoglou, MA, Choirat, C, Di, Q and Dominici, F (2019). Causal

inference in the context of an error prone exposure: Air pollution and mortality. Annals
of Applied Statistics 13(1): 520-547.

Wu, X, Braun, D, Schwartz, J, Kioumourtzoglou, MA and Dominici, F (2020). Evaluating the
impact of long-term exposure to fine particulate matter on mortality among the elderly.
Science Advances 6(29): eaba5692.

Wyatt, LH, Devlin, RB, Rappold, AG, Case, MW and Diaz-Sanchez, D (2020a). Low levels of
fine particulate matter increase vascular damage and reduce pulmonary function in young
healthy adults. Particle and fibre toxicology 17(1): 1-12.

Wyatt, LH, Peterson, GCL, Wade, TJ, Neas, LM and Rappold, AG (2020b). The contribution of
improved air quality to reduced cardiovascular mortality: Declines in socioeconomic
differences over time. Environment International 136: 105430.

Wyatt, LH, Xi, YZ, Kshirsagar, A, Di, Q, Ward-Caviness, C, Wade, TJ, Cascio, WE and
Rappold, AG (2020c). Association of short-term exposure to ambient PM2.5 with

3-243

-------
hospital admissions and 30-day readmissions in end-stage renal disease patients:
population-based retrospective cohort study. British Medical Journal Open 10(12):
e041177.

Yazdi, D, Wang, Y, Di, Q, Zanobetti, A and Schwartz, J (2019). Long-term exposure to PM2.5
and ozone and hospital admissions of Medicare participants in the Southeast USA.
Environment International 130: 104879.

Yazdi, DM, Wang, Y, Di, Q, Wei, YG, Requia, WJ, Shi, LH, Sabath, MB, Dominici, F, Coull,
BA, Evans, JS, Koutrakis, P and Schwartz, JD (2021). Long-term association of air
pollution and hospital admissions among Medicare participants using a doubly robust
additive model. Circulation 143(16): 1584-1596.

Yitshak-Sade, M, Kloog, I, Zanobetti, A and Schwartz, JD (2019). Estimating the causal effect
of annual PM2.5 exposure on mortality rates in the Northeastern and mid-Atlantic states.
Environmental Epidemiology 3(4): e052.

Yorifuji, T, Kashima, S and Doi, H (2016). Fine-particulate air pollution from diesel emission
control and mortality rates in Tokyo: a quasi-experimental study. Epidemiology 27(6):
769-778.

Zanobetti, A, Dominici, F, Wang, Y and 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. Environmental Health: A Global Access Science
Source 13(1): 38.

Zanobetti, A, Franklin, M, Koutrakis, P and Schwartz, J (2009). Fine particulate air pollution and
its components in association with cause-specific emergency admissions. Environmental
Health: A Global Access Science Source 8: 58.

Zanobetti, A and Schwartz, J (2009). The effect of fine and coarse particulate air pollution on
mortality: A national analysis. Environmental Health Perspectives 117(6): 1-40.

Zeger, S, Dominici, F, McDermott, A and Samet, J (2008). Mortality in the Medicare population
and chronic exposure to fine particulate air pollution in urban centers (2000-2005).
Environmental Health Perspectives 116(12): 1614-1619.

Zhang, Z, Wang, J, Kwong, JC, Burnett, RT, van Donkelaar, A, Hystad, P, Martin, RV, Bai, L,
McLaughlin, J and Chen, H (2021). Long-term exposure to air pollution and mortality in
a prospective cohort: The Ontario Health Study. Environment International 154: 106570.

Zigler, CM, Choirat, C and Dominici, F (2018). Impact of National Ambient Air Quality

Standards nonattainment designations on particulate pollution and health. Epidemiology
(Cambridge, Mass.) 29(2): 165.

3-244

-------
4 RECONSIDERATION OF THE PRIMARY STANDARD

FOR PMio

This chapter presents and evaluates the policy implications of the scientific and technical
information pertaining to reconsideration of the 2020 final decision on the primary PMio
standard. In so doing, the chapter presents key aspects of the health effects evidence of PM10-2.5,
as documented in the 2019 ISA, with support from the prior ISA and AQCDs, and associated
public health implications. This information provides the basis for our evaluation of the scientific
information regarding health effects of PMio in ambient air and the potential for effects to occur
under air quality conditions associated with the existing standard, as well as the associated
implications for public health. Our evaluation is framed around key policy-relevant questions
derived from the IRP (U.S. EPA, 2016, section 2.1) for the review completed in 2020, and the
scientific conclusions regarding the relationship between short- and long-term PM10-2.5 exposure
and health effects detailed in the 2019 ISA, while also taking into account conclusions reached in
previous reviews. In this way, we identify key policy-relevant issues and summary conclusions
regarding the public health protection provided by the current standard as the Administrator
reconsiders the final 2020 decision on the primary PMio standard.

As described in Chapter 1, the scope of the updated scientific evaluation of the health
effects evidence for PMio is based on those health effects categories where the 2019 ISA
concluded a causal relationship exists. Therefore, the ISA Supplement does not include an
evaluation of additional studies for PM10-2.5 and the 2019 ISA continues to serve as the scientific
foundation for assessing the adequacy of the primary PMio standard in this reconsideration of the
2020 final decision (U.S. EPA, 2019, section 1.7; U.S. EPA, 2022). As such, this chapter draws
heavily from the 2020 PA in identifying and summarizing key issues related to this
reconsideration of the primary PMio standard.

Within this chapter, background information on the current standard is summarized in
section 4.1. The general approach for evaluating the available information in this
reconsideration, including policy-relevant questions identified to frame our policy evaluation, is
summarized in section 4.2. Key aspects of the available health effects evidence presented in the
2019 ISA and considered in the 2020 PA are addressed in section 4.3. Section 4.5 summarizes
the CASAC's advice and public comments, while section 4.5 summarizes the key evidence-
based considerations identified in our evaluation and presents associated conclusions on the
adequacy of the current standard. Key remaining uncertainties and areas for future research are
identified in section 4.6.

4-1

-------
4.1 BACKGROUND ON THE CURRENT STANDARD

With the 2020 final decision on the PM NAAQS, the EPA retained the existing 24-hour
primary PMio standard, with its level of 150 |ig/m3 and its one-expected-exceedance form on
average over three years, to continue to provide public health protection against short-term
exposures to PM10-2.5 (85 FR 82725, December 18, 2020). This decision was based on the
scientific information available at that time, as well as the Administrator's judgments regarding
the health effects evidence and the appropriate degree of public health protection for the existing
standard.

The health effects evidence assessed in the 2019 ISA included an expanded body of
scientific evidence linking short-term PM10-2.5 to health outcomes such as premature death and
hospital visits (U.S. EPA, 2009, U.S. EPA, 2019). This evidence base assessed the causal nature
of relationships between short-term exposure to PM10-2.5 and a broad range of health effects (U.S.
EPA, 2020, section 1.4.2). These effects associated with short-term exposure ranged from
hospital admissions and emergency department visits for cardiovascular effects (documented in
epidemiologic studies that reported PM10-2.5 associations with cardiovascular hospital admissions
and emergency department visits in study locations with mean 24-hour average PM10-2.5
concentrations ranging from 7.4 to 13 |ig/m3) and respiratory effects (documented in
epidemiologic studies that reported PM10-2.5 associations with respiratory hospital admissions and
emergency department visits in study locations with mean 24-hour average concentrations
ranging from 5.6 to 16.2 |ig/m3) to mortality (documented in epidemiologic studies that reported
PM10-2.5 associations with mortality in study areas with mean 24-hour average concentrations
ranging from 6.1 |ig/m3 to 16.4 |ig/m3). In addition to the epidemiologic studies, the evidence
base included a small number of controlled human exposure studies and animal toxicologic
studies that provided insight into the biological plausibility of these effects. Collectively, the
epidemiologic studies, controlled human exposure, and animal toxicological studies, with their
inherent uncertainties, contributed to the causality determinations of "suggestive of, but not
sufficient to infer, a causal relationship" between short-term exposures to PM10-2.5 and
cardiovascular effects, respiratory effects, and mortality (U.S. EPA, 2009, U.S. EPA, 2019,
section 1.4.2).

Building on the evidence considered in the 2012 review, the primary focus in the 2020
review was on multi-city and single-city epidemiologic studies that evaluated associations
between short-term PM10-2.5 and mortality, cardiovascular effects (hospital admissions and
emergency department visits), and respiratory effects. Despite differences in the approaches used
to estimate ambient PM10-2.5 concentrations, the majority of the studies reported positive, though
often not statistically significant, associations with short-term PM10-2.5 exposures. Most PM10-2.5
effect estimates remained positive in copollutant models that included either gaseous pollutants

4-2

-------
or other particulate matter size fractions (e.g., PM2.5). In U.S. study locations likely to have met
the PM10 standard during the study period, a few studies reported positive associations between
PM10-2.5 and mortality that were statistically significant and remained so in copollutant models
(U.S. EPA, 2009, U.S. EPA, 2019).

In addition to the epidemiologic studies, there were a small number of controlled human
exposure studies assessed in the 2019 ISA that reported alterations in heart rate variability or
increased pulmonary inflammation following short-term exposure to PM10-2.5, providing some
support for the associations in the epidemiologic studies. Toxicological studies that examined the
effects of PM10-2.5 used intratracheal instillation as opposed to inhalation. Therefore, these studies
provided limited evidence for the biological plausibility of PMio-2 5-induced effects (U.S. EPA,
2009, U.S. EPA, 2019).

Although the scientific evidence available in the 2019 ISA expanded the understanding of
health effects associated with PM10-2.5 exposures, a number of important uncertainties remained.
These uncertainties, and their implications for interpreting the scientific evidence, include the
following:

• The potential for confounding by copollutants, notably PM2.5, was addressed with
copollutant models in a relatively small number of PM10-2.5 epidemiologic studies (U.S.
EPA, 2009, U.S. EPA, 2019). This was particularly important given the relatively small
body of experimental evidence (i.e., controlled human exposure and animal toxicological
studies) available to support the independent effect of PM10-2.5 on human health. This
increases the uncertainty regarding the extent to which PM10-2.5 itself, rather than one or
more cooccurring pollutants, is responsible for the mortality and morbidity effects
reported in epidemiologic studies.

• There was greater spatial variability in PM10-2.5 concentrations than PM2.5 concentrations,
resulting in increased exposure error for PM10-2.5 (U.S. EPA, 2009, U.S. EPA, 2019).
Available measurements did not provide sufficient information to adequately characterize
the spatial distribution of PM10-2.5 concentrations (U.S. EPA, 2009, U.S. EPA, 2019). The
limitations in estimates of ambient PM10-2.5 concentrations "would tend to increase
uncertainty and make it more difficult to detect effects of PM10-2.5 in epidemiologic
studies" (U.S. EPA, 2009, U.S. EPA, 2019).

• The distributions of PM10-2.5 concentrations over which reported health outcomes occur
remain highly uncertain. Only a relatively small number of PM10-2.5 monitoring sites were
operating at the time of the 2012 review and such sites had only been in operation for a
relatively short period of time, limiting the spatial and temporal coverage for routine
measurement of PM10-2.5 concentrations. Given these limitations in routine monitoring,
epidemiologic studies employed a number of different approaches for estimating PM10-2.5
concentrations. Given the relatively small number of PM10-2.5 monitoring sites, the
relatively large spatial variability in ambient PM10-2.5 concentrations, the use of different
approaches to estimating ambient PM10-2.5 concentrations across epidemiologic studies,
and the limitations inherent in such estimates, the distributions of PM10-2.5 concentrations

4-3

-------
over which reported health outcomes occur remain highly uncertain (U.S. EPA, 2009,
U.S. EPA, 2019).

• There was relatively little information on the chemical and biological composition of
PMio-2.5 and the effects associated with the various components (U.S. EPA, 2019).
Without more information on the chemical speciation of PM10-2.5, the apparent variability
in associations with health effects across locations was difficult to characterize (U.S.
EPA, 2009, U.S. EPA, 2019).

Consistent with the general approach routinely employed in NAAQS reviews, the initial
consideration in the 2020 review of the primary PM10 standard was with regard to the adequacy
of protection provided by the then-existing standard. Key aspects of that consideration are
summarized in section 4.1.1 below.

4.1.1 Considerations Regarding the Adequacy of the Existing Standards in the 2020
Review

In the 2020 final decision, the EPA retained the existing 24-hour primary PM10 standard
with its level of 150 |ig/m3 and its one-expected-exceedance form on average over three years to
continue to provide public health protection against exposures to PM10-2.5 (85 FR 82727,
December 18, 2020). In reaching his decision, the Administrator specifically noted that, while
the health effects evidence was somewhat expanded since the prior reviews, the overall
conclusions in the 2019 ISA, including uncertainties and limitations, were generally consistent
with what was considered in the 2012 review (85 FR 82725, December 18, 2020). In addition,
the Administrator recognized that there were still a number of uncertainties and limitations
associated with the available evidence.

With regard to the evidence on PM10-2.5-related health effects, the Administrator noted
that epidemiologic studies continued to report positive associations with mortality and morbidity
in cities across North America, Europe, and Asia, where PM10-2.5 sources and composition were
expected to vary widely. While significant uncertainties remained in the 2020 review, the
Administrator recognized that this expanded body of evidence had broadened the range of effects
that have been linked with PM10-2.5 exposures. The studies evaluated in the 2019 ISA expanded
the scientific foundation presented in the 2009 ISA and led to revised causality determinations
(and new determinations) for long-term PM10-2.5 exposures and mortality, cardiovascular effects,
metabolic effects, nervous system effects, and cancer (85 FR 82726, December 18, 2020).
Drawing from his consideration of this evidence, the Administrator concluded that the scientific
information available since the time of the last review supported a decision to maintain a primary
PM10 standard to provide public health protection against PM10-2.5 exposures, regardless of
location, source of origin, or particle composition (85 FR 82726, December 18, 2020).

4-4

-------
With regard to uncertainties in the available evidence, the Administrator first noted that a
number of limitations were identified in the 2012 review related to: (1) estimates of ambient
PMio-2.5 concentrations used in epidemiologic studies; (2) limited evaluation of copollutant
models to address the potential for confounding; and (3) limited experimental studies supporting
biological plausibility for PMio-2 5-related effects. Despite the expanded body of evidence for
PM10-2.5 exposures and health effects, the Administrator recognized that uncertainties in the 2020
review continued to include those associated with the exposure estimates used in epidemiologic
studies, the independence of the PM10-2.5 health effect associations, and the biologically plausible
pathways for PM10-2.5 health effects (85 FR 82726, December 18, 2020). These uncertainties
contributed to the 2019 ISA determinations that the evidence is "suggestive of, but not sufficient
to infer" causal relationships (85 FR 82726, December 18, 2020).

Further, consistent with the approach in reaching the 2012 decision, the approach for the
2020 PM NAAQS review did not include quantitative assessments of estimated exposures or
risks allowed by the existing standard or potential alternative standards. Further, the available
evidence in the 2019 ISA did not provide support for evaluating air quality distributions in
locations of individual epidemiologic studies as was done in the 2012 review (78 FR 3176,
January 15, 2013). The substantial uncertainty in such analyses, if conducted based on the
available PM10-2.5 health studies, would have been of limited utility for informing conclusions on
the primary PM10 standard.

In the 2020 decision, for all of the reasons discussed above and recognizing the CASAC
conclusion that the evidence provided support for retaining the current standard, the
Administrator concluded that it was appropriate to retain the existing primary PM10 standard,
without revision. His decision was consistent with the CASAC advice related to the primary
PM10 standard. Specifically, the CASAC agreed with the 2020 PA conclusions that, while these
effects are important, the "evidence does not call into question the adequacy of the public health
protection afforded by the current primary PM10 standard" and "supports consideration of
retaining the current standard in this review" (Cox, 2019a, p. 3 of letter). Thus, the Administrator
concluded that the primary PM10 standard (in all of its elements) was requisite to protect public
health with an adequate margin of safety against effects that have been associated with PM10-2.5.
In light of this conclusion, the EPA retained the existing PM10 standard.

4.2 GENERAL APPROACH AND KEY ISSUES IN THIS
RECONSIDERATION OF THE 2020 FINAL DECISION

This reconsideration of the 2020 final decision on the primary PM10 standard is most
fundamentally based on using the Agency's assessment of the scientific evidence and
quantitative information, if available, to inform the Administrator's judgments regarding a

4-5

-------
primary standard that is requisite to protect public health with an adequate margin of safety. The
approach for this reconsideration builds on the substantial assessments and evaluations
performed over previous reviews (U.S. EPA, 2011, U.S. EPA, 2020). As noted above, the ISA
Supplement does not include an evaluation of studies for PM10-2.5 and the 2019 ISA continues to
serve as the scientific foundation for this reconsideration. Given that there is no new evidence for
PMio-2 5-related health effects assessed in the ISA Supplement that would inform quantitative
assessments or conclusions on the current primary PM10 standard since the completion of the
2020 review, this PA draws from the evaluation of the health effects evidence for PM10-2.5-
related effects in the 2019 ISA and considerations of such effects in the 2020 PA (U.S. EPA,
2020).

The evaluations in this PA of the health effects evidence assessed in the 2019 ISA are
intended to inform the Administrator's public health policy judgments and conclusions as a part
of this reconsideration of the 2020 final decision, including his decision as to whether to retain or
revise the primary PM10 standard. The PA evaluations consider the potential implications of
various aspects of the scientific evidence and the associated uncertainties and limitations. In so
doing, the approach for this PA involves evaluating the available scientific and technical
information to address a series of key policy-relevant questions using evidence-based
considerations. Consideration of the full set of evidence in this reconsideration will inform the
answer to the following initial overarching question:

• Does the scientific evidence support or call into question the adequacy of the

protection afforded by the current 24-hour primary PM10 standard against health

effects associated with exposures to PM10-2.5?

In reflecting on this question, we consider the body of scientific evidence, assessed in the
2019 ISA, including whether it supports or calls into question the scientific conclusions reached
in previous reviews regarding health effects related to exposure to PM10-2.5 in ambient air.
Information available in the 2019 ISA that may be informative to public health judgments
regarding significance or adversity of key effects will also be considered. Further, in considering
this question with regard to the primary PM10 standard, as in all NAAQS reviews, we give
particular attention to exposures and health risks to at-risk populations (including at-risk
lifestages). Evaluation of the scientific information with regard to this consideration of the
current standard will focus on key policy-relevant issues by addressing a series of questions
including the extent to which the available scientific evidence supports retaining or altering the
conclusions in the prior reviews regarding health effects attributed to PM10-2.5 exposures.
Furthermore, this PA will examine whether the previously identified uncertainties have been
reduced and if new uncertainties have been identified.

4-6

-------
The general approach to reaching conclusions on the current primary PMio standard is
summarized in Figure 4-1:

F urther evaluate the scientific evidence and exposure/risk assessments to
inform Identification of potential alternatives

Figure 4-1. Overview of general approach for the reconsideration of the 2020 final decision
on the primary PMio standard.

4-7

-------
The Agency's approach to reviewing the primary standards is consistent with the
requirements of the provisions of the CAA related to the review of the NAAQS and with how the
EPA and the courts have historically interpreted the CAA. As discussed in section 1.1 above,
these provisions require the Administrator to establish primary standards that, in the
Administrator's judgment, are requisite (i.e., neither more nor less stringent than necessary) to
protect public health with an adequate margin of safety. Consistent with the Agency's approach
across all NAAQS reviews, the approach of this PA to informing these judgments is based on a
recognition that the available health effects evidence generally reflects continuums that include
ambient air exposures for which scientists generally agree health effects are likely to occur
through lower levels at which the likelihood and magnitude of response become increasingly
uncertain. The CAA does not require the Administrator to establish a primary standard at a zero-
risk level or at background concentration levels, but rather at a level that reduces risk sufficiently
so as to protect public health, including the health of sensitive groups,1 with an adequate margin
of safety.

The decisions on the adequacy of the current primary PMio standard and on any
alternative standards considered in a review are largely public health policy judgments made by
the Administrator. The four basic elements of the NAAQS (i.e., indicator, averaging time, form,
and level) are generally considered collectively in evaluating the health protection afforded by
the current standard, and by any alternatives considered. The Administrator's final decisions in a
review draw upon the scientific evidence for health effects, quantitative analyses of population
exposures and/or health risks, as available, and judgments about how to consider the
uncertainties and limitations that are inherent in the scientific evidence and quantitative analyses.

4.3 HEALTH EFFECTS EVIDENCE

This section draws from the EPA's synthesis and assessment of the scientific evidence
presented in the 2019 ISA (U.S. EPA, 2019) to consider the following policy-relevant questions:

• To what extent does the available scientific evidence strengthen, or otherwise alter, our
conclusions from previous reviews regarding health effects attributable to long- or
short-term PM10-2.5 exposures? Have previously identified uncertainties been reduced?
What important uncertainties remain and have new uncertainties been identified?

1 More than one population group may be identified as sensitive or at-risk in a NAAQS review. Decisions on
NAAQS reflect consideration of the degree to which protection is provided for these sensitive population groups.
To the extent that any particular population group is not among the identified sensitive groups, a decision that
provides protection for the sensitive groups would be expected to also provide protection for other population
groups.

4-8

-------
Answers to these questions will inform our response to the overarching question on the adequacy
of the current primary PMio standard, posed at the beginning of this chapter. In section 4.3.1
below, we consider the nature of the effects attributable to long-term and short-term PM10-2.5
exposures.

4.3.1 Nature of Effects

As noted above, for the health effect categories and exposure duration combinations
evaluated, the 2019 ISA concludes that the evidence supports causality determinations for
PM10-2.5 that are "suggestive of, but not sufficient to infer, a causal relationship." These health
effect categories, along with their corresponding causality determinations from the 2009 ISA, are
highlighted below in Table 4-1 (adapted from U.S. EPA, 2019, Table 1-4).

Table 4-1. Key Causality Determinations for PM10-2.5 Exposures

Health Outcome

Exposure
Duration

2009 PM ISA

2019 PM ISA

Mortality

Long-term

Inadequate

Short-term

Suggestive of, but not sufficient to infer

Cardiovascular

Long-term

Inadequate

effects

Short-term

Suggestive of, but not sufficient to infer

Respiratory effects

Short-term

Suggestive of, but not sufficient to infer

Suggestive of, but not
sufficient to infer

Cancer

Long-term

Inadequate

Nervous System
effects

Long-term

—

Metabolic effects

Long-term

—

While the evidence supporting the causal nature of relationships between exposure to
PM10-2.5 has been strengthened for some of the health effect categories listed in Table 4-1 since
the 2009 ISA, the 2019 ISA concludes that overall "the uncertainties in the evidence identified in
the 2009 PM ISA have, to date, still not been addressed" (U.S. EPA, 2019, section 1.4.2, p. 1-
41). Specifically, epidemiologic studies available in the 2012 review relied on various methods
to estimate PM10-2.5 concentrations, and these methods had not been systematically compared to
evaluate spatial and temporal correlations in PM10-2.5 concentrations. Methods included (1)
calculating the difference between PM10 and PM2.5 concentrations at co-located monitors, (2)
calculating the difference between county-wide averages of monitored PM10- and PIVh.s-based on
monitors that are not necessarily co-located, and (3) direct measurement of PM10-2.5 using a

4-9

-------
dichotomous sampler (U.S. EPA, 2019, section 1.4.2). As described in the 2019 ISA, there
continues to be variability across epidemiologic studies in the approaches used to estimate PMio-
2.5 concentrations. Additionally, some studies estimate long-term PM10-2.5 exposures as the
difference between PM10 and PM2.5 concentrations based on information from spatiotemporal or
land use regression (LUR) models, in addition to monitors. The various methods used to estimate
PM10-2.5 concentrations have not been systematically evaluated (U.S. EPA, 2019, section
3.3.1.1), contributing to uncertainty regarding the spatial and temporal correlations in PM10-2.5
concentrations across methods and in the PM10-2.5 exposure estimates used in epidemiologic
studies (U.S. EPA, 2019, section 2.5.1.2.3). Given the greater spatial and temporal variability of
PM10-2.5 and the lower number of PM10-2.5 monitoring sites, compared to PM2.5, this uncertainty is
particularly important for the coarse size fraction.

Beyond the uncertainty associated with PM10-2.5 exposure estimates in epidemiologic
studies, the limited information on the potential for confounding by copollutants and the limited
support available for the biological plausibility of health effects following PM10-2.5 exposures
also continue to contribute to uncertainty in the PM10-2.5 health evidence. Uncertainty related to
potential confounding stems from the relatively small number of epidemiologic studies that have
evaluated PM10-2.5 health effect associations in copollutants models with both gaseous pollutants
and other PM size fractions. On the other hand, uncertainty related to the biological plausibility
of effects attributed to PM10-2.5 exposures results from the small number of controlled human
exposure and animal toxicology2 studies that have evaluated the health effects of experimental
PM10-2.5 inhalation exposures. The evidence supporting the 2019 ISA's "suggestive of, but not
sufficient to infer, a causal relationship" causality determinations for PM10-2.5, including
uncertainties in this evidence, is summarized in sections 4.3.1.1 to 4.3.1.6 below.

4.3.1.1 Mortality

Long-term exposures

Due to the dearth of studies examining the association between long-term PM10-2.5

exposure and mortality, the 2009 ISA concluded that the evidence was "inadequate to determine
if a causal relationship exists" (U.S. EPA, 2009, U.S. EPA, 2019). As reported in the 2019 ISA,
some recent cohort studies conducted in the U.S. and Europe report positive associations
between long-term PM10-2.5 exposure and total (nonaccidental) mortality, though results are
inconsistent across studies (U.S. EPA, 2019, Table 11-11). The examination of copollutant
models in these studies remains limited and, when included, PM10-2.5 effect estimates were often

2 Compared to humans, rats and mice have small nasal passages, allowing smaller fractions of inhaled PM10-2.5 to
penetrate into the thoracic regions of the lungs of rats and mice (U.S. EPA, 2019, section 4.1.6), contributing to
the relatively limited evaluation of PMi 0-2.5 exposures in animal studies.

4-10

-------
attenuated after adjusting for PM2.5 (U.S. EPA, 2019, Table 11-11). Across studies, PM10-2.5
exposure concentrations were estimated using a variety of approaches, including direct
measurements from dichotomous samplers, calculating the difference between PM10 and PM2.5
concentrations measured at collocated monitors, and calculating the difference between area-
wide concentrations of PM10 and PM2.5. As discussed above, temporal and spatial correlations
between these approaches have not been evaluated, contributing to uncertainty regarding the
potential for exposure measurement error (U.S. EPA, 2019, section 3.3.1.1 and Table 11-11).
The 2019 ISA concludes that this uncertainty "reduces the confidence in the associations
observed across studies" (U.S. EPA, 2019, p. 11-125). The 2019 ISA additionally concludes that
the evidence for long-term PM10-2.5 exposures and cardiovascular effects, respiratory morbidity,
and metabolic disease evidence provides limited biological plausibility for PMio-2 5-related
mortality (U.S. EPA, 2019, sections 11.4.1 and 11.4). Taken together, the 2019 ISA concludes
that, "this body of evidence is suggestive of, but not sufficient to infer, a causal relationship
between long-term PM10-2.5 exposure and total mortality" (U.S. EPA, 2019, p. 11-125).
Short-term exposures

The 2009 ISA concluded that the evidence is "suggestive of a causal relationship between
short-term exposure to PM10-2.5 and mortality" (U.S. EPA, 2009). The 2019 ISA included
multicity epidemiologic studies conducted primarily in Europe and Asia which continue to
provide consistent evidence of positive associations between short-term PM10-2.5 exposure and
total (nonaccidental) mortality (U.S. EPA, 2019, Table 11-9). Although these studies contribute
to increasing confidence in the PMio-2 5-mortality relationship, the use of a variety of approaches
to estimate PM10-2.5 exposures continues to contribute uncertainty to the associations observed.
Studies considered in the 2019 ISA continue to expand the assessment of potential copollutant
confounding of the PMio-2 5-mortality relationship and provide evidence that PM10-2.5
associations generally remain positive in copollutant models, though associations are attenuated
in some instances (U.S. EPA, 2019, section 11.3.4.1, Figure 11-28, Table 11-10). The 2019 ISA
concludes that, overall, the assessment of potential copollutant confounding is limited due to the
lack of information on the correlation between PM10-2.5 and gaseous pollutants and the small
number of locations in which copollutant analyses have been conducted. Associations with
cause-specific mortality provide some support for associations with total (nonaccidental)
mortality, though associations with cause-specific mortality, particularly respiratory mortality,
are more uncertain (i.e., wider confidence intervals) and less consistent (U.S. EPA, 2019, section
11.3.7). The 2019 ISA concludes that the evidence for PMio-2 5-related cardiovascular and
respiratory effects provides only limited support for the biological plausibility of a relationship
between short-term PM10-2.5 exposure and cardiovascular mortality (U.S. EPA, 2019, section
11.3.7). Based on the overall evidence, the 2019 ISA concludes that, "this body of evidence is

4-11

-------
suggestive of, but not sufficient to infer, a causal relationship between short-term PM10-2.5
exposure and total mortality" (U.S. EPA, 2019, p. 11-120).

4.3.1.2 Cardiovascular Effects

Long-term exposures

In the 2009 ISA, the evidence describing the relationship between long-term exposure to
PM10-2.5 and cardiovascular effects was characterized as "inadequate to infer the presence or
absence of a causal relationship." The limited number of epidemiologic studies reported
contradictory results and experimental evidence demonstrating an effect of PM10-2.5 on the
cardiovascular system was lacking (U.S. EPA, 2019, section 6.4).

The evidence relating long-term PM10-2.5 exposures to cardiovascular mortality remains
limited, with no consistent pattern of associations across studies and, as discussed above,
uncertainty stemming from the use of various approaches to estimate PM10-2.5 concentrations
(U.S. EPA, 2019, Table 6-70). The evidence for associations with cardiovascular morbidity has
grown since the 2009 ISA and, while results across studies are not entirely consistent, some
epidemiologic studies report positive associations with ischemic heart disease (IHD) and
myocardial infarction (MI) (U.S. EPA, 2019, Figure 6-34); stroke (U.S. EPA, 2019, Figure 6-
35); atherosclerosis; venous thromboembolism (VTE); and blood pressure and hypertension
(U.S. EPA, 2019, Section 6.4.6). PM10-2.5 cardiovascular mortality effect estimates are often
attenuated, but remain positive, in models that adjust for PM2.5. For morbidity outcomes,
associations are inconsistent in models that adjust for PM2.5, NO2, and chronic noise pollution
(U.S. EPA, 2019, p. 6-276). The lack of toxicological evidence for long-term PM10-2.5 exposures
represents a substantial data gap (U.S. EPA, 2019, section 6.4.10), resulting in the 2019 ISA
conclusion that "evidence from experimental animal studies is of insufficient quantity to
establish biological plausibility" (U.S. EPA, 2019, p. 6-277). Based largely on the observation of
positive associations in some epidemiologic studies, the 2019 ISA concludes that "evidence is
suggestive of, but not sufficient to infer, a causal relationship between long-term PM10-2.5
exposure and cardiovascular effects" (U.S. EPA, 2019, p. 6-277).

Short-term exposures

The 2009 ISA concluded that the available evidence for short-term PM10-2.5 exposure and
cardiovascular effects was "suggestive of a causal relationship." This conclusion was based on
several epidemiologic studies reporting associations between short-term PM10-2.5 exposure and
cardiovascular effects, including IHD hospitalizations, supraventricular ectopy, and changes in
heart rate variability (HRV). In addition, dust storm events resulting in high concentrations of
crustal material were linked to increases in total cardiovascular disease emergency department
visits and hospital admissions. However, the prior reviews noted the potential for exposure
measurement error and copollutant confounding in these epidemiologic studies. In addition, there

4-12

-------
was only limited evidence of cardiovascular effects from a small number of experimental studies
(e.g., animal toxicological studies and controlled human exposure studies) that examined short-
term PMio-2.5 exposures (U.S. EPA, 2009, U.S. EPA, 2019). In the 2019 ISA, key uncertainties
include the potential for exposure measurement error, copollutant confounding, and limited
evidence of biological plausibility for cardiovascular effects following inhalation exposure (U.S.
EPA, 2019, section 6.3.13).

The evidence for short-term PM10-2.5 exposure and cardiovascular outcomes has expanded
since the 2009 ISA, though important uncertainties remain. The 2019 ISA notes that there are a
small number of epidemiologic studies reporting positive associations between short-term
exposure to PM10-2.5 and cardiovascular-related morbidity outcomes. However, the ISA also
notes that these epidemiological studies include a limited assessment of potential copollutant
confounding, and that there is limited experimental evidence to support biological plausibility of
these outcomes. The 2019 ISA also concludes that it remains unclear how the approaches used to
estimate PM10-2.5 concentrations in epidemiologic studies may impact exposure measurement
error. Taken together, the 2019 ISA concludes that "the evidence is suggestive of, but not
sufficient to infer, a causal relationship between short-term PM10-2.5 exposures and
cardiovascular effects" (U.S. EPA, 2019, p.6-254).

4.3.1.3 Respiratory Effects

Short-term exposures

Based on a small number of epidemiologic studies observing associations with some

respiratory effects and limited evidence from experimental studies to support biological
plausibility, the 2009 ISA concluded that the relationship between short-term exposure to PM10-
2.5 and respiratory effects is "suggestive of a causal relationship" (U.S. EPA, 2009).
Epidemiologic findings were consistent for respiratory infection and combined respiratory-
related diseases, but not for COPD. Studies were characterized by overall uncertainty in the
exposure assignment approach and limited information regarding potential copollutant
confounding. Controlled human exposure studies of short-term PM10-2.5 exposures found no lung
function decrements and inconsistent evidence for pulmonary inflammation. Animal
toxicological studies were limited to those using non-inhalation (e.g., intra-tracheal instillation)
routes of PM10-2.5 exposure.

Recent epidemiologic findings consistently link PM10-2.5 exposure to asthma exacerbation
and respiratory mortality, with some evidence that associations remain positive (though
attenuated in some studies of mortality) in copollutant models that include PM2.5 or gaseous
pollutants. Studies provide limited evidence for positive associations with other respiratory
outcomes, including COPD exacerbation, respiratory infection, and combined respiratory-related
diseases (U.S. EPA, 2019, Table 5-36). As noted above for other endpoints, one source of

4-13

-------
uncertainty in these epidemiologic studies is the lack of a systematic evaluation of the various
methods used to estimate PM10-2.5 concentrations as well as the resulting uncertainty in the
spatial and temporal variability in PM10-2.5 concentrations compared to PM2.5 (U.S. EPA, 2019,
sections 2.5.1.2.3 and 3.3.1.1). Taken together, the 2019 ISA concludes that "the collective
evidence is suggestive of, but not sufficient to infer, a causal relationship between short-term
PM10-2.5 exposure and respiratory effects" (U.S. EPA, 2019, p. 5-270).

4.3.1.4 Cancer

Long-term exposures

In the 2012 review, few studies examined cancer following inhalation exposures to PM10-
2.5. Thus, the 2009 ISA determined the evidence was "inadequate to assess the relationship
between long-term PM10-2.5 exposures and cancer" (U.S. EPA, 2009). The scientific information
assessed in the 2019 ISA of long-term PM10-2.5 exposure and cancer remains limited, with a few
recent epidemiologic studies reporting positive, but imprecise, associations with lung cancer
incidence (U.S. EPA, 2019). Additionally, uncertainty remains in these studies with respect to
exposure measurement error due to the use of PM10-2.5 predictions that have not been validated
by monitored PM10-2.5 concentrations (U.S. EPA, 2019, sections 3.3.2.3 and 10.3.4). Relatively
few experimental studies of PM10-2.5 have been conducted, though available studies indicate that
PM10-2.5 exhibits two key characteristics of carcinogens: genotoxicity and oxidative stress. While
limited, such experimental studies provide some evidence of biological plausibility for the
findings in a small number of epidemiologic studies (U.S. EPA, 2019, section 10.3.4). Taken
together, the small number of epidemiologic and experimental studies, along with uncertainty
with respect to exposure measurement error, contribute to the determination in the 2019 ISA that,
"the evidence is suggestive of, but not sufficient to infer, a causal relationship between long-term
PM10-2.5 exposure and cancer" (U.S. EPA, 2019, p. 10-87).

4.3.1.5 Metabolic Effects

Long-term exposures

The 2009 ISA did not make a causality determination for PMio-25-related metabolic
effects. One epidemiologic study is assessed in the 2019 ISA that reports an association between
long-term PM10-2.5 exposure and diabetes incidence, while additional cross-sectional studies
report associations with effects on glucose or insulin homeostasis (U.S. EPA, 2019, section 7.4).
As discussed above for other outcomes, uncertainties with the epidemiologic evidence include
the potential for copollutant confounding and exposure measurement error (U.S. EPA, 2019,
Tables 7-14 and 7-15). The evidence base to support the biological plausibility of metabolic
effects following PM10-2.5 exposures is limited, but a cross-sectional study that investigated
biomarkers of insulin resistance and systemic and peripheral inflammation may support a

4-14

-------
pathway leading to type 2 diabetes (U.S. EPA, 2019, sections 7.4.1 and 7.4.3). Based on the
expanded, though still limited evidence base, the 2019 ISA concludes that, "[ojverall, the
evidence is suggestive of, but not sufficient to infer, a causal relationship between [long]-term
PMio-2.5 exposure and metabolic effects" (U.S. EPA, 2019, p. 7-56).

4.3.1.6 Nervous system effects

Long-term exposures

The 2009 ISA did not make a causality determination for PMio-25-related nervous system
effects. In the 2019 ISA, available epidemiologic studies report associations between PM10-2.5
and impaired cognition and anxiety in adults in longitudinal analyses (U.S. EPA, 2019, Table 8-
25, section 8.4.5). Associations of long-term exposure with neurodevelopmental effects are not
consistently reported in children (U.S. EPA, 2019, sections 8.4.4 and 8.4.5). Uncertainties in
these studies include the potential for copollutant confounding, as no studies examined
copollutants models (U.S. EPA, 2019, section 8.4.5), and for exposure measurement error, given
the use of subtraction methods to estimate PM10-2.5 concentrations (U.S. EPA, 2019, Table 8-25,
section 3.4.2.3). In addition, there is only limited animal toxicological evidence supporting the
biological plausibility of nervous system effects (U.S. EPA, 2019, sections 8.4.1 and 8.4.5).
Overall, the 2019 ISA concludes that, "the evidence is suggestive of, but not sufficient to infer, a
causal relationship between long-term PM10-2.5 exposure and nervous system effects (U.S. EPA,
2019, p. 8-75).

4.3.1.7 Conclusions Drawn from the Evidence

With the evidence available in this reconsideration, as assessed in the 2019 ISA (U.S.
EPA, 2019) and summarized in subsections 4.3.1.1 to 4.3.1.6 above, we revisit the policy-
relevant questions posed at the beginning of this section:

For each of these categories of effects listed above, the 2019 ISA concludes that the
evidence is "suggestive of, but not sufficient to infer, a causal relationship" (U.S. EPA, 2019).
As summarized in the sections above, key uncertainties in the evidence result from limitations in
the approaches used to estimate ambient PM10-2.5 concentrations in epidemiologic studies, limited
examination of the potential for confounding by co-occurring pollutants, and limited support for
the biological plausibility of the serious effects reported in many epidemiologic studies. The
evidence base for several PMio-2 5-related health effects has expanded over time, broadening our
understanding of the range of health effects linked to PM10-2.5 exposures. This includes additional

4-15

-------
evidence for the relationships between long-term exposures and cardiovascular effects, metabolic
effects, nervous system effects, cancer, and mortality. However, the 2019 ISA identifies a
number of key limitations in the evidence, including the following:

• The use of multiple methods to estimate PM10-2.5 exposures in epidemiologic studies and
the lack of systematic evaluation of these methods, Overall, this lack of evaluation
contributes to uncertainty with respect to the spatial and temporal correlations in PM10-2.5
concentrations across methods, which may add to uncertainties in PM10-2.5 exposure
surrogates.

• The limited number of studies that evaluate PM10-2.5 health effect associations in
copollutant models, together with evidence from some studies for attenuation of
associations in such models, results in uncertainty in the independence of PM10-2.5 health
effect associations from co-occurring pollutants.

• The limited number of controlled human exposure and animal toxicology studies of
PM10-2.5 inhalation contribute to uncertainty in the biological plausibility of the PM10-2.5-
related effects reported in epidemiologic studies.

These uncertainties contribute to the conclusions in the 2019 ISA that the evidence for the PM10-
2.5-related health effects discussed in this section for both short- and long-term exposures is
"suggestive of, but not sufficient to infer, a causal relationship."

4.4 CASAC ADVICE AND PUBLIC COMMENTS

As part of its review of the draft PA, the CASAC has provided advice on the adequacy of
the public health protection afforded by the current primary PM10 standard. In its comments on
the draft PA, the CASAC notes that "the [d]raft PA concludes that the evidence reviewed in the

2019 ISA does not call into question the scientific judgments that informed the decision in the

2020 review to retain the current primary PM10 standard, without revision, in order to protect
against PM10-2.5 exposures" (Sheppard, 2022, p. 4 of consensus letter) and that "[t]he CASAC
supports this decision" and concurs with the draft PA's overall preliminary conclusion that it is
appropriate to consider retaining the current primary PM10 standard (Sheppard, 2022, p. 4 of
consensus letter). Additionally, the CASAC concurred that".. .at this time, PM10 is an
appropriate choice as the indicator for PM10-2.5" and "that it is important to retain the level of
protection afford by the current PM10 standard" (Sheppard, 2022, p. 4 of consensus letter).

The CASAC also recognized uncertainties associated with the scientific evidence,
including "compared to PM2.5 studies, the more limited number of epidemiology studies with
positive statistically significant findings, and the difficulty in extracting the sole contribution of
coarse PM to observed adverse health effects" (Sheppard, 2022, p. 19 of consensus responses).
The CASAC recommended several areas for additional research to reduce uncertainties in the
PM10-2.5 exposure estimates used in the epidemiologic studies, to assess the independence of
PM10-2.5 health effect associations, to evaluate the biological plausibility of PMio-2 5-related

4-16

-------
effects, and to increase the number of studies of PMio-25-related health effects in at-risk
populations (Sheppard, 2022, p. 20 of consensus responses). Furthermore, the CASAC
"recognizes a need for, and supports investment in research and deployment of measurement
systems to better characterize PM10-2.5" and to "provide information that can improve public
health" (Sheppard, 2022, p. 20 of consensus responses).

We also received a limited number of public comments on the adequacy of the primary
PM10 standard. Of those who provided comments on the primary PM10 standard, most
commenters supported the preliminary conclusion that it is appropriate to consider retaining the
current primary PM10 standard, without revision. However, one nonprofit organization suggested
that the primary PM10 standard should be strengthened to a level of 45 |ig/m3, consistent with the
World Health Organization Global Air Quality Guideline (WHO, 2021).

4.5 CONCLUSIONS ON THE ADEQUACY OF THE CURRENT
PRIMARY PM10 STANDARD

This section describes our conclusions regarding the adequacy of the current primary
PM10 standard. Our approach to reaching conclusions considers the EPA's assessment of the
scientific evidence for PMio-25-related health effects in the 2019 ISA. We revisit the overarching
question for this chapter:

• Does the available scientific evidence support or call into question the adequacy of
the protection afforded by the current primary PM10 standard against health effects
associated with exposures to PM10-2.5?

As an initial matter, we note that the scope of the updated scientific evaluation of the
health effects evidence in the ISA Supplement is based on those health effects categories where
the 2019 ISA concludes a causal relationship exists. Therefore, the ISA Supplement does not
include an evaluation of additional studies for PM10-2.5 and the 2019 ISA continues to serve as
the scientific foundation for assessing the adequacy of the primary PM10 standard in this
reconsideration of the 2020 final decision (U.S. EPA, 2019, section 1.7; U.S. EPA, 2022). As
such, this section describing our conclusions regarding the adequacy of the current primary PM10
standard draws heavily from the conclusions in the 2020 PA related to the primary PM10
standard (U.S. EPA, 2020), section 4.4) and on the evaluation of the health effects associated
with exposures to PM10-2.5 evaluated in the 2019 ISA (U.S. EPA, 2019, section 6.4.10). Lastly,
we recognize that a final decision on the primary PM10 standard in this reconsideration will be
largely a public health policy judgement in which the Administrator weighs the evidence,
including its associated uncertainties.

Regarding evidence for PMio-2 5-related health effects, we note that the evidence for
several PMio-2 5-related health effects has expanded, particularly for long-term exposures,

4-17

-------
broadening our understanding of the range of effects linked to PM10-2.5 exposures. The
epidemiologic studies considered in the 2019 ISA continue to report positive associations with
mortality or morbidity in cities across North America, Europe, and Asia, where PM10-2.5 sources
and composition are expected to vary widely. Such studies provide an important part of the body
of evidence supporting the strengthened causality determinations (and new determinations) for
long-term PM10-2.5 exposures and mortality, cardiovascular effects, metabolic effects, nervous
system effects and cancer (U.S. EPA, 2019). Although most of these studies examined PM10-2.5
health effect associations in urban areas, some studies have also linked mortality and morbidity
with relatively high ambient concentrations of particles of non-urban crustal origin from dust
storm events (U.S. EPA, 2019). Drawing from this evidence, we note continued support for
maintaining a standard that provides some measure of protection against exposures to PM10-2.5,
regardless of location, source of origin, or particle composition (78 FR 3176, January 15, 2013).
Thus, the scientific evidence evaluated for this reconsideration does not call into question the
decision in the 2020 review to maintain a primary standard that provides some measure of public
health protection against PM10-2.5 exposures, regardless of location, source of origin, or particle
composition.3

With regard to uncertainties, the 2019 ISA notes that important uncertainties remain in the
evidence base for PMio-2 5-related health effects. As summarized in section 4.3.1 above, these
include uncertainties in the PM10-2.5 exposure estimates used in epidemiologic studies, in the
independence of PM10-2.5 health effect associations, and in the biological plausibility of the
PMio-2 5-related effects. Thus, the evidence available in the 2019 ISA for consideration in
reaching conclusions in this reconsideration is subject to the same broad uncertainties present in
the 2012 review (U.S. EPA, 2019). Consistent with the assessment of the evidence in the 2009
ISA, these uncertainties contribute to the determinations in the 2019 ISA that the evidence for
key PMio-2 5-related health effects is "suggestive of, but not sufficient to infer" causal
relationships (U.S. EPA, 2019). Drawing from this information, we reach the conclusion that, as
in previous reviews, such uncertainties raise questions regarding the degree to which additional
public health improvements would be achieved by revising the existing PM10 standard.

With respect to indicator, we note that the evidence continues to support retaining the
PM10 indicator to provide public health protection against PMio-2 5-related effects. As the EPA
explained when it retained the PM10 indicator in 2006 and 2012, although PM10 includes both
fine and coarse PM, it is an appropriate indictor for thoracic particles because fine particles are

3 While we consider the full body of health evidence in reaching conclusions on the adequacy of the standard, we
recognize that the greatest emphasis is often placed on the health effects for which the evidence has been judged
to demonstrate a "causal" or "likely to be causal" relationship with PMi 0-2.5 exposures. The 2019 ISA did not
conclude that the available evidence for PMi0-2 5-related health effects supported such determinations.

4-18

-------
generally higher in urban areas. Given this, a PMio standard, set at a single unvarying level, will
generally result in lower allowable concentration of thoracic coarse particles in urban areas than
in nonurban areas. This targeting of protection is appropriate given that the strongest health
evidence associated with thoracic coarse particles comes from epidemiologic studies conducted
in urban areas, and higher fine particle concentrations in urban areas could result in increased
contamination of coarse particles and potentially increase the toxicity of thoracic coarse particles
in urban areas (71 FR 61195-97, October 17, 2006; 78 FR 3176-77, January 15, 2013).

When the above information is taken together, we reach the conclusion that the available
evidence does not call into question the scientific judgments that informed the decision in the
2020 review to retain the current primary PMio standard in order to protect against PM10-2.5
exposures. Specifically, while the evidence supports maintaining a PMio standard to provide
some measure of protection against PM10-2.5 exposures, uncertainties in the evidence lead to
questions regarding the potential public health implications of revising the existing PMio
standard. Thus, consistent with the approach taken in the previous reviews, we reach the
conclusion that the evidence does not call into question the adequacy of the public health
protection afforded by the current primary PMio standard. Furthermore, the available evidence in
this reconsideration of the 2020 final decision supports retaining the current standard. As such,
we have not evaluated alternative standards in this updated PA.

4.6 AREAS FOR FUTURE RESEARCH AND DATA COLLECTION

As discussed above, a number of key uncertainties and limitations in the health evidence
have been considered, consistent with those identified in the 2009 ISA and 2019 ISA, as well as
those noted by the CASAC. In this section, we highlight areas for future health-related research
and data collection activities from the 2020 PA to address these uncertainties and limitations in
the evidence (U.S. EPA, 2020, section 4.5). These efforts, if undertaken, could provide important
evidence for informing future reviews of the PM NAAQS. Key areas for future research efforts
are summarized below.

• The body of experimental inhalation studies of exposure to PM10-2.5 (e.g., controlled
human exposure and animal toxicology studies) is relatively sparse. While coarse PM
inhalation studies in rats and mice are complicated by substantial differences in nasal
anatomy that influences particle dosimetry (i.e., compared to humans), additional
experimental studies of short- or long-term PM10-2.5 exposures could play an important
role in informing biological plausibility in future ISAs.

• Existing epidemiologic studies have rarely examined associations with PM10-2.5 in
copollutant models, contributing to uncertainty in the degree to which reported health
effect associations are independent of potential confounding variables. Additional
epidemiologic studies that evaluate copollutants models would be informative.

4-19

-------
Epidemiologic studies use a variety of approaches to measure/estimate PM10-2.5
concentrations, including: (1) difference method with co-located monitors, (2) difference
method with area-wide averages of monitored PM10 and PM2.5, (3) difference method
with area-wide averages of modeled PM10 and PM2.5 or (4) direct measurement of
PM10-2.5 using a dichotomous sampler. It is important that we better understand how these
methods compare to one another, both in terms of absolute estimated concentrations and
in terms of the spatial and temporal correlations in those estimated concentrations
between methods.

Measurement capabilities and the availability of PM10-2.5 ambient concentration data have
greatly increased since the 2009 ISA (U.S. EPA, 2019, U.S. EPA, 2009, section
2.5.1.1.3). Starting in 2011, PM10-2.5 has been monitored at NCore stations, IMPROVE
stations, and several sites run by State and local agencies. Furthermore, there has been an
increase in the deployment of PM2.5 FEM monitors that also measure PM10-2.5. To date,
epidemiologic studies have used a variety of approaches to measure/estimate PM10-2.5
concentrations but have not used direct measurements from NCore or IMPROVE stations
to evaluate health effects associations with PM10-2.5 exposure. A body of epidemiologic
studies that evaluate health effect associations using monitoring data from these stations
could allow more direct comparisons of results across studies.

Evaluate and expand the PM10-2.5 network, along with speciation of PM10-2.5 including
multi-elements, major ions, carbon (including carbonate carbon), and bioaerosols.

Characterize PM10-2.5 in different health-relevant exposure environments (e.g., city center,
suburban, roadside, agricultural, and rural areas) for mass, elements (including potential
toxic species), carbonaceous materials (including selected organic compounds and
carbonate), water-soluble ions, and bioaerosols (including endotoxins, 1,3 beta glucans,
and total protein).

Additional areas of interest for future research include:

- Further evaluation of the potential for particular PM10-2.5 components, groups of
components, or other particle characteristics to contribute to exposure-related
health effects.

- Research to improve our understanding of concentration-response relationships
and the confidence bounds around these relationships, especially at lower ambient
PM10-2.5 concentrations.

- Studies focusing on populations, including children, that may be at increased risk
of PMio-2 5-related health effects. Additionally, studies that incorporate health
equity to assess both long- and short-term PM10-2.5 exposure burdens and health
effects.

- Modeling to estimate PM10-2.5 mass and composition in areas with sparse or less-
than-daily monitoring.

4-20

-------
REFERENCES

Sheppard, EA. (2022). Letter from Elizabeth A. (Lianne) Sheppard, Chair, Clean Air Scientific
Advisory Committee, to Administrator Michael S. Regan. Re: CASAC Review of the
EPA's Policy Assessment for the Review of the National Ambient Air Quality Standards
for Particulate Matter (External Review Draft - October 2021). March 18, 2022. EPA-
CAS AC-22-002. Office of the Administrator, Science Advisory Board U.S. EPA HQ,
Washington DC. Available at:

https://casac.epa.gov/ords/sab/f?p=105:18:10792850355838:::RP.18:P18 ID:2607#report

https://nepis.epa. gov/Exe/ZvPURL.cgi?Dockev=P 100AUMY.txt.

U.S. EPA (2016). Integrated review plan for the national ambient air quality standards for

U.S. EPA (2020). Policy Assessment for the Review of the National Ambient Air Quality

https://www.epa.gov/svstem/files/documents/2021-10/final-policv-assessment-for-the-
review-of-the-pm-naaqs-01-2020.pdf.

4-21

-------
WHO (2021). World Health Organization global air quality guidelines: particulate matter

(PM2.5 andPM10), ozone, nitrogen dioxide, sulfur dioxide and carbon monoxide. World
Health Organization. Geneva. Available at:
https://apps.who.int/iris/handle/10665/345329.

4-22

-------
5 RECONSIDERATION OF THE SECONDARY

STANDARDS FOR PM

This chapter presents and evaluates the policy implications of the scientific and technical
information pertaining to this reconsideration of the 2020 final decision on the secondary PM
standards. In so doing, the chapter presents key aspects of the evidence for the welfare effects of
PM documented in the 2019 ISA and ISA Supplement, with support from the prior ISA and
AQCDs, and associated public welfare implications, as well as key aspects of quantitative
analyses of recent air quality that are presented in the appendix associated with this chapter. As
described in detail in section 1.4.2, the ISA Supplement focuses on a thorough evaluation of
some studies that became available after the literature cutoff date of the 2019 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 ISA. The selection of the welfare
effects to evaluate within the ISA Supplement were based on the causality determinations
reported in the 2019 ISA and the subsequent use of scientific evidence in the 2020 PA.
Specifically, for welfare effects, the focus within the ISA Supplement is on visibility effects. The
ISA Supplement does not include an evaluation of studies on climate or materials effects.
Together, the scientific evidence and quantitative information provide the foundation for our
evaluation of welfare effects of PM in ambient air and the potential for welfare effects to occur
under air quality conditions associated with the current standards, as well as the associated public
welfare implications. Our evaluation is framed around key policy-relevant questions derived
from those included in the IRP (U.S. EPA, 2016) for the review completed in 2020 and also
takes into account the conclusions reached in the 2020 review. In this way we identify key
policy-relevant considerations and summary conclusions regarding the public welfare protection
provided by the current standards for the Administrator's consideration in this reconsideration of
the 2020 final decision on the secondary PM standards.

Within this chapter, background information on the current standards, including key
considerations in reaching the final decision in the 2020 review, is summarized in section 5.1.
The general approach for considering the information in this reconsideration of the 2020 final
decision, including policy-relevant questions identified to frame our policy evaluation, is
summarized in section 5.2. Key aspects of the welfare effects evidence, quantitative information,
and associated public welfare implications and uncertainties are addressed in section 5.3. Section
5.3.1 presents our consideration of the available scientific evidence and quantitative information
for visibility effects, while section 5.3.2 considers the scientific evidence for each of the non-

5-1

-------
visibility welfare effects (climate effects and materials effects) separately.1 Section 5.4
summarizes the advice and recommendations received from the CASAC during its review of the
draft PA, and public comments received on the draft document. Section 5.5 summarizes the key
evidence- and quantitative-based considerations identified in our evaluation and presents
associated summary conclusions of this analysis. Key remaining uncertainties and areas for
future research are identified in section 5.6.

5.1 BACKGROUND ON THE CURRENT STANDARDS

The current secondary PM standards were affirmed in 2020 based on the scientific and
technical information available at that time, as well as the Administrator's judgments regarding
the available welfare effects evidence, the appropriate degree of public welfare protection for the
existing standards, and available air quality information on visibility impairment that may be
allowed by such a standard (85 FR 82684, December 18, 2020). The welfare effects evidence
base available in the 2020 review included several decades of extensive research on the visibility
and non-visibility effects (climate effects and materials effects) of PM, conducted both in and
outside of the U.S., that documents the impacts of PM (U.S. EPA, 2019; U.S. EPA, 2009; U.S.
EPA, 2004b; U.S. EPA, 2004a). With the 2020 decision, the EPA retained the secondary 24-hour
PM2.5 standard, with its level of 35 |ig/m3, the annual PM2.5 standard, with its level of 15.0
|ig/m3, and the 24-hour PM10 standard, with its level of 150 |ig/m3. The sections below focus on
the key considerations, and the Administrator's conclusions, for climate and materials effects
(section 5.1.1) and visibility effects (section 5.1.2) in the 2020 review.

5.1.1 Non-Visibility Effects

In light of the robust evidence base, the 2019 ISA concluded there to be causal
relationships between PM and climate effects and material effects (U.S. EPA, 2019, sections
13.3.9 and 13.4.2). For climate effects, the 2019 ISA concluded that aerosols2 alter climate

1 Other welfare effects of PM, such as ecological effects, are being considered in the separate, on-going review of

the secondary NAAQS for oxides of nitrogen, oxides of sulfur and PM. Accordingly, the public welfare
protection provided by the secondary PM standards against ecological effects such as those related to deposition
of nitrogen- and sulfur-containing compounds in vulnerable ecosystems is being considered in that separate
review. Thus, the Administrator's conclusion in this reconsideration of the 2020 final decision will be focused
only and specifically on the adequacy of public welfare protection provided by the secondary PM standards from
effects related to visibility, climate, and materials.

2 In the climate sciences research community, PM is encompassed by what is typically referred to as aerosol. An

aerosol is defined as a solid or liquid suspended in a gas, but PM refers to the solid or liquid phase of an aerosol.
In this reconsideration of the 2020 final decision on the secondary PM NAAQS the discussion on climate effects
of PM uses the term PM throughout for consistency with the 2019 IS A (U. S. EPA, 2019) as well as to emphasize
that the climate processes altered by aerosols are generally altered by the PM portion of the aerosol. Exceptions to
this practice include the discussion of climate effects in the 2012 review, when aerosol was used when discussing

5-2

-------
processes directly through radiative forcing and by indirect effects on cloud brightness, changes
in precipitation, and possible changes in cloud lifetimes (U.S. EPA, 2019, section 13.3.9).
Additionally, the major aerosol components with the potential to affect climate processes (i.e.,
black carbon (BC), organic carbon (OC), sulfates, nitrates and mineral dusts) vary in their
reflectivity, forcing efficiencies, and direction of climate forcing (U.S. EPA, 2019, section
13.3.5). For materials effects, the 2019 ISA considered effects associated with the deposition of
PM (i.e., dry and wet deposition), including both physical damage (materials effects) and
aesthetic qualities (soiling effects). The deposition of PM can physically affect materials, adding
to the effects of natural weathering processes, by promoting or accelerating the corrosion of
metals; by degrading paints; and by deteriorating building materials such as stone, concrete, and
marble (U.S. EPA, 2019, section 13.4.2). Additionally, the deposition of PM from ambient air
can reduce the aesthetic appeal of buildings and objects through soiling.

The 2020 decision on the adequacy of the secondary standards for climate and materials
effects was a public welfare policy judgment made by the Administrator, which drew upon the
available scientific evidence for PM-attributable climate and materials effects and recognized
that the evidence did not support a quantitative assessment of exposures and public welfare risks
based on impacts to climate and materials. Noting the strong evidence indicating that aerosols
affect climate, the Administrator further considered what the available information indicated
regarding the adequacy of protection provided by the secondary PM standards. He noted that a
number of uncertainties in the scientific information affected our ability to quantitatively
evaluate the standards in this regard. For example, the 2019 ISA and 2020 PA noted the spatial
and temporal heterogeneity of PM components that contribute to climate forcing, uncertainties in
the measurement of aerosol components, inadequate consideration of aerosol impacts in climate
modeling, insufficient data on local and regional microclimate variations and heterogeneity of
cloud formations (U.S. EPA, 2019, section 13.3.9). In light of these uncertainties and the lack of
sufficient data, the 2020 PA concluded that "the data remain insufficient to conduct quantitative
analyses for PM effects on climate in the current review" (U.S. EPA, 2020, pp. 5-34 to 5-35) and
that there was insufficient information available to base a national ambient air quality standard
on climate impacts associated with ambient air concentrations of PM or its constituents (U.S.
EPA, 2020, section 5.4).

With regard to materials effects, the Administrator noted that the 2020 PA noted that
quantitative relationships were lacking between characteristics of PM and frequency of
repainting and repair of surfaces and that considerable uncertainty exists in the contributions of

suspending aerosol particles, and for certain acronyms that are widely used by the climate community that include
the term aerosol (e.g., aerosol optical depth, or AOD).

5-3

-------
co-occurring pollutants to materials damage and soiling processes (U.S. EPA, 2020, p. 5-35).
The 2020 PA concluded that none of the evidence available called into question the adequacy of
the existing secondary PM standards to protect against material effects (U.S. EPA, 2020, section
5.4).

The 2020 final decision was based on a thorough review in the 2019 ISA of the scientific
information on PM-induced climate and materials effects. The decision also took into account:
(1) assessments in the 2020 PA of the most policy-relevant information in the 2019 ISA
regarding evidence of adverse effects of PM to climate and materials, (2) uncertainties in the
available evidence to inform a quantitative assessment of PM-related climate and materials
effects, (3) CASAC advice and recommendations, and (4) public comments received during the
development of these documents and on the proposal notice.

Consistent with the general approach routinely employed in NAAQS reviews, the initial
consideration in the 2020 review of the secondary standards was with regard to the adequacy of
protection provided by the then-existing standards. Key aspects of that consideration are
summarized in section 5.1.1.1 below.

5.1.1.1 Considerations Regarding Adequacy of the Existing Standards for Non-
Visibility Effects in the 2020 Review

In considering non-visibility welfare effects in the 2020 review, as discussed above, the
Administrator concluded that, while it is important to maintain an appropriate degree of control
of fine and coarse particles to address non-visibility welfare effects, "it is generally appropriate
to retain the existing standards and that there is insufficient information to establish any distinct
secondary PM standards to address climate and materials effects of PM" (85 FR 82744,
December 18, 2020).

With regard to climate, the Administrator recognized that there were a number of
improvements and refinements to climate models since the 2012 review. However, while the
evidence continued to support a causal relationship between PM and climate effects, the
Administrator noted that significant limitations continued to exist related to quantifying the
contributions of direct and indirect effects of PM and PM components on climate forcing (U.S.
EPA, 2020, sections 5.2.2.1.1 and 5.4). He also recognized that the models continued to exhibit
considerable variability in estimates of PM-related climate impacts as regional scales (e.g., -100
km) as compared to simulations at global scales. Therefore, the resulting uncertainty led the
Administrator to conclude that the available scientific information in the 2020 review remained
insufficient to quantify climate impacts associated with particular concentrations of PM in
ambient air (U.S. EPA, 2020, section 5.2.2.2.1) or to evaluate or consider a level of PM air
quality in the U.S. to protect against climate effects and that there was insufficient information

5-4

-------
available to base a national ambient standard on climate impacts (85 FR 82744, December 18,
2020).

With regard to materials effects, the Administrator noted that the evidence available in
the 2019 ISA continued to support a causal relationship between materials effects and PM
deposition (U.S. EPA, 2019, section 13.4). He recognized that the deposition of fine and coarse
particles to materials can lead to physical damage and/or impaired aesthetic qualities. Particles
can contribute to materials damage by adding to the natural weathering processes and by
promoting the corrosion of metals, the degradation of building materials, and the weakening of
material components. While some new information was available in the 2019 ISA, the
information was from studies primarily conducted outside of the U.S. in areas where PM
concentrations in ambient air are typically higher than those observed in the U.S. (U.S. EPA,
2020, section 13.4). Additionally, the information assessed in the 2019 ISA did not support
quantitative analyses of PM-related materials effects in the 2020 review (U.S. EPA, section
5.2.2.2.2). Given the limited amount of information available and its inherent uncertainties and
limitations, the Administrator concluded that he was unable to relate soiling or damage to
specific levels of PM in ambient air or to evaluate or consider a level of air quality to protect
against such materials effects, and that there was insufficient information available to support a
distinct national ambient standard based on materials effects (85 FR 82744, December 18, 2020).

In the 2020 decision, for all of the reasons discussed above and recognizing the CASAC
conclusion that the evidence provided support for retaining the current secondary PM standards,
the Administrator concluded that it was appropriate to retain the existing secondary PM
standards, without revision. His decision was consistent with the CASAC advice related to non-
visibility effects. Specifically, the CASAC agreed with the 2020 PA conclusions that, while these
effects are important, "the available evidence does not call into question the protection afforded
by the current secondary PM standards" and recommended that the secondary standards "should
be retained" (Cox, 2019a, p. 3 of letter). For climate and materials effects, this conclusion
reflected his judgment that, although it remains important to maintain secondary PM2.5 and PM10
standards to provide some degree of control over long- and short-term concentrations of both
fine and coarse particles, there was insufficient information to establish distinct secondary PM
standards to address non-visibility PM-related welfare effects (85 FR 82744, December 18,
2020). Thus, the Administrator concluded that it was appropriate to retain all aspects of the
existing 24-hour PM2.5, annual PM2.5, and 24-hour PM10 secondary standards (85 FR 82744,
December 18, 2020).

5-5

-------
5.1.2 Visibility Effects

Visibility refers to the visual quality of a human's view with respect to color rendition
and contrast definition. It is the ability to perceive landscape form, colors, and textures. Visibility
involves optical and psychophysical properties involving human perception, judgment, and
interpretation. Light between the observer and the object can be scattered into or out of the sight
path and absorbed by PM or gases in the sight path. Given the strength of the evidence base, the
2019 ISA concluded that, "the evidence is sufficient to conclude that a causal relationship exists
between PM and visibility impairment" (U.S. EPA, 2019, section 13.2.6). Visibility impairment
is caused by light scattering and absorption by suspended particles and gases, including water
content of aerosols.3 The available evidence in the 2012 review indicated that specific
components of PM have been shown to contribute to visibility impairment. For example, at
sufficiently high relative humidity values, sulfate and nitrate are the PM components that scatter
more light and thus contribute most efficiently to visibility impairment. Elemental carbon (EC)
and OC are also important contributors, especially in the northwestern U.S. where their
contribution to PM2.5 mass is higher. Crustal materials can be significant contributors to visibility
impairment, particularly for remote areas in the arid southwestern U.S. (U.S. EPA, 2009, section
2.5.1; 2019 ISA, section 13.2.4.1).

Visibility impairment can have implications for people's enjoyment of daily activities
and for their overall sense of well-being (U.S. EPA, 2009, section 9.2). Consistent with the
evidence available in the 2012 review, the 2019 ISA evaluated available visibility preference
studies that were part of the overall body of evidence, and these preference studies were
considered in the 2020 PA (U.S. EPA, 2020, pp. 5-15 to 5-17). These preference studies
provided information about the potential public welfare implications of visibility impairment
from surveys in which participants were asked questions about their preferences or the values
they placed on various visibility conditions, as displayed to them in scenic photographs or in
images with a range of known light extinction levels.4

3 All particles scatter light and, although a larger particle scatters more light than a similarly shaped smaller particle

of the same composition, the light scattered per unit of mass is greatest for particles with diameters from ~0.3-1.0
|im (U.S. EPA, 2009, section 2.5.1; U.S. EPA, 2019, section 13.2.1). Particles with hygroscopic components
(e.g., particulate sulfate and nitrate) contribute more to light extinction at higher relative humidity than at lower
relative humidity because they change size in the atmosphere in response to relative humidity.

4 Preference studies were available in four urban areas. Three western preference studies were available, including

one in Denver, Colorado (Ely et al., 1991), one in the lower Fraser River valley near Vancouver, British
Columbia, Canada (Pryor, 1996), and one in Phoenix, Arizona (BBC Research & Consulting, 2003). A pilot focus
group study was also conducted for Washington, DC (Abt Associates, 2001), and a replicate study with 26
participants was also conducted for Washington, DC (Smith and Howell, 2009). More details about these studies
are available in Appendix D.

5-6

-------
The 2020 decision on the adequacy of the secondary standards with regard to visibility
effects was a public welfare policy judgment made by the Administrator, which drew upon the
available scientific evidence for PM-related visibility effects and on analyses of visibility
impairment, as well as judgments about the appropriate weight to place on the range of
uncertainties inherent in the evidence and analyses. Consistent with the approach in the 2012
review, the analyses utilized a PM2.5 visibility index based on an algorithm, known as the
IMPROVE algorithm,5 that provides for the estimation of light extinction (bext), in units of Mm"1,
using routinely monitored components of fine (PM2.5) and coarse (PM10-2.5) PM. The quantitative
analyses focused on PM2.5 based on conclusions in the 2019 ISA that fine particles scatter more
light than coarse particles on a per unit mass basis and include sulfates, nitrates, organics, light-
absorbing carbon, and soil (Malm et al., 1994). The 2019 ISA also concluded that hygroscopic
particles like ammonium sulfate, ammonium nitrate, and sea salt increase in size as relative
humidity increases, leading to increased light scattering (U.S. EPA, 2019, section 13.2.3).
Included in this decision were judgments on the weight to place on the visibility preference
studies; on the weight to give associated uncertainties, including those related to variability in
visibility preferences across the studies in different areas of the U.S.; variability in in occurrence
of visibility impairment in areas of the U.S., especially in urban areas; and on the extent to which
such effects in such areas may be considered adverse to public welfare.

The 2020 final decision was based on a thorough review in the 2019 ISA of the scientific
information on PM-related visibility effects. The decision also took into account: (1) assessments
in the 2020 PA of the most policy-relevant information in the 2019 ISA regarding evidence of
adverse effects of PM on visibility; (2) air quality analyses of the PM2.5 visibility index and
design values based on the form and averaging time of the existing secondary 24-hour PM2.5
standard; (3) CASAC advice and recommendations; and (4) public comments received during
the development of these documents and on the 2020 proposal notice.

Consistent with the general approach routinely employed in NAAQS reviews, the initial
consideration in the 2020 review of the secondary PM standards was with regard to the adequacy
of the protection provided by the then-existing standards. Key aspects of that consideration are
summarized in section 5.1.2.1 below.

5 The algorithm is referred to as the IMPROVE algorithm as it was developed specifically to use monitoring data
generated at IMPROVE network sites and with equipment specifically designed ot support the IMPROVE
program and was evaluated using IMPROVE optical measurements at the subset of monitoring sites that make
those measurements (Malm et al., 1994).

5-7

-------
5.1.2.1 Consideration Regarding the Adequacy of the Existing Standards for
Visibility Effects in the 2020 Review

In considering the visibility effects in the 2020 review, the Administrator noted the long-
standing body of evidence for PM-related visibility impairment. This evidence, which is based
on the fundamental relationship between light extinction and PM mass, demonstrated that
ambient PM can impair visibility in both urban and remote areas, and had changed very little
since the 2012 review (U.S. EPA, 2019, section 13.1; U.S. EPA, 2009a, section 9.2.5). The
evidence related to public perception of visibility impairment was from studies from four areas in
North America. These studies provided information to inform our understanding of levels of
visibility impairment that the public judged to be "acceptable" (U.S. EPA, 2010; 85 FR 24131,
April 30, 2020). In considering these public preference studies, the Administrator noted that, as
described in the 2019 ISA, no new visibility studies had been conducted in the U.S. and there
was little newly available information with regard to acceptable levels of visibility impairment in
the U.S. The Administrator recognized that visibility impairment can have implications for
people's enjoyment of daily activities and their overall well-being, and therefore, considered the
degree to which the current secondary standards protect against PM-related visibility
impairment.

Consistent with the 2012 review, in the 2020 review, the Administrator first concluded
that a target level of protection for a secondary PM standard is most appropriately defined in
terms of a visibility index that directly takes into account the factors (i.e., species composition
and relative humidity) that influence the relationship between PM2.5 in ambient air and PM-
related visibility impairment. In defining a target level of protection, the Administrator
considered the specific aspects of such an index, including the appropriate indicator, averaging
time, form and level (78 FR 82742-82744, December 18, 2020).

First, with regard to indicator, the Administrator noted that in the 2012 review, the EPA
used an index based on estimates of light extinction by PM2.5 components calculated using an
adjusted version of the IMPROVE algorithm, which allows the estimation of light extinction
using routinely monitored components of PM2.5 and PM10-2.5, along with estimates of relative
humidity. The Administrator recognized that, while there have been some revisions to the
IMPROVE algorithm since the time of the 2012 review, our fundamental understanding of the
relationship between PM in ambient air and light extinction had changed little and the various
IMPROVE algorithms appropriately reflected this relationship across the U.S. In the absence of
a monitoring network for direct measurement of light extinction, he concluded that calculated
light extinction indicator that utilizes the IMPROVE algorithms continued to provide a
reasonable basis for defining a target level of protection against PM-related visibility impairment
(78 FR 82742-82744, December 18, 2020).

5-8

-------
In further defining the characteristics of a visibility index, the Administrator next
considered the appropriate averaging time, form, and level of the index. Given the available
scientific information in the review, and in considering the CASAC's advice and public
comments, the Administrator concluded that, consistent with the decision in the 2012 review, a
visibility index with a 24-hour averaging time and a form based on the 3-year average of annual
90th percentile values remained reasonable. With regard to the averaging time and form of such
an index, the Administrator noted analyses conducted in the last review that demonstrated
relatively strong correlations between 24-hour and subdaily (i.e., 4-hour average) PM2.5 light
extinction (78 FR 3226, January 15, 2013), indicating that a 24-hour averaging time is an
appropriate surrogate for the sub-daily time periods of the perception of PM-related visibility
impairment and the relevant exposure periods for segments of the viewing public. This decision
in the 2020 review also recognized that a 24-hour averaging time may be less influenced by
atypical conditions and/or atypical instrument performance (78 FR 3226, January 15, 2013). The
Administrator recognized that there was no new information to support updated analyses of this
nature, and therefore, he believed these analyses continued to provide support for consideration
of a 24-hour averaging time for a visibility index in this review. With regard to the statistical
form of the index, the Administrator noted that, consistent with the 2012 review: (1) a multi-year
percentile form offers greater stability from the occasional effect of interannual meteorological
variability (78 FR 3198, January 15, 2013; U.S. EPA, 2011, p. 4-58); (2) a 90th percentile
represents the median of the distribution of the 20 percent worst visibility days, which are
targeted in Federal Class I areas by the Regional Haze Program; and (3) public preference
studies did not provide information to identify a different target than that identified for Federal
Class I areas (U.S. EPA, 2011, p. 4-59). Therefore, the Administrator judged that a visibility
index based on estimates of light extinction, with a 24-hour averaging time and a 90th percentile
form, averaged over three years, remained appropriate (78 FR 82742-82744, December 18,
2020).

With regard to the level of a visibility index, consistent with the 2012 review, the
Administrator judged that it was appropriate to establish a target level of protection of 30
deciviews (dv),6 7 reflecting the upper end of the range of visibility impairment judged to be
acceptable by at least 50% of study participants in the available public preference studies (78 FR
3226, January 15, 2013). The 2011 PA identified a range of levels from 20 to 30 dv based on the
responses in the public preference studies available at that time (U.S. EPA, 2011, section 4.3.4).

6 Deciview (dv) refers to a scale for characterizing visibility that is defined directly in terms of light extinction. The

deciview scale is frequently used in the scientific and regulatory literature on visibility.

7 For comparison, 20 dv, 25 dv, and 30 dv are equivalent to 64, 112, and 191 megameters (Mm1), respectively.

5-9

-------
At the time of the 2012 review, the Administrator noted a number of uncertainties and limitations
in public preference studies, including the small number of stated preference studies available,
the relatively small number of study participants, the extent to which the study participants may
not be representative of the broader study area population in some of the studies, and the
variations in the specific materials and methods used in each study. In considering the available
preference studies, with their inherent uncertainties and limitations, the prior Administrator
concluded that the substantial degree of variability and uncertainty in the public preference
studies should be reflected in a target level of protection based on the upper end of the range of
candidate protection levels (CPLs).

Given that there were no new preference studies available in 2020 review, the
Administrator's judgments were based on the same studies, with the same range of levels,
available in the 2012 review. As identified in the 2020 PA (U.S. EPA, 2020, section 5.5), there
were a number of limitations and uncertainties associated with these studies, including the
following:

• Available studies may not represent the full range of preferences for visibility in the U.S.
population, particularly given the potential variability in preferences based on the
conditions commonly encountered and the scenes being viewed.

• Available preference studies were conducted 15 to 30 years ago and may not accurately
represent the current day preferences of people in the U.S.

• The variety of methods used in the preference studies may potentially influence the
responses as to what level of impairment is deemed acceptable.

• Factors that are not captured in the methods of the preference studies, such as the time of
day when light extinction is the greatest or the frequency of impairment episodes, may
influence people's judgment on acceptable visibility (U.S. EPA, 2020, section 5.2.1.1).

Therefore, in considering the scientific information, with its uncertainties and limitations,
as well as public comments on the level of the target level of protection against visibility
impairment, the Administrator concluded that it is appropriate to again use a level of 30 dv for
the visibility index (78 FR 82742-82744, December 18, 2020).

Having concluded that the protection provided by a standard defined in terms of a PM2.5
visibility index, with a 24-hour averaging time, and a 90th percentile form, averaged over 3 years,
set at a level of 30 dv, was requisite to protect public welfare with regard to visual air quality, the
Administrator next considered the degree of protection from visibility impairment afforded by
the existing suite of secondary PM standards.

In this context, the Administrator considered the updated analyses of visibility
impairment presented in the 2020 PA (U.S. EPA, 2020, section 5.2.1.2), which reflected a
number of improvements since the 2012 review. Specifically, the updated analyses examined
multiple versions of the IMPROVE equation, including the version incorporating revisions since

5-10

-------
the time of the 2012 review. These updated analyses provided a further understanding of how
variation in the inputs to the algorithms affect the estimates of light extinction (U.S. EPA, 2020,
Appendix D). Additionally, for a subset of monitoring sites with available PM10-2.5 data, the
updated analyses better characterized the influence of coarse PM on light extinction than in the
2012 review (U.S. EPA, 2020, section 5.2.1.2).

The results of the updated analyses in the 2020 PA were consistent with those from the
2012 review. Regardless of which version of the IMPROVE equation was used, the analyses
demonstrated that, based on 2015-2017 data, the 3-year visibility metric was at or below about
30 dv in all areas meeting the current 24-hour PM2.5 standard, and below 25 dv in most of those
areas. In locations with available PM10-2.5 monitoring, which met both the current 24-hour
secondary PM2.5 and PM10 standards, 3-year visibility index metrics were at or below 30 dv
regardless of whether the coarse fraction was included as an input to the algorithm for estimating
light extinction (U.S. EPA, 2020, section 5.2.1.2). While the inclusion of the coarse fraction had
a relatively modest impact on the estimates of light extinction, the Administrator recognized the
continued importance of the PM10 standard given the potential for larger impacts on light
extinction in areas with higher coarse particle concentrations, which were not included in the
analyses in the 2020 PA due to a lack of available data (U.S. EPA, 2019, section 13.2.4.1; U.S.
EPA, 2020, section 5.2.1.2). He noted that the air quality analyses showed that all areas meeting
the existing 24-hour PM2.5 standard, with its level of 35 |ig/m3, had visual air quality at least as
good as 30 dv, based on the visibility index. Thus, the secondary 24-hour PM2.5 standard would
likely be controlling relative to a 24-hour visibility index set at a level of 30 dv. Additionally,
areas would be unlikely to exceed the target level of protection for visibility of 30 dv without
also exceeding the existing secondary 24-hour standard. Thus, the Administrator judged that the
24-hour PM2.5 standard provided sufficient protection in all areas against the effects of visibility
impairment, i.e., that the existing 24-hour PM2.5 standard would provide at least the target level
of protection for visual air quality of 30 dv which he judged appropriate (78 FR 82742-82744,
December 18, 2020).

5.2 GENERAL APPROACH AND KEY ISSUES IN THIS
RECONSIDERATION OF THE 2020 FINAL DECISION

This reconsideration of the 2020 final decision on the secondary PM standards is most
fundamentally based on using the Agency's assessment of the scientific evidence and associated
quantitative analyses to inform the Administrator's judgments regarding secondary standards that
are requisite to protect public welfare from known or anticipated adverse effects. This PA is
intended to help bridge the gap between the scientific evidence and information assessed in the
2019 ISA and ISA Supplement and the judgments required of the Administrator in determining

5-11

-------
whether it is appropriate to retain or revised the secondary PM NAAQS. The approach for this
reconsideration of the 2020 final decision on the secondary PM standards builds on previous
reviews, including the substantial assessments and evaluations performed in those reviews, and
taking into account scientific information and air quality data to inform our understanding of the
key policy-relevant issues in this reconsideration.

The evaluations in this PA, of the scientific assessments in the 2019 ISA and ISA
Supplement8 augmented by quantitative air quality analyses, are intended to inform the
Administrator's public welfare policy judgments and conclusions, including his decisions as to
whether to retain or revise these standards. The PA considers the potential implications of
various aspects of the scientific evidence, the air quality information, and the associated
uncertainties and limitations. In so doing, the approach for this PA involves evaluating the
scientific and technical information to address a series of key policy-relevant questions using
both evidence- and quantitative-based considerations. Together, consideration of the full set of
evidence and information in this reconsideration will inform the answer to the following initial
overarching question for the reconsideration:

• Do the scientific evidence and quantitative information support or call into question
the adequacy of the protection afforded by the current secondary PM standards?

In reflecting on this question in the remaining sections of this chapter, we consider the
body of scientific evidence assessed in the 2019 ISA and ISA Supplement and considered as
basis for developing or interpreting air quality analyses, including whether it supports or calls
into question the scientific conclusions reached in the 2020 review regarding welfare effects
related to exposure to PM in ambient air. Information in this reconsideration of the 2020 final
decision that may be informative to public policy judgments on the significance or adversity of
key effects on the public welfare is also considered. Additionally, the quantitative information,
whether newly developed in this reconsideration or predominantly developed in the past and
interpreted in light of current information, is considered, including with regard to the extent to
which it may continue to support judgments made in previous reviews.

The approach to reaching conclusions on the current secondary PM standards and, as
appropriate, on potential alternative standards, including consideration of policy-relevant
questions that frame the current reconsideration, is illustrated in Figure 5-1.

8 As noted above and described in detail in section 1.4.2, the ISA Supplement focuses on a thorough evaluation of
some studies that became available after the literature cutoff date of the 2019 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 ISA. The selection of the welfare effects to evaluate within the ISA Supplement were
based on the causality determinations reported in the 2019 ISA and the subsequent use of scientific evidence in
the 2020 PA. Specifically, for welfare effects, the focus within the ISA Supplement is on visibility effects. The
ISA Supplement does not include an evaluation of studies on climate or materials effects.

5-12

-------
I

Indicator

>Supportfor PM2 5 and/or PM1()?
>Supportforindicator(s) based
on other size fraction, PM
components, light extinction, etc,?

AveraainqTime

>Supportfor current 24-hour
and/or annual?

>Support for sub-daily,
seasonal, or other averaging
time(s)?

Form

>Supportfor retaining existing
forms?

>Supportfor alternative form based
on daylight hours or other metric?

Level

>Supportfor PM-attributable visibility, climate, or materials impacts at PM concentrations corresponding
to various potential standard levels?

> Support from quantitative exposure and/or risk assessments for public welfare improvements with
various potential standard levels?

>Uncertainties and limitations in the extent to which revised standard levels could result in public
welfare improvements, compared to existing standards

Identify range of potential alternative secondary standards for consideration

Figure 5-1. Overview of general approach for the reconsideration of the 2020 final decision
on the secondary PM standards.

5-13

-------
The Agency's approach in its reconsideration of the 2020 final decision on the secondary
standards is consistent with the requirements of the provisions of the CAA related to the review
of NAAQS and with how the EPA and the courts have historically interpreted the CAA. As
discussed in section 2.1 above, these provisions require the Administrator to establish secondary
standards that, in the Administrator's judgment, are requisite (i.e., neither more nor less stringent
than necessary) to protect the public welfare from known or anticipated adverse effects
associated with the presence of the pollutant in ambient air. In so doing, the Administrator
considers advice from the CASAC and public comment.

Consistent with the Agency's approach across all NAAQS reviews, the approach of this
PA to informing the Administrator's judgments in this reconsideration of the 2020 final decision
on the secondary PM standards is based on a recognition that the evidence generally reflects
continuums that include ambient air exposures for which scientists generally agree that effects
are likely to occur through lower levels at which the likelihood and magnitude of response
become increasingly uncertain. The CAA does not require that standards be set at a zero-risk
level, but rather at a level that reduces risk sufficiently so as to protect the public welfare from
known or anticipated adverse effects. The Agency's decisions on the adequacy of the current
secondary standards and, as appropriate, on any potential alternative standards considered in a
review, are largely public welfare policy judgments made by the Administrator. The four basic
elements of the NAAQS (i.e., indicator, averaging time, form, and level) are considered
collectively in evaluating the protection afforded by the current standard, or any alternative
standards considered. Thus, the Administrator's final decisions in such reviews draw upon the
scientific information and analyses about welfare effects, environmental exposures and risks, and
associated welfare significance, as well as judgments about how to consider the range and
magnitude of uncertainties that are inherent in the scientific evidence and analyses.

5.3 WELFARE EFFECTS AND QUANTITATIVE INFORMATION

In considering the evidence for welfare effects attributable to PM presented in the 2019
ISA and the ISA Supplement, this section poses the following policy-relevant questions:

• Does the scientific evidence and quantitative information support or call into

question the adequacy of the welfare protection afforded by the current secondary
PM standards?

In answering this question, we have posed a series of more specific questions to aid in
considering the scientific evidence and quantitative information, as discussed below. In
considering the scientific and technical information, we reflect upon both the information in
previous reviews and information that is assessed and presented in the 2019 ISA (U.S. EPA,
2019) and in the ISA Supplement (U.S. EPA, 2022), focusing on welfare effects for which the

5-14

-------
evidence supports either a "causal" or a "likely to be causal" relationship as described in the
Preamble to the ISA (U.S. EPA, 2015). Table 5-1 lists such causality determinations from the
2019 ISA for welfare effects. As in previous reviews, the evidence is sufficient to support a
causal relationship between PM and visibility effects (section 5.3.1), climate effects (section
5.3.2) and materials effects (section 5.3.2).

While the 2019 ISA provides the broad scientific foundation for this reconsideration, we
recognize that additional literature has become available since the cutoff date of the 2019 ISA
that expands the body of evidence related to visibility effects that can inform the Administrator's
judgments on the adequacy of the current secondary PM standards. As such, the ISA Supplement
builds on the information in the 2019 ISA with a target identification and evaluation of new
scientific information regarding visibility effects (U.S. EPA, 2022, section 1.2). As described in
Chapter 1, the selection of the welfare effects to evaluate within the ISA Supplement were based
on the causality determinations reported in the 2019 ISA and the subsequent use of scientific
evidence in the 2020 PA. The ISA Supplement focuses on U.S. and Canadian studies that
provide new information on public preference for visibility impairment and/or developed new
methodologies or conducted quantitative analyses of light extinction (U.S. EPA, 2022, section
1.2). Such studies of visibility effects and quantitative relationships between visibility
impairment and PM in ambient air were considered to be of greatest utility in informing the
Administrator's conclusions on the adequacy of the current secondary PM standards. The
visibility effects evidence presented within the 2019 ISA, along with the targeted identification
and evaluation of new scientific information in the ISA Supplement, provides the scientific basis
for the reconsideration of the 2020 final decision on the secondary PM standards. For climate
and materials effects, the 2020 PA concluded that there were substantial uncertainties associated
with the quantitative relationships with PM concentrations and the concentration patterns that
limited the ability quantitatively assess the public welfare protection provided by the standards
from these effects. Therefore, for climate and materials effects, we draw heavily from the 2020
PA in our evaluation of the information related to these effects and in reaching conclusions in
this PA.

Table 5-1. Key causality determinations for PM-related welfare effects.

Effect

2009 PM ISA

2019 PM ISA

Visibility effects

Causal

Climate effects

Causal

Materials effects

Causal

5-15

-------
5.3.1 Visibility Effects

In the sections below, we consider the nature of visibility-related effects attributable to
PM (section 5.3.1.1) and the quantitative assessment of visibility impairment (section 5.3.1.2).

5.3.1.1 Nature of Effects

In considering the evidence of visibility welfare effects attributable to PM as presented in
the 2019 ISA and the ISA Supplement, this section addresses the following policy-relevant
question:

• Does the available scientific evidence alter our conclusions from the 2020 review
regarding the nature of visibility effects attributable to PM in ambient air?

These subsequent questions consider the characterization and quantification of light
extinction and preferences associated with varying degrees of visibility impairment.

• To what extent is information available that changes or enhances our understanding
of the physics of light extinction and/or its quantification (e.g., through light
extinction or other monitoring methods or through algorithms such as IMPROVE)?

Our understanding of the relationship between light extinction and PM mass has changed
little since the completion of the 2009 ISA (U.S. EPA, 2009). The combined effect of light
scattering and absorption by particles and gases is characterized as light extinction, i.e., the
fraction of light that is scattered or absorbed per unit of distance in the atmosphere. Light
extinction is measured in units of 1/distance, which is often expressed in the technical literature
as visibility per megameter (abbreviated Mm"1). Higher values of light extinction (usually given
in terms of Mm"1 or dv) correspond to lower visibility. When PM is present in the air, its
contribution to light extinction is typically much greater than that of gases (U.S. EPA, 2019,
section 13.2.1). The impact of PM on light scattering depends on particle size and composition,
as well as relative humidity. All particles scatter light, as described by the Mie theory, which
relates light scattering to particle size, shape and index of refraction (U.S. EPA, 2019, section

5-16

-------
13.2.3; Van de Hulst, 1981; Mie, 1908). Fine particles scatter more light than coarse particles on
a per unit mass basis and include sulfates, nitrates, organics, light-absorbing carbon, and soil
(Malm et al., 1994). Hygroscopic particles like ammonium sulfate, ammonium nitrate, and sea
salt increase in size as relative humidity increases, leading to increased light scattering (U.S.
EPA, 2019, section 13.2.3).

Direct measurements of PM light extinction, scattering, and absorption are considered
more accurate for quantifying visibility impairment than PM mass-based estimates because they
do not depend on assumptions about particle characteristics (e.g., size, shape, density, component
mixture, etc.). Measurements of light extinction can be made with high time resolution, allowing
for characterization of subdaily temporal patterns of visibility impairment. Measurement
methods include transmissometers for measurement of light extinction and the determination of
visual range and integrating nephelometers for measurement of light scattering, as well as
teleradiometers and telephotometers, and photography and photographic modeling (U.S. EPA,
2019, section 13.2.2.2, Table 13.1; U.S. EPA, 2009; U.S. EPA, 2004b). While some recent
research confirms and adds to the body of knowledge regarding direct measurements as is
described in the 2019 ISA and ISA Supplement, no major new developments have been made
with these measurement methods since prior reviews (U.S. EPA, 2019, section 13.2.2.2; U.S.
EPA, 2022, section 4.2).

A theoretical relationship between light extinction and PM characteristics has been
derived from Mie theory (U.S. EPA, 2019, Equation 13-5) and can be used to estimate light
extinction by combining mass scattering efficiencies of particles with particle concentrations
(U.S. EPA, 2019, section 13.2.3; U.S. EPA, 2009, sections 9.2.2.2 and 9.2.3.1). However,
routine ambient air monitoring rarely includes measurements of particle size and composition
information with sufficient detail for these calculations. Accordingly, a much simpler algorithm
has been developed to make estimating light extinction more practical.

The algorithm, known as the IMPROVE algorithm,9 estimates light extinction (bext,
measured in units of Mm"1), using routinely monitored components of fine (PM2.5) and coarse
(PM10-2.5) PM. Relative humidity data are also needed to estimate the contribution by liquid
water that is in solution with the hygroscopic components of PM. To estimate each component's
contribution to light extinction, their concentrations are multiplied by extinction coefficients and
are additionally multiplied by a water growth factor that accounts for their expansion with
moisture. Both the extinction efficiency coefficients and water growth factors of the IMPROVE

9 The algorithm is referred to as the IMPROVE algorithm as it was developed specifically to use monitoring data
generated at IMPROVE network sites and with equipment specifically designed ot support the IMPROVE
program and was evaluated using IMPROVE optical measurements at the subset of monitoring sites that make
those measurements (Malm et al., 1994).

5-17

-------
algorithm have been developed by a combination of empirical assessment and theoretical
calculation using particle size distributions associated with each of the major aerosol components
(U.S. EPA, 2019, section 13.2.3.1, section 13.2.3.3).

The original IMPROVE algorithm (Equation D-l in Appendix D), so referenced here to
distinguish it from subsequent variations developed later, was found to underestimate the highest
light scattering values and overestimate the lowest values at IMPROVE monitors throughout the
U.S. (Malm and Hand, 2007; Ryan et al., 2005; Lowenthal and Kumar, 2004) and at sites in
China (U.S. EPA, 2019, section 13.2.3.3). To resolve these biases, a revised IMPROVE equation,
shown in Equation D-2 in Appendix D, was developed (Pitchford et al., 2007) that divides PM
components into smaller and larger sizes of particles in PM2.5, with separate mass scattering
efficiencies and hygroscopic growth functions for each size category. The revised IMPROVE
equation was described in detail in the 2009 ISA (U.S. EPA, 2009) and at that time, it both
reduced bias at the lowest and highest scattering values and improved the accuracy of the
calculated light bext. However, poorer precision was observed with the revised IMPROVE
equation compared to the original IMPROVE equation (U.S. EPA, 2009).10 Recent research
suggests that changes in PM composition in ambient air can impact the accuracy of estimating
light extinction using the IMPROVE algorithms (U.S. EPA, 2022 , section 4.2.2). As an
example, a study by Prenni et al. (2019) found that the relationship between directly measured
light scattering and estimated light scattering using the revised IMPROVE equation has changed
over time in recent years. In particular, Prenni et al. (2019) compared estimated light extinction
using the revised IMPROVE equation with measured light extinction using nephelometers from
2001-2016 and found that the revised IMPROVE equation underestimated light extinction at
many sites, especially for locations that experienced large decreases in sulfate and organic mass
concentrations. They further found that the underestimation results from splitting the components
into smaller and larger sizes of particles, with too much of the mass being allocated to the
smaller size fraction which has a lower dry mass scattering efficiency (U.S. EPA, 2022, section
4.2.2; Prenni et al., 2019).

Since the 2012 review, Lowenthal and Kumar (2016) have tested and evaluated a number
of modifications to the revised IMPROVE equation based on evaluations of monitoring data
from remote IMPROVE sites. In these locations, they observed that the multiplier to estimate the
concentration of organic matter, [OM], from the concentration of organic carbon, [OC], was

10 In the most recent IMPROVE report, a combination of the original and revised IMPROVE equations (the
modified original IMPROVE equation) was used (Hand et al., 2011). This equation uses the sea salt term of the
revised equation but does not subdivide the components into two size classes. Further, it uses a factor of 1.8 to
estimate organic matter from organic carbon concentrations and also replaces the constant value of 10 Mm"1 used
for Rayleigh scattering in the original and revised equations with a site-specific term based on elevation and mean
temperature.

5-18

-------
closer to 2.1 than the value of 1.8 used in the revised IMPROVE equation.11 They also observed
that water soluble organic matter absorbs water as a function of relative humidity, which is not
accounted for in either the original or revised IMPROVE equations and was therefore
underestimated in these equations. They further suggested that light scattering by sulfate was
overestimated because the assumption that all sulfate is fully neutralized ammonium sulfate is
not always true (U.S. EPA, 2019, section 13.2.3.3). Modifications based on these points are
reflected in Equation D-3 in Appendix D.

In summary, rather than altering our understanding from previous reviews, we continue
to recognize that direct measurements are better at characterizing light extinction than estimating
light extinction with an algorithm. However, in the absence of advances in the monitoring
methods and/or network for directly measuring light extinction, the use of the IMPROVE
equation for estimating light extinction continues to be supported by the evidence, with some
refinements to the inputs of the IMPROVE equation. Accordingly, as in previous reviews, this
reconsideration focuses on calculated light extinction when quantifying visibility impairment
resulting from recent concentrations of PM in ambient air.

• What does the information indicate with regard to factors that influence light
extinction and visibility, as well as variation in these factors and resulting light
extinction across the U.S.?

The 2019 ISA provides a comprehensive discussion of the spatial and temporal patterns
of PM2.5 composition and its contribution to light extinction from IMPROVE and CSN
monitoring sites, which are mostly rural and urban, respectively.12 The data from these sites for
the periods of 2005-2008 and 2011-2014 were used in the 2019 ISA to identify differences in
species contributing to light extinction in urban and rural areas by region and season. This is an
expansion over the analysis in the 2009 ISA, in that the measurements at that time were
primarily based measurements from monitors located in rural areas and at remote sites (U.S.
EPA, 2019, section 13.2.4.1, Figures 13-1 through 13-14).

Focusing on the time period of 2011-2014, some major differences in estimated light
extinction are apparent among regions of the U.S. Annual average calculated bext was
considerably greater in the East and Midwest than in the Southwest. Based on IMPROVE data,

11 In areas near sources, PM is often less oxygenated, and therefore, in these locations, much of the organic PM mass
is present as OC (Jimenez et al., 2009). In areas further away from PM sources, organic PM mass is often more
oxygenated as a result of photochemical activity and interactions with other PM and gaseous components in the
atmosphere (Jimenez et al., 2009). Under these conditions, the multiplier to convert OC to OM may be higher
than in locations with less aged organic PM.

12 Monitors were grouped into 28 IMPROVE regions and 31 CSN regions based on site location and PM
concentrations for major species. For comparison purposes, and where possible, CSN regions were defined
similarly to those forthe IMPROVE network (Hand et al., 2011; U.S. EPA, 2019, section 13.2.4.1).

5-19

-------
annual average bext was greater than 40 Mm"1 in the Southeast, East Coast, Mid-South, Central
Great Plains, and Appalachian regions, with the highest annual average bext (greater than 50 Mm"
') in the Ohio River Valley,13 while annual average bext was below 40 Mm"1 for all Western
IMPROVE regions. Annual average bext values were also generally higher in the East than the
West based on CSN data, although the highest annual average bext was in the Sacramento/San
Joaquin Valley and Los Angeles areas (U.S. EPA, 2019, section 13.2.4.1, Figure 13-1, Figure
13-3, Figure 13-5).

Consistent with the analysis in the 2019 ISA, a recent study analyzed national and
regional trends in light extinction based on reconstructed total light extinction estimated from
IMPROVE data using 5-year aggregates of annual mean b ext (Mm"1) for 2000-2004 and 2014-
2018 (U.S. EPA, 2022 , section 4.2.2). Hand et al. (2020) found that, for 2000-2004, the highest
levels of bext occurred in the Appalachian Mountains and Ohio River valley (-100 Mm"1 or
greater), with decreasing values in the central U.S. (-70 Mm"1). Values of bext in the East
significantly decreased over time, reduced to -50 Mm"1 in the 2014-2018 time period, likely
corresponding to decreases in sulfate concentrations over time. However, for 2014-2018, the
highest values of bext were in the central U.S. (50-60 Mm"1), which is an area with high
agricultural activity and nitrate and ammonium concentrations. During both time periods, lower
bext occurred in the western U.S. (20-30 Mm"1), with improvements in bext closer to the West
Coast in 2014-2018 compared to 2004-2008.

Moreover, Hand et al. (2020) also explored changes in bext over time as relative trends (%
yr"1) and found spatial variability in long-term and short-term trends. Generally, similar
magnitudes and spatial variability were found for both long-term and short-term trends, with the
strongest reductions in bext across the eastern U.S. (-4% yr"1 or greater) and along the West Coast,
particularly in Southern California. There was less improvement in the Intermountain West14 (-
2% yr"1), although air quality in these areas have been increasingly impacted by wildfire activity
and biomass smoke in recent years (Hand et al., 2020). Decreased trends also occurred across the
Southwest, but at a lower rate than in the Eastern U.S. Over the entire continental U.S., on
average, bext decreased at a rate of -2.8% yr"1 from 2002 to 2018 and -1.8% yr"1 from 1992 to
2018, with much of the improvement occurring in the eastern U.S. (U.S. EPA, 2022 , section
4.2.2; Hand et al., 2020).

Components of PM2.5 contributing to light extinction vary regionally. For example, in the
analysis completed in the 2019 ISA, in the Eastern regions, ammonium sulfate accounted for
approximately 35 to 60% of the annual average bext, with the greatest contributions typically

13 A bext value of 40 Mm"1 corresponds to a visual range of about 100 km.

14 The Intermountain West area includes Idaho, Montana, northern Wyoming, and portions of northern California.

5-20

-------
occurring in the summer (U.S. EPA, 2019, section 13.2.4.1). The second greatest contribution to
light extinction came from particulate organic matter (POM), ranging from about 20 to 30% of
annual average bext with less seasonal variation on average than ammonium sulfate. Ammonium
nitrate also contributed approximately 10% to 35% of annual average bext, with much higher
concentrations in the winter than in the summer (U.S. EPA, 2019, section 13.2.4.1). In the
Northwest, POM was the largest contributor to annual average bext, up to 70%, in most urban and
rural regions with the greatest contributions in the fall. This seasonal contribution of POM may
be related to wildfires. A few exceptions included Boise and sites in North Dakota, where
ammonium nitrate was the greatest contributor, and sites in the Alaska IMPROVE region, where
ammonium sulfate was the greatest contributor (U.S. EPA, 2019, section 13.2.4.1). In the
Southwest, based on IMPROVE data, ammonium sulfate or POM were generally the greatest
contributors to annual average bext, with nearly equivalent contributions in several regions. Based
on CSN data, ammonium nitrate was often the greatest contributor, with especially high bext
contributions in the winter. While PM10-2.5 mass scattering was relatively small in the eastern and
northwestern U.S., in the Southwest, PM10-2.5 mass scattering contributed to more than 20% of
light extinction (U.S. EPA, 2019, section 13.2.4.1).

Differences also exist between the urban CSN and the mainly rural IMPROVE data.

Light extinction is generally higher in CSN regions than the geographically corresponding
IMPROVE regions. Annual average bext was greater than 50 Mm"1 in 11 CSN regions, compared
to only one IMPROVE region, and was greater than 20 Mm"1 in all CSN regions, compared to
just over half of the IMPROVE regions. Light absorbing carbon was the greatest contributor to
light extinction in several Western CSN regions but was not a large contributor in any of the
IMPROVE regions (U.S. EPA, 2019, Figure 13-11). Ammonium nitrate also accounted for more
light extinction in the CSN regions, while it was only a top contributor to bext in one IMPROVE
region (U.S. EPA, 2019, section 13.2.4.1).

From the 2005-2008 time period to the 2011-2014 time period, the annual average bext in
most CSN regions in the Eastern U.S. decreased by more than 20 Mm"1. This corresponds to an
improvement in average visual range in most Eastern U.S. regions of more than 6 Mm"1 (or 15
km) from 2005-2008 to 2011-2014. Additionally, the contribution of ammonium sulfate to light
extinction has also changed over this period. Due to decreased atmospheric sulfate
concentrations, the impact on visibility impairment is evident with a smaller fraction of the total
bext accounted for by ammonium sulfate in 2011-2014 compared to 2005-2008 (U.S. EPA, 2019,
section 13.2.4.1).

Additionally, Hand et al. (2020) observed that changes in PM composition in ambient air
also affect trends for annual, regional mean speciated bext at IMPROVE monitoring locations
across the U.S. In the East, annual mean total bext decreased by -4.3% yr"1 from 2002 to 2018,

5-21

-------
much of which is attributable to reductions of light extinction from ammonium sulfate. Light
extinction was also decreased for ammonium nitrate, although at a lower rate and a lower
magnitude than ammonium sulfate. Light extinction by POM, EC, and fine dust also decreased
over time, while light extinction by coarse PM increased slightly. In the Intermountain West and
Southwest, annual mean total bext decreased by -0.9% yr"1 from 2002 to 2018. The composition
of PM in these regions are different than in the East, and while light extinction from ammonium
sulfate and ammonium nitrate generally decreased over these time periods, their contribution to
light extinction in the Intermountain West and Southwest is less than in the East. Light extinction
by POM, EC, and fine dust decreased over time, while the trend for coarse PM remained
relatively the same, although the composition of the particles responsible for light extinction in
these areas shifted towards a more carbon-dominated composition over time. It is also important
to note that the trends observed in the Intermountain West and Southwest regions are likely
influenced by biomass smoke, as wildfire smoke emissions are the largest contributor to light
extinction by POM and the impacts of wildfires on air quality in these regions has increased in
recent years (Hand et al., 2020). Light extinction levels in the West Coast region were higher
than in the Intermountain West and Southwest regions, but generally decreased over time (-1.5%
yr"1). Light extinction by ammonium nitrate decreased at the highest rate in the West Coast
region, and it was the only area where the rate decreased at a greater rate than ammonium
sulfate. Light extinction by EC and fine dust also decreased, while the trend for POM generally
remained flat and light extinction by coarse mass increased slightly. The mix of positive and
negative trends in the West Coast region are likely due to the influence of biomass smoke in
northern California and Oregon, in particular during 2017 and 2018, as well as reductions in
NOx emissions in Southern California and reductions in light extinction by ammonium sulfate
across the region (U.S. EPA, 2022, section 4.2.2; Hand et al., 2020).

Since the completion of the 2019 ISA, additional research has emerged that explores the
impact of wildfire smoke and biomass smoke on PM composition in the U.S. The increases in
PM emissions from these sources coincides with decreases in SO2 and NOx emissions, which
influences the contribution of different PM species to light extinction. The evidence suggests that
PM emissions from wildfire and biomass smoke can impact visibility impairment due to general
changes in the dominant PM species in the ambient air during these events, as well as the
influence of particle size and aging of the PM over time (U.S. EPA, 2022 , section 4.2.2; Laing et
al., 2016; Kleinman et al., 2020).

In summary, the spatial and temporal analysis of PM monitoring network data in the
2019 ISA and recent evidence presented in the ISA Supplement emphasize that the extent of
light extinction by PM2.5 depends on PM2.5 composition and relative humidity. Regional
differences in PM2.5 composition greatly influence light extinction spatially and temporally.

5-22

-------
Changes in PM2.5 composition over time can also affect light extinction based on concentrations
of specific PM components in ambient air.

• To what extent are recent studies available that might inform judgments about the
potential adversity to public welfare of PM-attributable visibility impairment and
the nature of the relationship between PM-attributable visibility impairment and
public perceptions of such impairment?

In the 2012 review, visibility preference studies were available from four areas in North
America,15 as described in section 5.1.2 above. Study participants were queried regarding
multiple images that, depending on the study, were either photographs of the same location and
scenery that had been taken on different days on which measured extinction data were available
or digitized photographs onto which a uniform "haze" had been superimposed. Results of these
studies indicated a wide range of judgments on what study participants considered to be
acceptable visibility across the different study areas, depending on the setting depicted in each
photograph. As a part of the 2010 Urban Focused Visibility Assessment (UFVA), each study was
evaluated separately, and figures were developed to display the percentage of participants that
rated the visual air quality depicted as "acceptable" (U.S. EPA, 2010). Figure 5-2 represents a
graphical summary of the results of the studies in the four cities and identifies a range
encompassing the PM2.5 visibility index values from images that were judged to be acceptable by
at least 50% of study participants across all four of the urban preference studies (U.S. EPA,
2010, p. 4-24).16 The "50% acceptability" criterion is a useful index for comparison between the
preferences studies and is designed to identify the visual air quality level (defined in terms of
deciviews or light extinction) that best divides the photographs into two groups: those with a
visual air quality rated as acceptable by the majority of the participants, and those rated not
acceptable by the majority of participants (U.S. EPA, 2011, p. 4-23). As shown in Figure 5-2,
much lower visibility (considerably more haze resulting in higher values of light extinction) was
considered acceptable in Washington, D.C. than was in Denver. The median judgment for the
study groups in the two areas differed by 9.2 dv (which roughly corresponds to about 30 |ig/m3
ofPM)(U.S. EPA, 2010).

15 As noted above, preference studies were available in four urban areas in the last review: Denver, Colorado (Ely et
al., 1991, Pryor, 1996), Vancouver, British Columbia, Canada (Pryor, 1996), Phoenix, Arizona (BBC Research &
Consulting, 2003), and Washington, DC (Abt Associates, 2001; Smith and Howell, 2009). More details about
these studies are available in Appendix D.

16 Figure 5-2 shows the results of a logistical regression analysis using a logit model of the acceptable or
unacceptable ratings from participants of the studies. The logit model is a generalized linear model used for
binomial regression analysis which fits explanatory data about binary outcomes (in this case, a person rating an
image as acceptable or unacceptable) to a logistic function curve. A detailed description is available in Appendix
J of the 2010 UFVA (U.S. EPA, 2010).

5-23

-------
Light Extinction (Mm'1)
100 200

20 25 30

PM2.5 Visibility Index (dv)

a Denver

~ Phoenix

~ BC

~ Washington

Denver Logit

Phoenix Logit

BC Logit

DC Logit

Figure 5-2. Relationship of viewer acceptability ratings to light extinction. (Source: U.S.

EPA, 2011, Figure 4-2; U.S. EPA, 2010, Figure 2-16)

Since the completion of the 2012 review, there has been very little research on visibility
preferences, with one visibility preference study conducted in the Grand Canyon, AZ (Malm et
aL 2019) and one in Beijing, China (Fajardo et al., 2013). The Grand Canyon study, conducted
by Malm et al. (2019), has a similar study design to that used in the public preference studies
discussed above, however, there are several important differences that make it difficult to
directly compare the results of the Malm et al. (2019) study with other public preference studies.
As an initial matter, the Grand Canyon study was conducted in a Federal Class I area, as opposed
to in an urban area, with a scene depicted in the photographs that did not include urban
features.1' There is currently a lack of information to inform our understanding of how public
preferences of visibility impairment differ among study participants when the scenes depicted
include urban features compared to scenes without urban features, and it is not clear how
individual preferences could vary between such scene types. Further, the Malm et al. (2019)
study also used a much lower range of superimposed "haze" than the preference studies

17 The Grand Canyon study used a single scene looking west down the canyon with a small landscape feature of a
100-km-distant mountain (Mount Trumbull), along with other closer landscape features. The scenes presented in
the previously available visibility preference studies are presented in more detail in Table D-9 in Appendix D.

5-24

-------
discussed above, which may bias the participant responses given the generally lower visibility
range presented compared to the other studies.18 It is unclear whether the participant preferences
for the Grand Canyon study were generally lower than the other preference studies because of
the lower range of superimposed "haze" for the images in that study, or if their preferences
would vary if presented with images with a range of superimposed "haze" more comparable to
the levels used in the other studies (i.e., more "haze" superimposed on the images).

The study conducted in Beijing found a higher range of acceptable visibility impairment
among participants than was found in preference studies previously conducted in the U.S. This
finding may be related to the common occurrence of higher PM2.5 concentrations in Beijing (with
associated visibility impairment) than is typical in the U.S. (U.S. EPA, 2019, section 13.2.5).

Similarly, there is little recent information regarding acceptable levels of visibility
impairment in the U.S. One study explored alternate methods for evaluating "acceptable" levels
of visual air quality from the preference studies, including the use of scene-specific visibility
indices as potential indicators of visibility levels as perceived by the observer (Malm et al.,
2019). In addition to measures of atmospheric haze, such as atmospheric extinction, used in
previously available preference studies, other indices for visual air quality include color and
achromatic contrast of single landscape figures, average and equivalent contrast of an entire
scene, edge detection algorithms such as the Sobel index, and just-noticeable difference or
change indexes. The results reported by Malm et al. (2019) suggest that scene-dependent metrics,
such as contrast, may be useful alternate predictors of preference levels compared to universal
metrics like light extinction (U.S. EPA, 2022, section 4.2.1). This is because extinction alone is
not a measure of "haze," but of light attenuation per unit distance, and visible "haze" is
dependent on both light extinction and distance to a landscape feature (U.S. EPA, 2022, section
4.2.1).

• To what extent have important uncertainties in the evidence from the previous
reviews been addressed, and have new uncertainties emerged?

Since the 2012 review, some refinements have been made to the IMPROVE equation to
better estimate light extinction, but there has been no expansion of monitoring efforts for direct
measurement of light extinction. At the time of the 2012 review, it was noted that a PM2.5 light
extinction monitoring program could help with characterizing visibility conditions and the
relationships between PM component concentrations and light extinction.

18 The Grand Canyon study superimposed light extinction ranging from 3 dv to 20 dv on the image slides shown to
participants compared to the previously available preference studies. In those studies, the visibility ranges
presented were as low as 9 dv and as high as 45 dv. The visibility ranges presented in the previously available
visibility preference studies are described in more detail in Table D-9 in Appendix D.

5-25

-------
Little new research is available that helps to expand our understanding of visibility
preferences or our characterization of visibility conditions. Uncertainties and limitations
consistent with those identified in the past reviews persist in this reconsideration.

• Given the potential for people to have different preferences based on the visibility they are
used to based on conditions that they commonly encounter, and the potential for them to
also have different preferences for different types of scenes, the preference studies may
not capture the range of preferences of people in the U.S.

• Most of the preference studies were conducted 15 to 30 years ago and may not reflect the
visibility preferences of the U.S. population today. Given that air quality has improved
over the last several decades, the older studies may not reflect current preferences of
people in the U.S. Newer studies may not capture the extent to which preferences may be
changing over time.

• The preference studies have used different methods to evaluate what level of visibility
impairment is acceptable. Variability in study methodology may influence an individual's
response as to what level of visibility impairment is deemed acceptable, and thereby
influence the results of the study.

• Many factors that are not captured by the methods used in the preference studies may
influence people's judgments on acceptable visibility. For example, an individual's
perception of an acceptable level of visibility impairment could be influenced by the
duration of visibility impairment experienced, the time of day during which light
extinction is greatest, and the frequency of episodes of visibility impairment, as well as
the intensity of the visibility impairment (i.e., the focus of the studies).

• Methods for quantitatively evaluating people's judgments on acceptability are evolving
but are still inconsistent in their application across studies and there is no evidence to
suggest that one method more accurately quantifies such judgments compared to other
methods. Variability in quantitative methods for comparing visual air quality in public
preference studies may influence the consistency and comparability of results and the
interpretation of these results in the context of regional or national preferences for
visibility impairment in urban, non-urban, and Federal Class I areas.

Overall, the body of evidence regarding visibility effects remains largely unchanged since
the time of the 2012 review. While one new study provides refinements to the methods for
estimating light extinction, uncertainties and limitations in the scientific evidence during the
previous reviews remain.

5.3.1.2 Quantitative and Air Quality Information

Beyond our consideration of the scientific evidence, discussed in section 5.3.1.1 above,
we have also considered quantitative analyses of PM air quality and visibility impairment with
regard to the extent they could inform conclusions on the adequacy of the public welfare
protection provided by the current secondary PM standards. In the 2012 review, quantitative
analyses focused on daily visibility impairment, given the short-term nature of PM-related
visibility effects. Such quantitative analyses conducted as part of the 2012 review informed the

5-26

-------
decision on the secondary standards in that review (U.S. EPA, 2010, U.S. EPA, 2011; 78 FR
3189-3192, January 15, 2013). Consistent with the approaches used in previous reviews, we
utilized a two-phase assessment for visibility impairment. First, we consider the appropriateness
of the visibility index and its target level of protection against PM-related visibility effects.
Second, we evaluate the protection afforded by the current secondary 24-hour PM2.5 standard
compared to the target level of protection defined in terms of the visibility index. The
information available since the 2012 review includes an updated equation for estimating light
extinction, summarized in section 5.3.1.1 above and described in the 2019 ISA, as well as more
recent air monitoring data, that together allow for development of an updated assessment with
the potential to substantially add to our understanding of PM-related visibility impairment. Thus,
we have conducted updated analyses for this reconsideration based on the technical information,
tools, and methods.

• To what extent does the available scientific information alter our conclusions

regarding the visibility index and the appropriate target level of protection against
PM-related visibility effects?

In evaluating the adequacy of the current secondary PM standards, we first evaluate the
target level of protection identified for a distinct secondary standard to protect against visibility
effects. In previous reviews, the visibility index was set at a level of 30 dv, with estimated light
extinction as the indicator, a 24-hour averaging time, and a 90th percentile form, averaged over
three years.

With regard to an indicator for the visibility index, we recognize the lack of availability
of methods and an established network for directly measuring light extinction (section 5.3.1.1
above). Therefore, consistent with previous reviews, we consider a visibility index based on
estimates of light extinction by PM2.5 components derived from an adjusted version of the
original IMPROVE algorithm to again be the most appropriate indicator. As described in section
5.3.1.1 above, the IMPROVE algorithm estimates light extinction using routinely monitored
components of PM2.5 and PM10-2.5, along with estimates of relative humidity.

With regard to averaging time, we note that the evidence continues to provide support for
the short-term nature of PM-related visibility effects. Given that there is no new information
available regarding the time periods during which visibility impairment occurs or public
preferences related to specific time periods for visibility impairment, we continue to focus on
daily visibility impairment. Analyses were conducted in the 2012 review that showed relatively
strong correlations between 24-hour and sub-daily (i.e., 4-hour average) PM2.5 light extinction
that indicated that a 24-hour averaging time is an appropriate surrogate for the sub-daily time
periods relevant for visual perception (U.S. EPA, 2011, Figures G-4 and G-5; Frank, 2012)

These analyses continue to provide support for a 24-hour averaging time for the visibility index

5-27

-------
in this reconsideration. Consistent with previous reviews, we also note that the 24-hour
averaging time may be less influenced by atypical conditions and/or atypical instrument
performance than a sub-daily averaging time (85 FR 82740, December 18, 2020; 78 FR 3226,
January 15, 2013).

With regard to the form for the visibility index, the available information continues to
provide support for a 3-year average of annual 90th percentile values. Given that there is no new
information to inform selection of an alternate form, as in previous review, we note that the 3-
year average form provides stability from the occasional effect of inter-annual meteorological
variability that can result in unusually high pollution levels for a particular year (85 FR 82741,
December 18, 2020; 78 FR3198, January 15, 2013; U.S. EPA, 2011, p. 4-58). With regard to the
statistical form to be averaged over 3-years, we do not have new information available in this
reconsideration to inform an updated analyses of alternative statistical forms; therefore, we
consider the evaluation in the 2010 UFVA of three different statistical forms: 90th, 95th, and 98th
percentiles (U.S. EPA, 2010, Chapter 4). In considering this evaluation of statistical forms from
the 2010 UFVA, we note that, consistent with the 2011 PA, the Regional Haze Program targets
the 20 percent most impaired days for visibility improvements in visual air quality in Federal
Class I areas and that the median of the distribution of these 20 percent worst days would be the
90th percentile. The 2011 PA also noted that strategies that are implemented so that 90 percent of
days would have visual air quality that is at or below the level of the standard would reasonably
be expected to lead to improvements in visual air quality for the 20 percent most impaired days.
Additionally, as in the 2011 PA, we recognize that the available public preference studies do not
address frequency of occurrence of different levels of visibility and do not identify a basis for a
different target for urban areas than for Federal Class I areas (U.S. EPA, 2011, p. 4-59).
Therefore, the analyses and consideration for the form of a visibility index from the 2011 PA
continue to provide support for a 90th percentile form, averaged across three years, in defining
the characteristics of a visibility index in this reconsideration.

With regard to the level for the visibility index, the evidence continues to support that it
is appropriate to establish a target level of protection based on the upper end of the range of
levels of visibility impairment judged to be acceptable by at least 50% of study participants in
the visibility preference studies (section 5.3.1.1 above). The range of levels were identified in the
2011 PA based on the responses in the public preference studies available at that time. Given that
there have been no new public preference studies to inform our understanding of the
"acceptable" level of visibility impairment, we continue to rely on the same studies and the range
of 20 to 30 dv identified from those studies in previous reviews.

There are a number of uncertainties and limitations associated with the public preference
studies that continue to persist from previous reviews, as described in section 5.3.1.1 above. As

5-28

-------
such, there are substantial degrees of variability and uncertainty in the public preference studies
that should be considered in selecting an appropriate target level of protection for the visibility
index. Given the lack of new preferences studies in this reconsideration that might reduce these
uncertainties, the evidence continues to support a target level of protection at the upper end of
the range. Therefore, consistent with previous reviews, a target level of protection of 30 dv is an
appropriate level for a visibility index in this reconsideration.

Taken together, the available information continues to support a visibility index with a
level of 30 dv, with estimated light extinction as the indicator, a 24-hour averaging time, and a
90th percentile form, averaged over three years.

• How much visibility impairment is estimated to occur in areas that meet the current
secondary PM standards? What are the factors contributing to the estimates in areas
with higher values?

Consistent with the analyses conducted in the 2012 and 2020 reviews, we have conducted
analyses examining the relationship between PM mass concentrations and calculated light
extinction using the 3-year design values19 for the current secondary standards and a 3-year
average visibility metric based on light extinction estimated using IMPROVE equations using air
quality data for 2017 to 2019. Given that there has been little new research since the time of the
2012 review to better inform our understanding of visibility preferences in the U.S., there is no
new information available to inform selection of a visibility metric for evaluating visibility
impairment in this PA different from the one identified in the 2012 review, as described above.20
These analyses are intended to inform our understanding of visibility impairment in the U.S.
under recent air quality conditions, particularly those conditions that meet the current standards,
and the relative influence of various factors on light extinction. Given the relationship of
visibility with short-term PM, we focus particularly on the short-term PM standards.

Given that visibility-related effects are often associated with short-term PM
concentrations, and recognizing the relatively larger role of PM2.5 and its components in light
extinction and as inputs to the IMPROVE equation, we have given somewhat more attention to
consideration of the 24-hour PM2.5 standard. Analyses were conducted using three versions of
the IMPROVE equation (Equations D-l through D-3 in Appendix D) to estimate light extinction

19 A design value is a statistic that summarizes the air quality data for a given area in terms of the indicator,
averaging time, and form of the standard. Design values can be compared to the level of the standard and are
typically used to designate areas as meeting or not meeting the standard and assess progress towards meeting the
NAAQS.

20 While the results presented in the figures below focus on the relationship between PM mass concentrations and
calculated light extinction based on the current secondary 24-hour PM2 5 standard with its level of 35 |ig/m3 and a
3-year visibility average visibility metric with a level of 30 dv, the figures can also be used to evaluate recent
visibility conditions for other levels as well.

5-29

-------
to better understand the influence of variability in inputs across the three equations. This analysis
included 60 monitoring sites that are geographically distributed across the U.S. in both urban and
rural areas (see Figure D-l in Appendix D). These sites are those that have a valid 24-hour PM2.5
design value for the 2017-2019 period and met strict criteria for PM species for this analysis.21
We present results for these 60 sites using the original IMPROVE equation, with modifications
to the equation consistent with those made in evaluating light extinction in the 2012 review
(described in detail in section D. 1 of Appendix D). We also present results for these 60 sites with
light extinction calculated using the revised IMPROVE equation and the Lowenthal and Kumar
(2016) IMPROVE equation described in section 5.3.1.1 above.

In considering the relationship between the 24-hour PM2.5 mass-based design value and
the 3-year visibility metric using recent air quality data, we first examine the relationship using
the original IMPROVE equation, consistent with the methods used in the 2012 review (Kelly et
al., 2012; 78 FR 3201, January 15, 2013; Appendix D). In those areas that meet the current 24-
hour PM2.5 standard, all sites have light extinction estimates at or below 26 dv (Figure 5-3). For
the four locations that exceed the current 24-hour PM2.5 standard, light extinction estimates range
from 22 dv to 27 dv (Figure 5-3). These findings are consistent with the findings of the analyses
using the same IMPROVE equation in the 2012 review with data from 102 sites with data from
2008-2010 and in the 2020 review with data from 67 sites with data from 2015-2017. The
analyses presented in this PA indicate similar findings to those from the analyses in the 2012 and
2020 reviews, i.e., the updated quantitative analysis shows that the 3-year visibility metric was
no higher than 30 dv22 at sites meeting the current secondary PM standards, and at most such
sites the 3-year visibility index values are much lower (e.g., an average of 20 dv across the 60
sites).

21 For this analysis, completeness criteria for speciated PM data at these sites included having all 12 quarters in the
2017-2019 period with at least 11 days in each quarter with a valid PM2.5 and PMi0-2.5 mass, sulfate, nitrate,
organic carbon, elemental carbon, sea salt (chlorine or chloride), and fine soil (aluminum, silica, calcium, iron,
and titanium) measurement.

22 For comparison purposes in these air quality analyses, we use a 3 -year visibility metric with a level of 30 dv,
which is the highest level of visibility impairment judged to be acceptable by at least 50 percent of the
participants in the preference studies that were available at the time of the 2012 review (78 FR 3191, January 15,
2013).

5-30

-------
40 i

35-

£ 30-

25-

20-

15-

10-

• •

• Northeast (n = 13)

• Southeast (n = 8)

• IndustMidwest (n = 15)

• UpperMidwest (n=9)
Southwest (n = 2)

• Northwest (n=8)

• SoCal (n=4)

• Alaska (n = l)

40 45 50 55 60
¦25 concentration,

0 5 10 15 20 25 30 35

98th percentile of daily PM-

averaged over 3 years (pg nrf3)

Figure 5-3. Comparison of 90th percentile of daily light extinction, averaged over three
years, and 98th percentile of daily PM2.5 concentrations, averaged over three years, for
2017-2019 using the original IMPROVE equation. (Note: Dashed lines indicate the level
of current 24-hour PM2.5 standard (35 jig/m3) and the target level of protection
identified for the 3-year visibility metric (30 dv).)

When light extinction was calculated using the revised IMPROVE equation, the resulting
3-year visibility metrics are nearly identical to light extinction estimates calculated using the
original IMPROVE equation (Figure 5-4), but some sites are just slightly higher. As noted in
section 5.3.1.1, the revised IMPROVE equation divides PM components into smaller and larger
sizes of particles in PM2.5, with separate mass scattering efficiencies and hygroscopic growth
functions for each size category. Using the revised IMPROVE equation, for those sites that meet
the current 24-hour PM2.5 standard, the 3-year visibility metric is at or below 26 dv. For the four
locations that exceed the current 24-hour PM2.5 standard, light extinction estimates range from 22

5-31

-------
dv to 29 dv (Figure 5-4). These results are similar to those for light extinction calculated using
the original IMPROVE equation, and those from previous reviews.

40 i

• Northeast (n = 13)

• Southeast (n = 8)
IndustMidwest (n = 15)

• UpperMidwest (n=9)
Southwest (n = 2)

• Northwest (n=8)

• SoCal (n=4)

• Alaska (n = l)

5 10 15 20 25 30 35 40 45 50 55 60

98th percentile of daily PM2 5 concentration,
averaged over 3 years (pg nrf3)

Figure 5-4. Comparison of 90th percentile of daily light extinction, averaged over three
years, and 98th percentile of daily PM2.5 concentrations, averaged over three years, for
2017-2019 using the revised IMPROVE equation. (Note: Dashed lines indicate the level of
current 24-hour PM2.5 standard (35 |ig/m3) and the target level of protection identified for the
3-year visibility metric (30 dv).)

When light extinction was calculated using the refined equation from Lowenthal and
Kumar (2016), the resulting 3-year visibility metrics are slightly higher at all sites compared to
light extinction estimates calculated using the original IMPROVE equation (Figure 5-5). As
noted in section 5.3.1.1, this version of the IMPROVE equation uses a multiplier of 2.1 to

5-32

-------
convert the measured OC to OM for input into the equation and also accounts for water
absorption by water soluble organic matter as a function of relative humidity, likely contributing
to the slightly higher estimates of light extinction. As noted in section 5.3.1.1, the Lowenthal and
Kumar (2016) refinements to the IMPROVE equation are based on evaluations of monitoring
data from remote IMPROVE sites. More remote areas tend to have more aged organic particles
than urban areas, and these adjustments to the IMPROVE equation account for the higher
concentration of organic matter as a result of more aged organic particles at these sites. It is
important to note that, since the Lowenthal and Kumar (2016) refinements to the IMPROVE
equation likely result in one of the higher estimates of light extinction, this equation may
overestimate light extinction in non-remote areas, including those urban areas in our analyses.

Using the Lowenthal and Kumar (2016) equation, for those sites that meet the current 24-
hour PM2.5 standard, the 3-year visibility metric is at or below 28 dv when light extinction is
calculated. For those sites that exceed the current 24-hour PM2.5 standard, three of these sites
have a 3-year visibility metric ranging between 26 dv and 30 dv, while one site in Fresno,
California that exceeds the current 24-hour PM2.5 standard and has a 3-year visibility index value
of 32 dv (compared to 29 dv when light extinction is calculated with the original IMPROVE
equation) (see Table D-3 in Appendix D). At this site, it is likely that the 3-year visibility metric
using the Lowenthal and Kumar (2016) equation would be below 30 dv if PM2.5 concentrations
were reduced such that the 24-hour PM2.5 level of 35 |ig/m3 was attained.

5-33

-------
40 i

.9 LO

CD
"43 >

X "(j

QJ OJ
¦4—' T3

fD
T3

ru
O)
>>

• Northeast (n = 13)

• Southeast (n=8)

• IndustMidwest (n = 15)

• UpperMidwest (n = 9)
Southwest (n = 2)

• Northwest (n = 8)

• SoCal (n=4)

• Alaska (n = l)

5 10 15 20 25 30 35 40 45 50 55 60

98th percentile of daily PM2 5 concentration,
averaged over 3 years (pg nrr3)

Figure 5-5. Comparison of 90th percentile of daily light extinction, averaged over three
years, and 98th percentile of daily PM2.5 concentrations, averaged over three years, for
2015-2017 using the Lowenthal and Kumar equation. (Note: Dashed lines indicate the
level of current 24-hour PM2.5 standard (35 |ig/m3) and the target level of protection
identified for the 3-year visibility metric (30 dv).)

In considering visibility impairment under recent air quality conditions, we recognize that
the differences in the inputs to equations estimating light extinction can influence the resulting
values. For example, given the varying chemical composition of emissions from different
sources, the 2.1 multiplier in the Lowenthal and Kumar (2016) equation may not be appropriate
for all source types. At the time of the 2012 review, the EPA judged that a 1.6 multiplier for
converting OC to OM was more appropriate, for the purposes of estimating visibility index at
sites across the U.S., than the 1.4 or 1.8 multipliers used in the original and revised IMPROVE
equations, respectively. A multiplier of 1.8 or 2.1 would account for the more aged and
oxygenated organic PM that tends to be found in more remote regions than in urban regions,

5-34

-------
whereas a multiplier of 1.4 may underestimate the contribution of organic PM found in remote
regions when estimating light extinction (78 FR 3206, January 15, 2013; U.S. EPA, 2012a, p.
IV-5). The information and analyses indicate that it may be appropriate to select inputs to the
IMPROVE equation (e.g., the multiplier for OC to OM) on a regional basis rather than a national
basis when calculating light extinction. This is especially true when comparing sites with
localized PM sources (such as sites in urban or industrial areas) to sites with PM derived largely
from biogenic precursor emissions (that contribute to widespread secondary organic aerosol
formation), such as those in the southeastern U.S. We note, however, that conditions involving
PM from such different sources have not been well studied in the context of applying a multiplier
to estimate light extinction, contributing uncertainty to estimates of light extinction for such
conditions.

At the time of the 2012 review, the EPA noted that PM2.5 is the size fraction of PM
responsible for most of the visibility impairment in urban areas (77 FR 38980, June 29, 2012).
Data available at the time of the 2012 review suggested that, generally, PM10-2.5 was a minor
contributor to visibility impairment most of the time (U.S. EPA, 2010) although the coarse
fraction may be a major contributor in some areas in the desert southwestern region of the U.S.
Moreover, at the time of the 2012 review, there were few data available from PM10-2.5 monitors
to quantify the contribution of coarse PM to calculated light extinction. Since that time, an
expansion in PM10-2.5 monitoring efforts has increased the availability of data for use in
estimating light extinction with both PM2.5 and PM10-2.5 concentrations included as inputs in the
equations. The analysis in the 2020 review addressed light extinction at 20 of the 67 PM2.5 sites
where collocated PM10-2.5 monitoring data were available. Since the 2020 review, PM10-2.5
monitoring data are available at more locations and the analyses presented in this PA include
those for light extinction estimated with coarse and fine PM at all 60 sites. Generally, the
contribution of the coarse fraction to light extinction at these sites is minimal, contributing less
than 1 dv to the 3-year visibility metric (U.S. EPA, 2020, section 5.2.1.2). However, we note that
in our analysis, only a few sites were in locations that would be expected to have high
concentrations of coarse PM, such as the Southwest. These results are consistent with those in
the analyses in the 2019 ISA, which found that mass scattering from PM10-2.5 was relatively
small (less than 10%) in the eastern and northwestern U.S., whereas mass scattering was much
larger in the Southwest (more than 20%) particularly in southern Arizona and New Mexico (U.S.
EPA, 2019, section 13.2.4.1, p. 13-36).

In summary, the findings of these updated quantitative analyses are generally consistent
with those in the 2012 and 2020 reviews. The 3-year visibility metric was generally below 26 dv
in most areas that meet the current 24-hour PM2.5 standard. Small differences in the 3-year
visibility metric were observed between the variations of the IMPROVE equation, which may

5-35

-------
suggest that it may be more appropriate to use one version over another in different regions of
the U.S. based on PM characteristics such as particle size and composition to more accurately
estimate light extinction.

5.3.2 Non-Visibility Effects
5.3.2.1 Nature of Effects

In considering the evidence for non-visibility welfare effects attributable to PM as
presented in the 2019 ISA, this section poses the following policy-relevant questions:

• To what extent has the scientific evidence improved our understanding of the nature
and magnitude of non-visibility welfare effects of PM in ambient air, including the
variability associated with such effects? To what extent have important uncertainties
in the evidence from the last review been addressed, and have new uncertainties
emerged?

As an initial matter, we note that the ISA Supplement does not include an evaluation of
additional studies for climate and materials effects and the causality determinations from PM-
related climate and materials effects presented in the 2019 ISA continue to serve as the scientific
foundation for these effects. As such, the sections below that address these questions for PM and
climate effects (section 5.3.2.1.1) and materials effects (section 5.3.2.1.2) draw from the
evaluation of the welfare effects evidence for PM-related climate and materials effects in the
2019 ISA and considerations of such effects in the 2020 PA (U.S. EPA, 2020).

5.3.2.1.1 Climate Effects

In considering the evidence of climate effects attributable to PM, this section poses the
following policy-relevant question:

• To what extent is information available that changes or enhances our understanding
of the climate impacts of PM-related aerosols, particularly regarding a quantitative
relationship between PM concentrations and effects on climate (e.g., through
radiative forcing)?

In the 2012 review, the 2009 PM ISA concluded that there was "sufficient evidence to
determine a causal relationship between PM and climate effects - specifically on the radiative
forcing of the climate system, including both direct effects of PM on radiative forcing and
indirect effects that involve cloud feedbacks that influence precipitation formation and cloud
lifetimes" (U.S. EPA, 2009, section 9.3.10).23 Since the 2012 review, climate impacts have been

23 Radiative forcing (RF) for a given atmospheric constituent is defined as the perturbation in net radiative flux, at
the tropopause (or the top of the atmosphere) caused by that constituent, in watts per square meter (Wm~2), after
allowing for temperatures in the stratosphere to adjust to the perturbation but holding all other climate responses
constant, including surface and tropospheric temperatures (Fiore et al., 2015, Myhre et al., 2013). A positive

5-36

-------
extensively studied and the 2019 ISA concludes that "overall the evidence is sufficient to
conclude that a causal relationship exists between PM and climate effects" (U.S. EPA, 2019,
section 13.3.9). Recent research reinforces and strengthens the evidence evaluated in the 2009
ISA. Recent evidence provides greater specificity about the details of these radiative forcing
effects and increased understanding of additional climate impacts driven by PM radiative effects.
The Intergovernmental Panel on Climate Change (IPCC) assesses the role of anthropogenic
activity in past and future climate change. In the 2012 review, the 2009 ISA relied heavily on the
Fourth IPCC Assessment Report (AR4); since the 2012 review, the IPCC issued an updated
report, as described in the 2019 ISA. The Fifth IPCC Assessment Report (AR5; IPCC, 2013)
reports on the key scientific advances in understanding the climate effects of PM since AR4. The
2019 ISA draws substantially upon AR5 in summarizing these effects.

Atmospheric PM has the potential to affect climate in multiple ways, including absorbing
and scattering of incoming solar radiation, alterations in terrestrial radiation, effects on the
hydrological cycle, and changes in cloud properties (U.S. EPA, 2019, section 13.3.1).
Atmospheric PM interacts with incoming solar radiation. Many species of PM (e.g., sulfate and
nitrate) efficiently scatter solar energy. By enhancing reflection of solar energy back to space,
scattering PM exerts a cooling effect on the surface below. Certain species of PM such as black
carbon (BC), brown carbon (BrC), or dust can also absorb incoming sunlight. One study found
that whether absorbing PM warms or cools the underlying surface depends on several factors,
including the altitude of the PM layer relative to cloud cover and the albedo of the surface (Ban-
Weiss et al., 2014). PM also perturbs incoming solar energy by influencing cloud cover and
cloud lifetime. For example, PM provides nuclei upon which water vapor condenses, forming
cloud droplets. Finally, absorbing PM deposited on snow and ice can diminish surface albedo
and lead to regional warming (U.S. EPA, 2019, section 13.3.2).

PM has direct and indirect effects on climate processes. PM interactions with solar
radiation through scattering and absorption, collectively referred to as aerosol-radiation
interactions (ARI), are also known as the direct effects of PM on climate, as opposed to the
indirect effects that involve aerosol-cloud interactions (ACI). The direct effects of PM on climate
result primarily from particles scattering light away from Earth and sending a fraction of solar
energy back into space, decreasing the transmission of visible radiation to the surface of the
Earth and resulting in a decrease in the heating rate of the surface and the lower atmosphere. The
IPCC AR5, taking into account both model simulations and satellite observations, reports a
radiative forcing from aerosol-radiation interactions (RFari) from anthropogenic PM of -0.35 ±

forcing indicates net energy trapped in the Earth system and suggests warming of the Earth's surface, whereas a
negative forcing indicates net loss of energy and suggests cooling (U.S. EPA, 2019, section 13.3.2.2).

5-37

-------
0.5 watts per square meter (Wm'2) (Boucher, 2013), which is slightly reduced compared to AR4.
Estimates of effective radiative forcing24 from aerosol-radiation interactions (ERFari), which
include the rapid feedback effects of temperature and cloud cover, rely mainly on model
simulations, as this forcing is complex and difficult to observe (U.S. EPA, 2019, section

13.3.4.1). The IPCC AR5 best estimate for ERFari is -0.45 ± 0.5 Wm"2, which reflects this
uncertainty (Boucher, 2013).

By providing cloud condensation nuclei, PM increases cloud droplet number, thereby
increasing cloud droplet surface area and albedo (Twomey, 1977). The climate effects of these
perturbations are more difficult to quantify than the direct effects of aerosols with RF but likely
enhance the cooling influence of clouds by increasing cloud reflectivity (traditionally referred to
as the first indirect effect) and lengthening cloud lifetime (the second indirect effect). These
effects are reported as the radiative forcing from aerosol-cloud interactions (RFaci) and the
effective radiative forcing from aerosol-cloud interactions (ERFaci) (U.S. EPA, 2019, section

13.3.3.2). IPCC AR5 estimates ERFaci at -0.45 Wm"2, with a 90% confidence interval of -1.2 to
0 Wm"2 (U.S. EPA, 2019, section 13.3.4.2).25 Studies have also calculated the combined
effective radiative forcing from aerosol-radiation and aerosol-cloud interactions (ERFari+aci)
(U.S. EPA, 2019, section 13.3.4.3). IPCC AR5 reports a best estimate of ERFari+aci of -0.90 (-
1.9 to -0.1) Wm"2, consistent with these estimates (Boucher, 2013).

PM can also strongly reflect incoming solar radiation in areas of high albedo, such as
snow- and ice-covered surfaces. The transport and subsequent deposition of absorbing PM such
as BC to snow- and ice-covered regions can decrease the local surface albedo, leading to surface
heating. The absorbed energy can then melt the snow and ice cover and further depress the
albedo, resulting in a positive feedback loop (U.S. EPA, 2019, section 13.3.3.3; Bond et al.,
2013; U.S. EPA, 2012b). Deposition of absorbing PM, such as BC, may also affect surface
temperatures over glacial regions (U.S. EPA, 2019, section 13.3.3.3). The IPCC AR5 best
estimate of RF from the albedo effect is +0.04 Wm"2, with an uncertainty range of +0.02 to +0.09
Wm"2 (Boucher, 2013).

While research on PM-related effects on climate has expanded since the 2012 review,
there are still significant uncertainties associated with the accurate measurement of PM
contributions to the direct and indirect effects of PM on climate.

24 Effective radiative forcing (ERF), new in the IPCC AR5, takes into account not just the instantaneous forcing but
also a set of climate feedbacks, involving atmospheric temperature, cloud cover, and water vapor, that occur
naturally in response to the initial radiative perturbation (U.S. EPA, 2019, section 13.3.2.2).

25 While the 2019 ISA includes estimates of RFaci and ERFaci from a number of studies (U.S. EPA, 2019, sections
13.3.4.2, 13.3.4.3, 13.3.3.3), this PA focuses on the single best estimate with a range of uncertainty, as reported in
IPCC AR5 (Boucher, 2013).

5-38

-------
• To what extent does the information provide evidence of a quantitative relationship

between specific PM constituents (i.e., BC, OC, sulfate) and climate-related effects?

Since the 2012 review, a number of studies have examined the individual climate effects
associated with key PM components, including sulfate, nitrate, OC, BC, and dust, along with
updated quantitative estimates of the radiative forcing associated with the individual species.

Sulfate particles form through oxidation of SO2 by OH in the gas phase and in the
aqueous phase by a number of pathways, including in particular those involving ozone and H2O2
(U.S. EPA, 2019, section 13.3.5.1). The main source of anthropogenic sulfate is from coal-fired
power plants, and global trends in the anthropogenic SO2 emissions are estimated to have
increased dramatically during the 20th and early 21st centuries, although the recent
implementation of more stringent air pollution controls on sources has led to a reversal in such
trends in many places (U.S. EPA, 2019, section 13.3.5.1). Sulfate particles are highly reflective.
Consistent with other recent estimates, on a global scale, the IPCC AR5 estimates that sulfate
contributes more than other PM types to RF, with RFari of -0.4 (-0.6 to -0.2) Wm'2, where the
5% and 95% uncertainty range is represented by the numbers in the parentheses (Myhre et al.,
2013). This uncertainty range indicates the challenges associated with estimating SO2 from
sources in developing regions and estimating the lifetime of sulfate against wet deposition.

Sulfate is also a major contributor to the influence of PM on clouds (Takemura, 2012). A total
effective radiative forcing (ERFari+aci) for anthropogenic sulfate has been estimated to be nearly
-1.0 Wm"2 (Adams et al., 2001, Zelinka et al., 2014).

Nitrate particles form through the oxidation of nitrogen oxides and occur mainly in the
form of ammonium nitrate. Ammonium preferentially associates with sulfate rather than nitrate,
leading to formation of ammonium sulfate at the expense of ammonium nitrate (Adams et al.,
2001). As anthropogenic emissions of SO2 decline, more ammonium will be available to react
with nitrate, potentially leading to future increases in ammonium nitrate particles in the
atmosphere (U.S. EPA, 2019, section 13.3.5.2; Hauglustaine et al., 2014; Lee et al., 2013;
Shindell et al., 2013). Warmer global temperatures, however, may decrease nitrate abundance
given that it is highly volatile at higher temperatures (Tai et al., 2010). The IPCC AR5 estimates
RFari of nitrate of -0.11 (-0.3 to -0.03) Wm"2 (Boucher, 2013), which is one-fourth of the RFari
of sulfate.

Primary organic carbonaceous PM, including BrC, are emitted from wildfires,
agricultural fires, and fossil fuel and biofuel combustion. Secondary organic aerosols (SOA)
form when anthropogenic or biogenic nonmethane hydrocarbons are oxidized in the atmosphere,
leading to less volatile products that may partition into PM (U.S. EPA, 2019, section 13.3.5.3).
Organic particles are generally reflective, but in the case of BrC, a portion is significantly
absorbing at shorter wavelengths (<400 nm). The IPCC AR5 estimates an RFari for primary

5-39

-------
organic PM from fossil fuel combustion and biofuel use of -0.09 (-0.16 to -0.03) Wm'2 and an
RFari estimate for SOA from these sources of -0.03 (-0.27 to +0.20) Wm"2 (Myhre et al., 2013).
The wide range in these estimates, including inconsistent signs for forcing, reflect uncertainties
in the optical properties of organic PM and its atmospheric budgets, including the production
pathways of anthropogenic SOA (Scott et al., 2014; Myhre et al., 2013; McNeill et al., 2012;
Heald et al., 2010). The IPCC AR5 also estimates an RFari of -0.2 Wm"2 for primary organic PM
arising from biomass burning (Boucher, 2013).

Black carbon (BC) particles occur as a result of inefficient combustion of carbon-
containing fuels. Like directly emitted organic PM, BC is emitted from biofuel and fossil fuel
combustion and by biomass burning. BC is absorbing at all wavelengths and likely has a large
impact on the Earth's energy budget (Bond et al., 2013). The IPCC AR5 estimates a RFari from
anthropogenic fossil fuel and biofuel use of +0.4 (+0.5 to +0.8) Wm"2 (Myhre et al., 2013).
Biomass burning contributes an additional +0.2 (+0.03 to +0.4) Wm"2 to BC RFari, while the
albedo effect of BC on snow and ice adds another +0.04 (+0.02 to +0.09) Wm"2 (Myhre et al.,
2013; U.S. EPA, 2019, section 13.3.5.4, section 13.3.4.4).

Dust, or mineral dust, is mobilized from dry or disturbed soils as a result of both
meteorological and anthropogenic activities. Dust has traditionally been classified as scattering,
but a recent study found that dust may be substantially coarser than currently represented in
climate models, and thus more light-absorbing (Kok et al., 2017). The IPCC AR5 estimates
RFari as -0.1 ± 0.2 Wm"2 (Boucher, 2013), although the results of the study by Kok et al. (2017)
would suggest that in some regions dust may have led to warming, not cooling (U.S. EPA, 2019,
section 13.3.5.5).

Recent research expands upon the evidence from the 2012 review. Consistent with the
evidence in the 2012 review, the key PM components, including sulfate, nitrate, OC, BC, and
dust, that contribute to climate processes vary in their reflectivity, forcing efficiencies, and
direction of forcing.

• To what extent does the evidence change or improve our understanding of the spatial

and temporal variation in climate responses to PM?

Radiative forcing due to PM elicits a number of responses in the climate system that can
lead to significant effects on weather and climate over a range of spatial and temporal scales,
mediated by a number of feedbacks that link PM and climate. Since the 2012 review, the
evidence base has expanded with respect to the mechanisms of climate responses and feedbacks
to PM radiative forcing, described below, although considerable uncertainties continue to exist.
We focus our discussion primarily on the climate impacts in the U.S.

Unlike well-mixed, long-lived greenhouse gases in the atmosphere, PM has a very
heterogenous distribution across the Earth. As such, patterns of RFari and RFaci tend to correlate

5-40

-------
with PM loading, with the greatest forcings centralized over continental regions. The climate
response is more complicated since the perturbation to one climate variable (e.g., temperature,
cloud cover, precipitation) can lead to a cascade of effects on other variables. While the initial
PM radiative forcing may be concentrated regionally, the eventual climate response can be much
broader spatially or be concentrated in remote regions (U.S. EPA, 2019, section 13.3.6). The
complex climate system interactions lead to variation among climate models, with some studies
showing relatively close correlation between forcing and surface response temperatures (e.g.,
Leibensperger et al., 2012), while other studies show much less correlation (e.g., Levy et al.,
2013). Many studies have examined observed trends in PM and temperature in the U.S. Climate
models have suggested a range of factors which can influence large-scale meteorological
processes and may affect temperature, including local feedback effects involving soil moisture
and cloud cover, changes in the hygroscopicity of the PM, and interactions with clouds alone
(U.S. EPA, 2019, section 13.3.7). While evidence described in the 2019 ISA suggests that PM
influenced temperature trends across the southern and eastern U.S. in the 20th century,
uncertainties continue to exist and further research is needed to better characterize the effects of
PM on regional climate in the U.S.

• To what extent have important uncertainties identified in prior reviews been
reduced and/or have new uncertainties emerged?

Since 2009, significant progress has been made in evaluating PM-related climate effects
and uncertainties. The IPCC AR5 states that "climate-relevant aerosol processes are better
understood, and climate-relevant aerosol properties are better observed, than at the time of the
AR4" (Boucher, 2013). However, significant uncertainties remain that make it difficult to
quantify the climate effects of PM. Such uncertainties include those related to our understanding
of:

• The magnitude of PM radiative forcing and the portion of that associated with
anthropogenic emissions;

• The contribution of regional differences in PM concentrations, and of individual
components, to radiative forcing;

• The mechanisms of climate responses and feedbacks resulting from PM-related radiative
forcing; and,

• The process by which PM interacts with clouds and how to represent such interactions in
climate models.

While research has progressed significantly since the 2012 review, substantial
uncertainties still remain with respect to key processes linking PM and climate, because of the
small scale of PM-relevant atmospheric processes compared to the resolution of state-of-the-art
models, and because of the complex cascade of indirect impacts and feedbacks in the climate

5-41

-------
system that result from an initial PM-related radiative perturbation (U.S. EPA, 2019, section
13.3.9).

5.3.2.1.2 Materials Effects

In considering the evidence on materials effects attributable to PM, this section poses the
following policy-relevant question:

• To what extent is information available to link PM to materials effects, including
degradation of surfaces, and deterioration of materials such as metal, stone, concrete
and marble?

In the 2012 review, the 2009 ISA concluded that there was "a causal relationship between
PM and effects on materials" (U.S. EPA, 2009, sections 2.5.4 and 9.5.4). Rather than altering our
conclusions from the 2012 review, the evidence in the 2019 ISA continues to support prior
conclusions regarding materials effects associated with PM deposition. Effects of deposited PM,
particularly sulfates and nitrates,26 to materials include both physical damage and impaired
aesthetic qualities. Because of their electrolytic, hygroscopic, and acidic properties and their
ability to sorb corrosive gases, particles contribute to materials damage by adding to the effects
of natural weathering processes, by potentially promoting or accelerating the corrosion of metals,
degradation of painted surfaces, deterioration of building materials, and weakening of material
components. The majority of the evidence on materials effects of PM are from outside the U.S.
on buildings and other items of cultural heritage; however, they provide limited new data for
consideration. (U.S. EPA, 2019, section 13.4).

Materials damage from PM generally involves one or both of two processes: soiling and
corrosion (U.S. EPA, 2019, section 13.4.2). Soiling and corrosion are complex, interdependent
processes, typically beginning with deposition of atmospheric PM or SO2 to exposed surfaces.
Constituents of deposited PM can interact directly with materials or undergo further chemical
and/or physical transformation to cause soiling, corrosion, and physical damage. Weathering,
including exposure to moisture, ultraviolet (UV) radiation and temperature fluctuations, affects
the rate and degree of damage (U.S. EPA, 2019, section 13.4.2).

Soiling is the result of PM accumulation on an object that alters its optical characteristics
or appearance. These soiling effects can affect the aesthetic value of a structure or result in
reversible or irreversible damage to the surface. The presence of air pollution can increase the
frequency and duration of cleaning and can enhance biodeterioration processes on the surface of
materials. For example, deposition of carbonaceous components of PM can lead to the formation

26 In the case of materials effects, it is difficult to isolate the effects of gaseous and particulate nitrogen and sulfur
wet deposition so both will be considered along with other PM-related deposition effects on materials.

5-42

-------
of black crusts on surfaces, and the buildup of microbial biofilms27 can discolor surfaces by
trapping PM more efficiently (U.S. EPA, 2009, p. 9-195; U.S. EPA, 2019, section 13.4.2). The
presence of PM may alter light transmission or change the reflectivity of a surface. Additionally,
the organic or nutrient content of deposited PM may enhance microbial growth on surfaces.

Since the 2012 review, very little evidence has become available related to deposition of
SO2 to materials such as limestone, granite, and metal. Deposition of SO2 onto limestone can
transform the limestone into gypsum, resulting in a rougher surface, which allows for increased
surface area for accumulation of deposited PM (Camuffo and Bernardi, 1993; U.S. EPA, 2019,
section 13.4.2). Oxidation of deposited SO2 that contributes to the transformation of limestone to
gypsum can be enhanced by the formation of surface coatings from deposited carbonaceous PM
(both elemental and organic carbon) (Grossi et al., 2007, McAlister et al., 2008). Ozga et al.
(2011) characterized damage to two concrete buildings in Poland and Italy. Gypsum was the
main damage product on surfaces of these buildings that were sheltered from rain runoff, while
PM embedded in the concrete, particularly carbonaceous particles, were responsible for
darkening of the building walls (Ozga et al., 2011).

Building on the evidence in the 2009 ISA, research has progressed on the theoretical
understanding of soiling of cultural heritage in a number of studies. Barca et al. (2010)
developed and tested a new methodological approach for characterizing trace elements and
heavy metals in black crusts on stone monuments to identify the origin of the chemicals and the
relationship between the concentrations of elements in the black crusts and local environmental
conditions. Recent research has also used isotope tracers to distinguish between contributions
from local sources versus atmospheric pollution to black crusts on historical monuments in
France (Kloppmann et al., 2011). A study in Portugal found that biological activity played a
major role in soiling, specifically in the development of colored layers and in the detachment
process (de Oliveira et al., 2011). Another study found damage to cement renders, often used for
restoration, consolidation, and decorative purposes on buildings, following exposure to sulfuric
acid, resulting in the formation of gypsum (Lanzon and Garcia-Ruiz, 2010).

Corrosion of stone and the decay of stone building materials by acid deposition and
sulfate salts were described in the 2009 ISA (U.S. EPA, 2009, section 9.5.3). Since that time,
advances have been made on the quantification of degradation rates and further characterization
of the factors that influence damage of stone materials (U.S. EPA, 2019, section 13.4.2). Decay
rates of marble grave stones were found to be greater in heavily polluted areas compared to a
relatively pristine area (Mooers et al., 2016). The time of wetness and the number of

27 Microbial biofilms are communities of microorganisms, which may include bacteria, algae, fungi and lichens, that
colonize an inert surface. Microbial biofilms can contribute to biodeterioration of materials via modification of
the chemical environment.

5-43

-------
dissolution/crystallization cycles were identified as hazard indicators for stone materials, with
greater hazard during the spring and fall when these indicators are relatively high (Casati et al.,
2015).

A study examining the corrosion of steel as a function of PM composition and particle
size found that changes in the composition of resulting rust gradually changed with particle size
(Lau et al., 2008). In a study of damage to metal materials in Hong Kong, which generally has
much higher PM concentrations than those observed in the U.S., Liu et al. (2015) found that iron
and steel were corroded by both PM and gaseous pollutants (SO2 and NO2), while copper and
copper alloys were mainly corroded by gaseous pollutants (SO2 and O3) and aluminum and
aluminum alloy corrosion was mainly attributed to PM and NO2.

A number of studies have also found materials damage from PM components besides
sulfate and black carbon and atmospheric gases besides SO2. Studies have characterized impacts
of nitrates, NOx, and organic compounds on direct materials damage or on chemical reactions
that enhance materials damage (U.S. EPA, 2019, section 13.4.2). Other studies have found that
soiling of building materials can be attributed to enhanced biological processes and colonization,
including the development and thickening of biofilms, resulting from the deposition of PM
components and atmospheric gases (U.S. EPA, 2019, section 13.4.2).

Since the 2012 review, other materials have been studied for damage attributable to PM,
including glass and photovoltaic panels. Soiling of glass can affect its optical and thermal
properties, and can lead to increased cleaning costs and frequency. The development of haze28 on
modern glass has been measured and modeled, with a strong correlation between the size
distribution of particles and the evolution of the mass deposited on the surface of the glass.
Measurements showed that, under sheltered conditions, mass deposition accelerated regularly
with time in areas closest to sources of PM (i.e., near roadways) and coarse mineral particles
were more prevalent compared to other sites (Alfaro et al., 2012). Model predictions were found
to correctly simulate the development of haze at site locations when compared with
measurements (Alfaro et al., 2012).

Soiling of photovoltaic panels can lead to decreased energy efficiency. For example,
soiling by carbonaceous PM decreased solar efficiency by nearly 38%, while soil particles
reduced efficiency by almost 70% (Radonjic et al., 2017). The rate of photovoltaic power output
can also be degraded by soiling and has been found to be related to the rate of dust accumulation.

28 In this discussion of non-visibility welfare effects (section 5.3.2), haze is used as it has been defined in the
scientific literature on soiling of glass, i.e., the ratio of diffuse transmitted light to direct transmitted light
(Lombardo et al., 2010). This differs from the definition of haze as used in the discussion of visibility welfare
effects in section 5.3.1, where it is used as a qualitative description of the blockage of sunlight by dust, smoke,
and pollution.

5-44

-------
In five sites in the U.S. representing different meteorological and climatological conditions,29
photovoltaic module power transmission was reduced by approximately 3% for every g/m2 of
PM deposited on the cover plate of the photovoltaic panel, independent of geographical location
(Boyle et al., 2017). Another study found that photovoltaic module power output was reduced by
40% after 10 months of exposure without cleaning, although a number of anti-reflective coatings
can generally mitigate power reduction resulting from dust deposition (Walwil et al., 2017).
Energy efficiency can also be impacted by the soiling of building materials, such as light-colored
marble panels on building exteriors, that are used to reflect a large portion of solar radiation for
passive cooling and to counter the urban heat island effect. Exposure to acidic pollutants in urban
environments have been found to reduce the solar reflectance of marble, decreasing the cooling
effect (Rosso et al., 2016). Highly reflective roofs, or cool roofs, have been designed and
constructed to increase reflectance from buildings in urban areas, to both decrease air
conditioning needs and urban heat island effects, but these efforts can be impeded by soiling of
materials used for constructing cool roofs. Methods have been developed for accelerating the
aging process of roofing materials to better characterize the impact of soiling and natural weather
on materials used in constructing cool roofs (Sleiman et al., 2014).

• To what extent has information emerged for quantifying material damage

attributable to PM through dose-response relationships or damage functions? Are
there studies linking perceptions of reduced aesthetic appeal of buildings and other
objects to PM or wet deposition of nitrogen and sulfur species?

Some progress has been made since the 2012 review in the development of dose-response
relationships for soiling of building materials, although some key relationships remain poorly
characterized. The first general dose-response relationships for soiling of materials were
generated by measuring contrast reflectance of a soiled surface to the reflectance of the unsoiled
substrate for different materials, including acrylic house paint, cedar siding, concrete, brick,
limestone, asphalt shingles, and window glass with varying total suspended particulate (TSP)
concentrations (Beloin and Haynie, 1975; U.S. EPA, 2019, section 13.4.3). Continued efforts to
develop dose-response curves for soiling have led to some advancements for modern materials,
but these relationships remain poorly characterized for limestone. One study quantified the dose-
response relationships between PMio and soiling for painted steel, white plastic, and
polycarbonate filter material, but there was too much scatter in the data to produce a dose-
response relationship for limestone (Watt et al., 2008). A dose-response relationship for silica-

29 Of the five sites studied, three were in rural, suburban, and urban areas representing a semi-arid environment
(Front Range of Colorado), one site represented a hot and humid environment (Cocoa, Florida), and one
represented a hot and arid environment (Albuquerque, New Mexico) (U.S. EPA, 2019, section 13.4.2; Boyle et
al., 2017).

5-45

-------
soda-lime window glass soiling by PMio, NO2, and SO2 was quantified based on 31 different
locations (Lombardo et al., 2010; U.S. EPA, 2019, section 13.4.3, Figure 13-32, Equation 13-8).
The development of this dose-response relationship required several years of observation time
and had inconsistent data reporting across the locations.

Since the 2012 review, there has also been progress in developing methods to more
rapidly evaluate soiling of different materials by PM mixtures. Modern buildings typically have
simpler lines, less detailed surfaces, and a greater use of glass, tile, and metal, which are easier to
clean than stone. There have also been major changes in the types of materials used for
buildings, including a variety of polymers available for use as coatings and sealants. New
economic and environmental considerations beyond aesthetic appeal and structural damage are
emerging (U.S. EPA, 2019, section 13.4.3). Changes in building materials and design, coupled
with new approaches in quantifying the dose-response relationship between PM and materials
effects, may reduce the amount of time needed for observations to support the development of
material-specific dose-response relationships.

In addition to dose-response functions, damage functions have also been used to quantify
material decay as a function of pollutant type and load. Damage can be determined from sample
surveys or inspection of actual damage and a damage function can be developed to link the rate
of material damage to time of replacement or maintenance. A cost function can then link the time
for replacement and maintenance to a monetary cost, and an economic function links cost to the
dose of pollution based on the dose-response relationship (U.S. EPA, 2019, section 13.4.3).
Damage functions are difficult to assess because it depends on human perception of the level of
soiling deemed to be acceptable and evidence in this area remains limited. As described in the
2019 ISA, damage functions for a wide range of building materials (i.e., stone, aluminum, zinc,
copper, plastic, paint, rubber, stone) have been developed and reviewed (Brimblecombe and
Grossi, 2010). One study estimated long-term deterioration of building materials and found that
damage to durable building material (such as limestone, iron, copper, and discoloration of stone)
is no longer controlled by pollution as was historically documented but rather that natural
weathering is a more important influence on these materials in modern times (Brimblecombe and
Grossi, 2009). Even as PM-attributable damage to stone and metals has decreased over time, it
has been predicted that there will be potentially higher degradation rates for polymeric materials,
plastic, paint, and rubber due to increased oxidant concentrations and solar radiation
(Brimblecombe and Grossi, 2009).

• To what extent have important uncertainties identified in prior reviews been

reduced and/or have new uncertainties emerged?

While there are a number of studies in the 2019 ISA that investigate the effect of PM on
newly studied materials and further characterize the effects of PM on previously studied

5-46

-------
materials, there remains insufficient evidence to relate soiling or damage to specific PM levels or
to establish a quantitative relationship between PM in ambient air and materials degradation.
Uncertainties that were identified in the 2012 review still largely remain with respect to
quantitative relationships between particle size, concentration, chemical concentrations, and
frequency of repainting and repair. No new studies are assessed in the 2019 ISA that link
perceptions of reduced aesthetic appeal of buildings and other objects to PM-related materials
effects. Moreover, uncertainties about the deposition rates of airborne PM to surfaces and the
interaction of co-pollutants still remain.

5.3.2.2 Quantitative Information

Beyond our consideration of the scientific evidence, discussed above in section 5.3.2.1
above, we also consider the extent to which quantitative analyses of PM air quality and
quantitative assessments for climate and materials effects could inform conclusions on the
adequacy of the public welfare protection provided by the current secondary PM standards. We
have evaluated the potential support for conducting new analyses of PM air quality
concentrations and non-visibility welfare effects.

5.3.2.2.1 Climate Effects

While expanded since the 2012 review, our current understanding of PM-related climate
effects is still limited by significant uncertainties. Large spatial and temporal heterogeneities in
direct and indirect PM climate forcing can occur for a number of reasons, including the
frequency and distribution of emissions of key PM components contributing to climate forcing,
the chemical and microphysical processing that occurs in the atmosphere, and the atmospheric
lifetime of PM relative to other pollutants contributing to climate forcing (U.S. EPA, 2019,
section 13.3). These issues particularly introduce uncertainty at the local and regional scales in
the U.S. that would likely be most relevant to a quantitative assessment of the potential effects of
a national PM standard on climate in this review. Limitations and uncertainties in the evidence
make it difficult to quantify the impact of PM on climate and in particular how changes in the
level of PM mass in ambient air would result in changes to climate in the U.S. Thus, as in the
2012 review, the data remain insufficient to conduct quantitative analyses for PM effects on
climate.

5.3.2.2.2 Materials Effects

As at the time of the 2012 review, sufficient evidence is not available to conduct a
quantitative assessment of PM-related soiling and corrosion effects. While soiling associated
with PM can lead to increased cleaning frequency and repainting of surfaces, no quantitative
relationships have been established between characteristics of PM or the frequency of cleaning
or repainting that would help inform our understanding of the public welfare implications of

5-47

-------
soiling (U.S. EPA, 2019, section 13.4). Similarly, while some information is available with
regard to microbial deterioration of surfaces and the contribution of carbonaceous PM to the
formation of black crusts that contribute to soiling, the available evidence does not support
quantitative analyses (U.S. EPA, 2019, section 13.4). While some evidence is available with
respect to PM-attributable materials effects, the data are insufficient to conduct quantitative
analyses for PM effects on materials.

5.4 CASAC ADVICE AND PUBLIC COMMENTS

As part of its review of the draft PA, the CASAC has provided advice on the adequacy of
the current PM standards. In its comments on the draft PA, the CASAC first recognized the
scientific evidence is sufficient to support a causal relationship between PM and visibility
effects, climate effects and materials effects.

With regard to visibility effects, the CASAC recognized that the identification of a target
level of protection for the visibility index is based on a limited number of studies and suggest
that "additional region- and view-specific visibility preference studies and data analyses are need
to support a more refined visibility target" (Sheppard, 2022, p. 21 of consensus responses).

While the CASAC did not recommend revising either the target level of protection for the
visibility index or the level of the current 24-hour PM2.5 standard, they did state that "[i]f a
value of 20-25 deciviews is deemed to be an appropriate visibility target level of protection, then
a secondary 24-hour PM2.5 standard in the range of 25-35 |ig/m3 should be considered"
(Sheppard, 2022, p. 21 of consensus responses).

The CASAC also recognized the limited availability of monitoring methods and networks
for directly measuring light extinction. As such, they suggest that "[a] more extensive technical
evaluation of the alternatives for visibility indicators and practical measurement methods
(including the necessity for a visibility FRM) is need for future reviews" (Sheppard, 2022, p. 22
of consensus letter). The majority of the CASAC "recommend[ed] that an FRM for a directly
measured PM2.5 light extinction indicator be developed" to inform the consideration of the
protection afforded by the secondary PM standards against visibility impairment, the minority of
the CASAC "believe that a light extinction FRM is not necessary to set a secondary standard
protective of visibility" (Sheppard, 2022, p. 22 of consensus responses).

With regard to climate and materials effects, the CASAC noted that substantial
uncertainties remain in the scientific evidence for these effects. The CASAC suggested a number
of areas for future research to further inform our understanding of these effects, including
research that would allow for quantitative assessment of the relationship between climate and
materials effects and PM in ambient air.

5-48

-------
We also received a limited number of public comments on the draft PA focused on the
secondary PM standards. Of those who provided comments on the secondary PM standards, the
majority of commenters support the preliminary conclusion that it is appropriate to consider
retaining the current secondary PM standards, without revision. These commenters generally cite
to a lack of newly available evidence and information that would inform consideration of
alternative secondary PM standards to protect against PM-related effects on visibility, climate,
and materials. One commenter, however, supported the revision of the secondary PM standards
to provide additional protection against PM-related visibility effects.

5.5 CONCLUSIONS REGARDING THE ADEQUACY OF THE
SECONDARY PM STANDARDS

This section discusses staff conclusions for the Administrator's consideration in judging
the adequacy of the current secondary PM standards. These conclusions are based on
consideration of the assessment and integrative synthesis of evidence presented in the 2019 ISA
and ISA Supplement, as well as analyses of recent air quality, CASAC advice and
recommendations, and public comments received on the draft PA. Taking into consideration the
responses to specific questions discussed above, we revisit the overarching policy question for
this chapter:

• Does the scientific evidence and quantitative information support or call into
question the adequacy of the protection afforded by the current secondary PM
standards?

As provided in section 109(b)(2) of the CAA, the secondary standard is to "specify a
level of air quality the attainment and maintenance of which in the judgment of the
Administrator.. .is requisite to protect public welfare from any known or anticipated adverse
effects associated with the presence of such air pollutant in the ambient air." Effects on welfare
include, but are not limited to, "effects on soils, water, crops, vegetation, man-made materials,
animals, wildlife, weather, visibility, and climate, damage to and deterioration of property, and
hazards to transportation, as well as effects on economic values and on personal comfort and
well-being" (CAA section 302(h)). The secondary standards are not meant to protect against all
known or anticipated PM-related effects, but rather those that are judged to be adverse to the
public welfare (78 FR 3212, January 15, 2013). Similarly, the extent to which secondary
standards are concluded to provide adequate protection from such effects also depends on
judgments by the Administrator.

Therefore, we recognize that, as is the case in NAAQS reviews in general, the extent to
which the current secondary PM standards are judged to be adequate will depend on a variety of
factors and judgments to be made by the Administrator. Such judgments include those

5-49

-------
concerning the extent or severity of welfare effects that may be considered adverse to the public
welfare, and accordingly, what level of protection from such known or anticipated effects may be
judged requisite. In general, the public welfare significance of PM-related effects for different air
quality conditions and in different locations depend upon the type and severity of the effects, as
well as the strength of the underlying information and associated uncertainties. Thus, in the
discussion below, our intention is to focus on such aspects of the evidence and quantitative
analyses.

With regard to visibility, climate, and materials effects of PM, our response to the
question above takes into consideration the discussions that address the specific policy-relevant
questions in prior sections of this chapter (see sections 5.3.1 and 5.3.2) and the approach
described in section 5.2 that builds on the approach from previous reviews. With respect to the
evidence-based considerations, we note that the evidence, while somewhat expanded since
previous reviews, does not include evidence of effects at lower concentrations or other welfare
effects of PM than those identified at the time of prior reviews. There continue to be significant
uncertainties related to quantifying the relationships between PM mass concentrations in ambient
air and welfare effects, including visibility impairment, climate effects, and materials effects.

With respect to the visibility effects of PM, the evidence continues to support a causal
relationship. With respect to evidence for visibility effects of PM, we note that the evidence,
while somewhat expanded since the 2012 review, does not include evidence of effects at lower
concentrations than those identified at the time of the 2012 review. Consistent with the evidence
available at the time of the 2012 review, significant limitations remain in directly measuring light
extinction. However, a number of small refinements have been made to the algorithm commonly
used to estimate light extinction (U.S. EPA, 2019, section 13.2.3.3; section 5.3.1.1 above). Light
extinction by PM2.5 is dependent on PM2.5 composition and relative humidity, which varies
regionally, with component contributions to light extinction also changing over time with
changes in emissions, as can be seen in analyses of recent air quality. We also note that limited
new research is available on methods of characterizing visibility or on how visibility is valued by
the public, such as visibility preference studies. Thus, while limited new research has further
informed our understanding of the influence of atmospheric components of PM2.5 on light
extinction, the available evidence to inform consideration of the public welfare implications of
PM-related visibility impairment remains relatively unchanged.

With respect to quantitative-based considerations, analyses using recent air quality and
considering updated and alternative methods for estimating visibility impairment provide results
generally similar to those given a focus in the decision for the 2012 and 2020 reviews. We
recognize that conclusions reached regarding visibility in previous reviews were based primarily
on the quantitative analyses that considered the relationship of estimated visibility impairment

5-50

-------
(light extinction) with design values for the secondary 24-hour PM2.5 standard. These analyses
demonstrated that visibility index values were below 30 dv - the value identified as the target
level of protection for visibility-related welfare effects - at all locations that met the daily
standard. In our evaluation in this chapter, we have considered the information regarding the
equations to estimate light extinction and the inputs to the equations and regarding identification
of the target level of protection. With regard to the equations, consistent with the approach in the
2020 review, we have utilized both the most recently published equations as well as alternatives
considered in the 2012 review in recognition of the uncertainties inherent in the quantitative
relationship between PM and light extinction and the variability in applicability to different
locations. Further, we have considered key coefficients in estimating and adjusting
concentrations of specific PM2.5 components, a key example of which is the multiplier used to
estimate the concentration of organic matter from the concentration of organic carbon. Given the
lack of new information and consistent with the analyses on which the decisions were based in
the 2012 and 2020 reviews, we have focused on a 3-year average of the 90th percentile of daily
light extinction (calculated using old and new algorithms) in considering visibility impairment at
the analyzed locations.

In reaching a conclusion in the 2012 and 2020 reviews with regard to the adequacy of
visibility protection provided by the secondary PM standards, both Administrators identified 30
dv as an appropriate target level of protection. We have not identified new information available
since the completion of the 2020 review in this reconsideration of the 2020 final decision that
would challenge this public policy. Thus, in our consideration of the current information and
analyses in this document, we have compared the results of the updated analyses to the value of
30 dv, finding that all sites meet this target level of protection while also meeting the current
daily standards. In so finding, we additionally note the uncertainties recognized above regarding
estimation of OM for use in the IMPROVE equations, and also the variability across sites in
characteristics that affect the relationship between PM in ambient air and light extinction, and in
characteristics that affect human visibility and preferences in that regard. Based on the findings
of this comparison, in light of all of these considerations, we find it reasonable to conclude that
the quantitative information available in this reconsideration of the 2020 final decision does not
call into question the adequacy of visibility-related public welfare protection provided by the
current secondary PM standards. Given this conclusion, we have not conducted additional
analyses to evaluate the level of visibility protection that might be afforded by potential
alternative standards.

With respect to the non-visibility welfare effects of PM, the available evidence continues
to support causal relationships between climate effects and PM and materials effects and PM.
The evidence related to climate effects and PM, while expanded since previous reviews, has not

5-51

-------
appreciably improved our understanding of the spatial and temporal heterogeneity of PM
components that contribute to climate forcing. We note that, as at the time of the 2012 review,
the evidence describes differences among individual PM components in their reflective
properties and direction of climate forcing. We also note that, while climate research has
continued, there are still significant limitations in our ability to quantify contributions of PM, and
of individual PM components, to the direct and indirect effects of PM on climate (e.g. changes to
the pattern of rainfall, changes to wind patterns, effects on vertical mixing in the atmosphere).
While climate models have been improved and refined since the 2012 review, climate models
simulating aerosol-climate interactions on regional scales (e.g., -100 km) tend to have more
variability in estimates of the PM-related climate effects than simulations at the global scale, and
fewer studies are available that simulate specific regions (e.g., the U.S.) than that provide global-
scale simulations. While recent research has added to the understanding of climate forcing on a
global scale, there remain significant limitations to quantifying potential adverse effects from
PM on climate in the U.S. and how they would vary in response to changes in PM concentrations
in the U.S. That is, the information with regard to climate does not provide a clear understanding
of a quantitative relationship between concentrations of PM mass in ambient air and associated
climate-related effects, and consequently, precludes a quantitative evaluation of the level of
protection provided by a PM concentration-based secondary standard from adverse climate-
related effects on the public welfare in the U.S. Thus, on the whole, we do not find the
information to provide support for different conclusions than were reached in the 2012 and 2020
reviews with regard to climate-related effects of PM in ambient air.

In considering the evidence related to materials effects and PM, we note that there is
some evidence that informs our understanding on the soiling process and types of materials
affected, and provides limited information on dose-response relationships and damage functions,
although most of the recent evidence comes from studies outside of the U.S. In particular, there
is a growing body of research on PM and energy efficiency-related materials, such as solar
panels and passive cooling building materials, affecting the optical and thermal properties,
thereby impacting the intended energy efficiency of these materials. While recent research has
added to the understanding of PM-related materials effects, there remains a lack of research
related to quantifying materials effects and understanding the public welfare implications of such
effects.

In summary, with regard to the PM-related non-visibility effects - climate effects and
materials effects - the available evidence, as in previous reviews, documents a causal role for
PM in ambient air. This evidence, however, as in the 2012 and 2020 reviews, also includes
substantial uncertainties with regard to quantitative relationships with PM concentrations and
concentration patterns that limit our ability to quantitatively assess the public welfare protection

5-52

-------
provided by the standards from these effects. Thus, as a whole, the available information does
not call into question the adequacy of protection provided by the current standards for these
effects.

Based on all of the above considerations, we find that the available evidence does not call
into question the protection afforded by the current secondary PM standards against PM-related
welfare effects. Thus, our conclusion for the Administrator's consideration is that it is
appropriate to consider retaining the current secondary PM standards, without revision. In so
concluding, we recognize, as noted above, that the final decision on this reconsideration of the
secondary PM standards to be made by the Administrator is largely a public welfare judgment,
based on his judgment as to the requisite protection of the public welfare from any known or
anticipated adverse effects. This final decision will draw upon the available scientific evidence
and quantitative analyses on PM-attributable welfare effects, and on judgments about the
appropriate weight to place on the range of uncertainties inherent in the evidence and analyses.

5.6 AREAS FOR FUTURE RESEARCH AND DATA COLLECTION

In this section, we highlight key uncertainties in the available information related to the
effects of PM on public welfare. Such key uncertainties and areas for future research, model
development, and data gathering are outlined below. We note, however, that a full set of research
recommendations is beyond the scope of this discussion. Rather, listed below are key
uncertainties, research questions and data gaps that have been thus far highlighted in this review
of the secondary PM standards.

• A critical aspect of our consideration of the evidence and quantitative information for
visibility impairment is our understanding of human perception of visibility impairment
in the preference studies. This is essential to the Administrator's consideration of the
public welfare implications of visibility effects and to decisions on the adequacy of
protection provided by the secondary PM standards from them. Additional information
related to several areas would reduce uncertainty in our interpretation of the available
information for purposes of characterizing visibility impairment. These areas include the
following:

- Expanding the number and geographic coverage of preference studies in urban,
rural and Class I areas to account for the potential for people to have different
preferences based on the conditions that they commonly encounter and potential
differences in preferences based on the scene types;

- Evaluating visibility preferences of the U.S. population today, given that the
preference studies were conducted more than 15 years ago, during which time air
quality in the U.S. has improved;

- Accounting for the influence that varying study methods may have on an
individual's response as to what level of visibility impairment is acceptable; and

5-53

-------
- Providing insights regarding people's judgments on acceptable visibility based on
those factors that can influence an individual's perception of visibility
impairment, including the duration of visibility impairment experiences, the time
of day during which light extinction is greatest, and the frequency of episodes of
visibility impairment, as well as the intensity of the visibility impairment.

The development and implementation of direct monitoring of PM2.5 light extinction would
help to characterize visibility and the relationships between PM component
concentrations and light extinction and to evaluate and refine light extinction calculation
algorithms for use in areas near anthropogenic sources, and would provide measurements
for future visibility effects assessments.

Improvements to the estimation of light extinction using various IMPROVE equations is
evolving over time. Additional evaluation of the influence of changes in PM components
over time, as well as changes in the sources of PM, would inform our understanding of
regional differences in estimated light extinction, and the appropriateness of different
IMPROVE equations in different regions. Further, the evaluation and validation of
alternative methods for estimating light extinction, such as those that combine ambient
air quality and meteorological measurements, could improve our understanding of the
factors that influence spatial and temporal variability in light scattering and absorption
efficiencies.

Increased availability of monitoring data for coarse PM could further inform our
understanding of the effects and importance of coarse PM on light extinction compared to
PM2.5, particularly in areas that are more likely to be impacted by the coarse fraction such
as the Southwest. Generally, monitoring data is lacking in these areas, limiting the ability
to quantitatively evaluate the role of coarse PM in light extinction in these areas.

Substantial uncertainties still remain with respect to key processes linking PM and
climate, because of the small scale of PM-relevant atmospheric processes compared to
the resolution of state-of-the-art models, and because of the complex cascade of indirect
impacts and feedbacks in the climate system that result from an initial PM-related
radiative perturbation. Such uncertainties include those related to our understanding of:

- The magnitude of PM radiative forcing and the portion of that associated with
anthropogenic emissions;

- The contribution of regional differences in PM concentrations, and of individual
components, to radiative forcing; and,

- The process by which PM interacts with clouds and how to represent such
interactions in climate models.

Research on more accurate U.S. and global emission inventories would provide source-
specific data on PM and PM component contributions to climate effects, particularly
those effects resulting from climate forcing.

Insufficient evidence is available to relate soiling or damage to specific PM
concentrations or to establish a quantitative relationship between PM concentrations in
ambient air and materials degradation. Additional information would reduce uncertainty
in our interpretation of the available information, including in the following areas:

5-54

-------
- Identifying quantitative relationships between particle size, PM concentration,
chemical concentrations, and frequency of repainting and repair;

- Understanding human perceptions of reduced aesthetic appeal of buildings, and
other objects to PM-related materials effects; and

- Characterizing deposition rates of airborne PM to surfaces and the interaction of
co-pollutants.

• Research on combining ground-based measurements with satellite observations to
establish dose-response relationships to provide additional information regarding the
corrosion and/or soiling effects of PM on various materials.

5-55

-------
REFERENCES

Abt Associates, Inc. (2001). Assessing public opinions on visibility impairment due to air

pollution: Summary report. U.S. Environmental Protection Agency. Research Triangle
Park, NC.

Adams, PJ, Seinfeld, JH, Koch, D, Mickley, L and Jacob, D (2001). General circulation model
assessment of direct radiative forcing by the sulfate-nitrate-ammonium-water inorganic
aerosol system. Journal of Geophysical Research 106(D1): 1097-1111.

Alfaro, SC, Chabas, A, Lombardo, T, Verney-Carron, A and Ausset, P (2012). Predicting the
soiling of modern glass in urban environments: A new physically-based model.
Atmospheric Environment 60: 348-357.

Ban-Weiss, GA, Jin, L, Bauer, SE, Bennartz, R, Liu, X, Zhang, K, Ming, Yi, Guo, H and Jiang,
JH (2014). Evaluating clouds, aerosols, and their interactions in three global climate
models using satellite simulators and observations. Journal of Geophysical Research:
Atmospheres 119(18): 10876-10901.

Barca, D, Belfiore, CM, Crisci, GM, La Russa, MF, Pezzino, A and Ruffolo, SA (2010).

Application of laser ablation ICP-MS and traditional techniques to the study of black
crusts on building stones: a new methodological approach. Environmental Science and
Pollution Research 17(8): 1433-1447.

BBC Research & Consulting (2003). Phoenix area visibility survey. Denver, CO.

http ://www. azdeq. gov/environ/air/dowmload/vi s 021903f. pdf.

Beloin, NJ and Haynie, FH (1975). Soiling of building materials. Journal of the Air and Waste
Management Association 25(4): 399-403.

Bond, TC, Doherty, SJ, Fahey, DW, Forster, PM, Berntsen, T, Deangelo, BJ, Flanner, MG,

Ghan, S, Karcher, B, Koch, D, Kinne, S, Kondo, Y, Quinn, PK, Sarofim, MC, Schultz,
MG, Schulz, M, Venkataraman, C, Zhang, H, Zhang, S, Bellouin, N, Guttikunda, SK,
Hopke, PK, Jacobson, MZ, Kaiser, JW, Klimont, Z, Lohmann, U, Schwarz, JP, Shindell,
D, Storelvmo, T, Warren, SG and Zender, CS (2013). Bounding the role of black carbon
in the climate system: A scientific assessment. Journal of Geophysical Research:
Atmospheres 118(11): 5380-5552.

Boucher, O (2013). Clouds and Aerosols. In Climate Change 2013: The Physical Science Basis.
Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental
Panel on Climate Change. Cambridge University Press. Cambridge, United Kingdom
and New York, NY, USA.

Boyle, L, Burton, PD, Danner, V, Hannigan, MP and King, B (2017). Regional and national
scale spatial variability of photovoltaic cover plate soiling and subsequent solar
transmission losses. 7(5): 1354-1361.

5-56

-------
Brimblecombe, P and Grossi, CM (2009). Millennium-long damage to building materials in
London. Science of the Total Environment 407(4): 1354-1361.

Brimblecombe, P and Grossi, CM (2010). Potential damage to modern building materials from
21st century air pollution. The Scientific World Journal 10: 116-125.

Camuffo, D and Bernardi, A (1993). Microclimatic factors affecting the trajan column. Science
of the Total Environment 128(2-3): 227-255.

Casati, M, Rovelli, G, D'Angelo, L, Perrone, MG, Sangiorgi, G, Bolzacchini, E and Ferrero, L
(2015). Experimental measurements of particulate matter deliquescence and
crystallization relative humidity: Application in heritage climatology. Aerosol and Air
Quality Research 15(2): 399-409.

de Oliveira, BP, de la Rosa, JM, Miller, AZ, Saiz-Jimenez, C, Gomez-Bolea, A, Braga, MAS and
Dionisio, A (2011). An integrated approach to assess the origins of black films on a
granite monument. Environmental Earth Sciences 63(6-7 SI): 1677-1690.

Ely, DW, Leary, JT, Stewart, TR and Ross, DM (1991). The establishment of the Denver
Visibility Standard. Colorado Department of Health. Denver, Colorado.

Fajardo, OA, Jiang, J and Hao, J (2013). Assessing young people's preferences in urban visibility
in beijing. Aerosol and Air Quality Research 13(5): 1536-1543.

Fiore, AM, Naik, V and Leibensperger, EM (2015). Air quality and climate connections. Journal
of the Air and Waste Management Association 65(6): 645-685.

Frank, N. (2012). Memorandum to PMNAAQS Review Docket (EPA-HQ-OAR-2007-0492)
regarding the Differences between maximum and composite monitor annual PM2.5
design values by CBSA. Dec 14, 2012. Docket ID No. EPA-HQ-OAR-2007-0492. Office
of Air Quality Planning and Standards Research Triangle Park, NC. Available at:

http s: //www, regul ati on s. gov/docum ent/EP A.~H.( 37-0492-10099.

Grossi, CM, Brimblecombe, P, Esbert, RM and Javier Alonso, F (2007). Color changes in

architectural limestones from pollution and cleaning. Color Research and Application
32(4): 320-331.

Hand, JL, Copeland, SA, Day, DA, Dillner, AM, Indresand, H, Malm, WC, McDade, CE,

Moore, CT, Jr., Pitchford, ML, Schichtel, BA and Watson, JG (2011). Spatial and
seasonal patterns and temporal variability of haze and its constituents in the United
States, IMPROVE Report V. Colorado State University. Fort Collins, CO.

Hand, JL, Prenni, AJ, Copeland, S, Schichtel, BA and Malm, WC (2020). Thirty years of the
Clean Air Act Amendments: Impacts on haze in remote regions of the United States
(1990-2018). Atmospheric Environment 243: 117865.

5-57

-------
Hauglustaine, DA, Balkanski, Y and Schulz, M (2014). A global model simulation of present
and future nitrate aerosols and their direct radiative forcing of climate. Atmospheric
Chemistry and Physics 14(20): 11031-11063.

Heald, CL, Ridley, DA, Kreidenweis, SM and Drury, EE (2010). Satellite observations cap the
atmospheric organic aerosol budget. Geophysical Research Letters 37.

IPCC (2013). Climate change 2013: The physical science basis. Contribution of working group I
to the fifth assessment report of the Intergovernmental Panel on Climate Change. T. F.
Stocker, D. Qin, G. K. Plattner et al. Cambridge University Press. Cambridge, UK.

Jimenez, JL, Canagaratna, MR, Donahue, NM, Prevot, AS, Zhang, Q, Kroll, JH, Decarlo, PF,
Allan, JD, Coe, H, Ng, NL, Aiken, AC, Docherty, KS, Ulbrich, IM, Grieshop, AP,
Robinson, AL, Duplissy, J, Smith, JD, Wilson, KR, Lanz, VA, Hueglin, C, Sun, YL,

Tian, J, Laaksonen, A, Raatikainen, T, Rautiainen, J, Vaattovaara, P, Ehn, M, Kulmala,
M, Tomlinson, JM, Collins, DR, Cubison, MJ, Dunlea, EJ, Huffman, JA, Onasch, TB,
Alfarra, MR, Williams, PI, Bower, K, Kondo, Y, Schneider, J, Drewnick, F, Borrmann,
S, Weimer, S, Demeijian, K, Salcedo, D, Cottrell, L, Griffin, R, Takami, A, Miyoshi, T,
Hatakeyama, S, Shimono, A, Sun, JY, Zhang, YM, Dzepina, K, Kimmel, JR, Sueper, D,
Jayne, JT, Herndon, SC, Trimborn, AM, Williams, LR, Wood, EC, Middlebrook, AM,
Kolb, CE, Baltensperger, U and Worsnop, DR (2009). Evolution of organic aerosols in
the atmosphere. Science 326(5959): 1525-1529.

Kelly, J, Schmidt, M, Frank, N, Timin, B, Solomon, D and Venkatesh, R. (2012). Memorandum
to PM NAAQS Review Docket (EPA-HQ-OAR-2007-0492). Technical Analyses to
Support Surrogacy Policy for Proposed Secondary PM2.5 NAAQS under NSR/PSD
Programs. June 14, 2012. Docket ID No. EPA-HQ-OAR-2007-0492. Office of Air
Quality Planning and Standards Research Triangle Park, NC. Available at:
https://www3.epa. gov/ttn/naaas/standards/pm/data/20120614Kellv.pdf.

Kleinman, LI, Sedlacek Iii, AJ, Adachi, K, Buseck, PR, Collier, S, Dubey, MK, Hodshire, AL,
Lewis, E, Onasch, TB, Pierce, JR, Shilling, J, Springston, SR, Wang, J, Zhang, Q, Zhou,
S and Yokelson, RJ (2020). Rapid evolution of aerosol particles and their optical
properties downwind of wildfires in the western US. Atmosheric Chemistry and Physics
20(21): 13319-13341.

Kloppmann, W, Bromblet, P, Vallet, JM, Verges-Belmin, V, Rolland, O, Guerrot, C and

Gosselin, C (2011). Building materials as intrinsic sources of sulphate: A hidden face of
salt weathering of historical monuments investigated through multi-isotope tracing (B, O,
S). Science of the Total Environment 409(9): 1658-1669.

Kok, JF, Ridley, DA, Zhou, Q, Miller, RL, Zhao, C, Heald, CL, Ward, DS, Albani, S and

Haustein, K (2017). Smaller desert dust cooling effect estimated from analysis of dust
size and abundance. Nature Geoscience 10(4): 274-278.

Laing, JR, Jaffe, DA and 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. Atmospheric Chemistry and Physics 16(23): 15185-15197.

5-58

-------
Lanzon, M and Garcia-Ruiz, PA (2010). Deterioration and damage evaluation of rendering
mortars exposed to sulphuric acid. Materials and Structures 43(3): 417-427.

Lau, NT, Chan, CK, Chan, L and Fang, M (2008). A microscopic study of the effects of particle
size and composition of atmospheric aerosols on the corrosion of mild steel. Corrosion
Science 50(10): 2927-2933.

Lee, YH, Lamarque, JF, Flanner, MG, Jiao, C, Shindell, DT, Berntsen, T, Bisiaux, MM, Cao, J,
Collins, WJ, Curran, M, Edwards, R, Faluvegi, G, Ghan, S, Horowitz, LW, McConnell,
JR, Ming, J, Myhre, G, Nagashima, T, Naik, V, Rumbold, ST, Skeie, RB, Sudo, K,
Takemura, T, Thevenon, F, Xu, B and Yoon, JH (2013). Evaluation of preindustrial to
present-day black carbon and its albedo forcing from Atmospheric Chemistry and
Climate Model Intercomparison Project (ACCMIP). Atmospheric Chemistry and Physics
13(5): 2607-2634.

Leibensperger, EM, Mickley, LJ, Jacob, DJ, Chen, WT, Seinfeld, JH, Nenes, A, Adams, PJ,

Streets, DG, Kumar, N and Rind, D (2012). Climatic effects of 1950-2050 changes in US
anthropogenic aerosols - Part 2: Climate response. Atmospheric Chemistry and Physics
12(7): 3349-3362.

Levy, H, Horowitz, LW, Schwarzkopf, MD, Ming, Yi, Golaz, JC, Naik, V and Ramaswamy, V
(2013). The roles of aerosol direct and indirect effects in past and future climate change.
Journal of Geophysical Research: Atmospheres 118(10): 4521-4532.

Liu, B, Wang, DW, Guo, H, Ling, ZH and Cheung, K (2015). Metallic corrosion in the polluted
urban atmosphere of Hong Kong. Environmental Monitoring and Assessment 187(1):
4112.

Lombardo, T, Ionescu, A, Chabas, A, Lefevre, RA, Ausset, P and Candau, Y (2010). Dose-

response function for the soiling of silica-soda-lime glass due to dry deposition. Science
of the Total Environment 408(4): 976-984.

Lowenthal, DH and Kumar, N (2004). Variation of mass scattering efficiencies in IMPROVE.
Journal of the Air and Waste Management Association (1990-1992) 54(8): 926-934.

Lowenthal, DH and Kumar, N (2016). Evaluation of the IMPROVE Equation for estimating
aerosol light extinction. Journal of the Air and Waste Management Association 66(7):
726-737.

Malm, WC and Hand, JL (2007). An examination of the physical and optical properties of

aerosols collected in the IMPROVE program. Atmospheric Environment 41(16): 3407-
3427.

Malm, WC, Schichtel, B, Molenar, J, Prenni, A and Peters, M (2019). Which visibility indicators
best represent a population's preference for a level of visual air quality? Journal of the
Air & Waste Management Association 69(2): 145-161.

5-59

-------
Malm, WC, Sisler, JF, Huffman, D, Eldred, RA and Cahill, TA (1994). Spatial and seasonal
trends in particle concentration and optical extinction in the United States. Journal of
Geophysical Research 99(D1): 1347-1370.

McAlister, J, Smith, BJ and Torok, A (2008). Transition metals and water-soluble ions in
deposits on a building and their potential catalysis of stone decay. Atmospheric
Environment 42(33): 7657-7668.

McNeill, VF, Woo, JL, Kim, DD, Schwier, AN, Wannell, NJ, Sumner, AJ and Barakat, JM
(2012). Aqueous-phase secondary organic aerosol and organosulfate formation in
atmospheric aerosols: a modeling study. Environmental Science and Technology 46(15):
8075-8081.

Mie, G (1908). Beitrage zur Optik truber Medien, speziell kolloidaler Metallosungen [Optics of
cloudy media, especially colloidal metal solutions], Annalen der Physik 25(3): 377-445.

Mooers, HD, Cota-Guertin, AR, Regal, RR, Sames, AR, Dekan, AJ and Henkels, LM (2016). A
120-year record of the spatial and temporal distribution of gravestone decay and acid
deposition. Atmospheric Environment 127: 139-154.

Myhre, G, Shindell, D, Breon, FM, Collins, W, Fuglestvedt, J, Huang, J, Koch, D, Lamarque, JF,
Lee, D, Mendoza, B, Nakajima, T, Robock, A, Stephens, G, Takemura, T and Zhang, H,
Eds. (2013). Anthropogenic and natural radiative forcing. Cambridge University Press
Cambridge, UK.

Ozga, I, Bonazza, A, Bernardi, E, Tittarelli, F, Favoni, O, Ghedini, N, Morselli, L and Sabbioni,
C (2011). Diagnosis of surface damage induced by air pollution on 20th-century concrete
buildings. Atmospheric Environment 45(28): 4986-4995.

Pitchford, M, Maim, W, Schichtel, B, Kumar, N, Lowenthal, D and Hand, J (2007). Revised

algorithm for estimating light extinction from IMPROVE particle speciation data. Journal
of the Air and Waste Management Association 57(11): 1326-1336.

Prenni, AJ, Hand, JL, Malm, WC, Copeland, S, Luo, G, Yu, F, Taylor, N, Russell, LM and

Schichtel, BA (2019). An examination of the algorithm for estimating light extinction
from IMPROVE particle speciation data. Atmospheric Environment 214: 116880.

Pryor, SC (1996). Assessing public perception of visibility for standard setting exercises.
Atmospheric Environment 30(15): 2705-2716.

Radonjic, IS, Pavlovic, TM, Mirjanic, DLj, Radovic, MK, Milosavljevic, DD and Pantic, LS
(2017). Investigation of the impact of atmospheric pollutants on solar module energy
efficiency. Thermal Science 21(5): 2021-2030.

Rosso, F, Pisello, AL, Jin, WH, Ghandehari, M, Cotana, F and Ferrero, M (2016). Cool marble
building envelopes: The effect of aging on energy performance and aesthetics.
Sustainability 8(8): Article #753.

5-60

-------
Ryan, PA, Lowenthal, D and Kumar, N (2005). Improved light extinction reconstruction in

interagency monitoring of protected visual environments. Journal of the Air and Waste
Management Association 55(11): 1751-1759.

Scott, CE, Rap, A, Spracklen, DV, Forster, PM, Carslaw, KS, Mann, GW, Pringle, KJ, Kivekas,
N, Kulmala, M, Lihavainen, H and Tunved, P (2014). The direct and indirect radiative
effects of biogenic secondary organic aerosol. Atmospheric Chemistry and Physics 14(1):
447-470.

https://casac.epa.eov/ords/sab/t7p i0s IN 10 • >2850 '< a '<* * * * * * 11 * _ 07#report

Shindell, DT, Lamarque, JF, Schulz, M, Flanner, M, Jiao, C, Chin, M, Young, PJ, Lee, YH,
Rotstayn, L, Mahowald, N, Milly, G, Faluvegi, G, Balkanski, Y, Collins, WJ, Conley,
AJ, Dalsoren, S, Easter, R, Ghan, S, Horowitz, L, Liu, X, Myhre, G, Nagashima, T, Naik,
V, Rumbold, ST, Skeie, R, Sudo, K, Szopa, S, Takemura, T, Voulgarakis, A, Yoon, JH
and Lo, F (2013). Radiative forcing in the ACCMIP historical and future climate
simulations. Atmospheric Chemistry and Physics 13(6): 2939-2974.

Sleiman, M, Kirchstetter, TW, Berdahl, P, Gilbert, HE, Quelen, S, Marlot, L, Preble, CV, Chen,
S, Montalbano, A, Rosseler, O, Akbari, H, Levinson, R and Destaillats, H (2014). Soiling
of building envelope surfaces and its effect on solar reflectance - Part II: Development of
an accelerated aging method for roofing materials. Solar Energy Materials and Solar
Cells 122: 271-281.

Smith, AE and Howell, S (2009). An assessment of the robustness of visual air quality
preference study results. CRA International. Washington, DC.
https://www.regulations.gov/document/EPA-HO-Q. >85.

Tai, APK, Mickley, LJ and Jacob, DJ (2010). Correlations between fine particulate matter

(PM2.5) and meteorological variables in the United States: Implications for the sensitivity
of PM2.5 to climate change. Atmospheric Environment 44(32): 3976-3984.

Takemura, T (2012). Distributions and climate effects of atmospheric aerosols from the

preindustrial era to 2100 along Representative Concentration Pathways (RCPs) simulated
using the global aerosol model SPRINTARS. Atmospheric Chemistry and Physics
12(23): 11555-11572.

Twomey, S (1977). The influence of pollution on the shortwave albedo of clouds. Journal of the
Atmospheric Sciences 34(7): 1149-1152.

5-61

-------
U.S. EPA (2004a). Air Quality Criteria for Particulate Matter. (Vol II of II). Office of Research
and Development. Research Triangle Park, NC. U.S. EPA. EPA-600/P-99-002bF.

October 2004. Available at:

https://nepis.epa. gov/Exe/ZyPURL.cgi?Dockey=P 100LG7Q.txt.

U.S. EPA (2004b). Air Quality Criteria for Particulate Matter. (Vol I of II). Office of Research
and Development. Research Triangle Park, NC. U.S. EPA. EPA-600/P-99-002aF.

October 2004. Available at:

https://nepis.epa.gov/Exe/ZyPURL.cgi?Dockey=P100I t.

U.S. EPA (2009). Integrated Science Assessment for Particulate Matter (Final Report). Office of
Research and Development, National Center for Environmental Assessment. Research
Triangle Park, NC. U.S. EPA. EPA-600/R-08-139F. December 2009. Available at:
http s: //cfpub. epa. gov/ncea/ri sk/recordi spl ay. cfm ? dei d=216546.

U.S. EPA (2010). Particulate Matter Urban-Focused Visibility Assessment (Final Document).
Office of Air Quality Planning and Standards, Health and Environmental Impacts
Division. Research Triangle Park, NC. U.S. EPA. EPA-452/R-10-004. July 2010.
Available at: https://nepis.epa.gov/Exe/ZyPURL.cgi?Dockey=P100F< t.

https://nepis. epa. gov/Exe/ZyPURL.cgi?Dockey=P 100AUMY.txt.

U.S. EPA (2012a). Responses to Significant Comments on the 2012 Proposed Rule on the

National Ambient Air Quality Standards for Particulate Matter (June 29, 2012; 77 FR
38890). Research Triangle Park, NC. U.S. EPA. Docket ID No. EPA-HQ-OAR-2007-
0492. Available at: https://www3.epa.gov/ttn/naaqs/standards/pm/dat; pdf.

U.S. EPA (2012b). Report to Congress on Black Carbon. U.S. Environmental Protection

Agency, Office of Air and Radition. Washington, DC. U.S. EPA. EPA-450/R-12-001.
March 2012. Availble at:

https://nepis. epa. gov/Exe/ZyPURL.cgi?Dockey=P 100EIJZ.TXT.

U.S. EPA (2015). Preamble to the integrated science assessments. U.S. Environmental Protection
Agency, Office of Research and Development, National Center for Environmental
Assessment, RTP Division. Research Triangle Park, NC. U.S. EPA. EPA/600/R-15/067.
November 2015. Available at:

https://cfpiib.epa.gov/ncea/isa/recordisplav.cfm7dei

U.S. EPA (2016). Integrated review plan for the national ambient air quality standards for

particulate matter. Office of Air Quality Planning and Standards. Research Triangle Park,
NC. U.S. EPA. EPA-452/R-16-005. December 2016. Available at:

https://www3.epa.gov/ttn/naaqs/standards/pm/data/ . al -integrated-revi ew-

plan.pdf.

5-62

-------
U.S. EPA (2019). Integrated Science Assessment (ISA) for Particulate Matter (Final Report).
U.S. Environmental Protection Agency, Office of Research and Development, National
Center for Environmental Assessment. Washington, DC. U.S. EPA. EPA/600/R-19/188.
December 2019. Available at: https://www.epa.gov/naaqs/particulate-matter-pm-

standards-integrated-science-assessments-current-review.

U.S. EPA (2020). Policy Assessment for the Review of the National Ambient Air Quality

https://www.epa.gov/sYStem/files/documents/20 inal-policv-assessment-for-the-
revi ew-of-th e-pm -n aaq 320.pdf.

https://www.epa.gov/naaqs/particulate-matter-pm-standards-integrated-science-
assessments-current-review.

Van de Hulst, H (1981). Light scattering by small particles. Dover Publications, Inc. New York.

Walwil, HM, Mukhaimer, A, Al-Sulaiman, FA and Said, SAM (2017). Comparative studies of
encapsulation and glass surface modification impacts on PV performance in a desert
climate. Solar Energy 142: 288-298.

Watt, J, Jarrett, D and Hamilton, R (2008). Dose-response functions for the soiling of heritage
materials due to air pollution exposure. Science of the Total Environment 400(1-3): 415-
424.

Zelinka, MD, Andrews, T, Forster, PM and Taylor, KE (2014). Quantifying components of

aerosol-cloud-radiation interactions in climate models. Journal of Geophysical Research:
Atmospheres 119(12): 7599-7615.

5-63

-------
APPENDIX A. SUPPLEMENTAL INFORMATION ON
PM AIR QUALITY ANALYSES

This appendix provides supplemental information on the data sources and methods used
to generate the figures and table presented in Chapter 2 of this PA. Sections A. 1 to A. 3 describe
the data sources and methods used to generate figures and tables in section 2.3.2. Section A.4
describes the data sources and methods used to generate figures and tables in section 2.3.3.
Section A. 5 describes the data sources and methods used to generate figures and tables in section
2.4. Section A.6 described the methods used for the comparison on PM2.5 fields in estimating
exposure and relative to design values.

A.l DATA SOURCES AND METHODS FOR GENERATING NATIONAL
PM2 5, PM10, PM10-2 5, AND PM2 5 SPECIATION FIGURES

• PM2.5 annual average and 98th percentile mass concentrations: calculated from regulatory-
quality (Federal Reference Method or Federal Equivalent Method) 24-hour average
values from monitors with at least 75% completeness for each year. For the FEM
monitors, the 24-hour averages only included days with at least 18 valid hourly
measurements. When a single site has multiple monitors, the figure shows the average of
the annual averages and 98th percentiles from each monitor at the site. We downloaded
the monitor-level concentrations for all sites in the United States for all available days
(including potential exceptional events) for 2000-2019 from the EPA's Air Quality
System (AQS, https://www.epa.eov/aqs)

• PM10 annual average and 2nd highest mass concentrations: calculated from regulatory-
quality (Federal Reference Method or Federal Equivalent Method) 24-hour average
values from monitors with at least 75% completeness for each year. For the FEM
monitors, the 24-hour averages only included days with at least 18 valid hourly
measurements. When a single site has multiple monitors, the figure shows the average of
the annual averages and 2nd highest mass concentrations from each monitor at the site.
We downloaded the monitor-level concentrations for all sites in the United States for all
available days (including potential exceptional events) for 2000-2019 from the EPA's Air
Quality System (AQS, https://www.epa.gov/aqs)

• PM10-7 5 annual average and 98th percentile mass concentrations: calculated from both
regulatory and non-regulatory methods using 24-hour average values from monitors with
at least 75% completeness for each year. When a single site has multiple monitors, the
figure shows the average of the annual averages and 98th percentiles from each monitor at
the site. We downloaded the monitor-level concentrations for all sites in the United
States for all available days (including potential exceptional events) for 2000-2019 from
the EPA's Air Quality System (AQS, https://www.epa.gov/aq_s)

• PM? 5 speciated annual average mass concentrations: calculated from filter-based, 24-hour
averages from monitors with at least 75% completeness for each year. We downloaded

A-l

-------
data from monitors that are part of the Interagency Monitoring of Protected Visual
Environments (IMPROVE) network, Chemical Speciation Network (CSN), and the
NCore Multipollutant Monitoring Network for 2017-2019.

• The 2000-2019 trends are calculated from the Pearson correlation coefficient for monitors
having at least 75% of the available years with 75% completeness within each year.

When a single site has multiple monitors, the average of the annual averages and 98th
percentiles from each monitor at the site is taken prior to calculation of the Pearson
correlation coefficient.

A.2 DATA SOURCES FOR SUB-DAILY PM2 s CONCENTRATION
FIGURE

• PM2.5 hourly average mass concentrations: calculated from regulatory-quality Federal
Equivalent Method monitors. The 2-, 4-, and 5-hour averages were calculated for periods
with each hourly average available. Only sites with a valid annual and 24-hour design
value for 2017-2019 are shown in the figure. The percentages of 2-hour average PM2.5
mass concentrations above 140 ng/in3 at individual sites are illustrated in Figure A-l.
Frequency distributions of 4- and 5-hour averages are presented in Figures A-2 and A-3,
respectively.

A-2

-------
Figure A-l. Percentages of 2017-2019 2-hour average PM2.5 mass concentrations above 140
Hg/rn3.

Sites meeting both NAAQS

Sites violating either NAAQS

Concentration (|ig m"3)

30 60 90 120 150
Concentration (|ig m"3)

Figure A-2. Frequency distribution of 2017-2019 4-hour averages for sites meeting both or
violating either PM2.5 NAAQS for October to March (blue) and April to September
(red).

A-3

-------
Sites meeting both NAAQS

Percentiles (|jg nr3)

Percentiles (|jg m 3)

Sites violating either NAAQS

o
c
CD
ZD

cr
CD

1 (T

10'

10°

10"'

10"

0 30 60 90 120 150 180

Concentration (jug m"3) Concentration (|ig m"'3)

Figure A-3. Frequency distribution of 2017-2019 5-hour averages for sites meeting both or
violating either PM2.5 NAAQS for October to March (blue) and April to September
(red).

A.3 DATA SOURCES FOR ULTRAFINE FRACTION OF PM2 5 MASS
FIGURE

• Annual average particle number and mass concentrations for Bondville. IL: calculated
from 24-hour average values for years with 66% data completion in 75% of the months
of the year from 2000-2019. We downloaded the mass concentrations from the EPA's
Air Quality System (AQS, https://www.epa.gov/aqs) and particle number concentrations
from NOAA's Earth System Research Laboratory's Global Monitoring Division
(https://www.esrl.noaa.gov/gmd).

A.4 METHODS FOR PREDICTING AMBIENT PM2.5 BASED ON HYBRID
MODELING APPROACHES

A.4.1 Data Sources for 2011 PM2.5 Spatial Fields

• The "HU2017" fields were provided by Professor Yang Liu of Emory University in the
form of comma-separated-values files (*.csv) of daily average PM2.5 on a national grid.

• The "DI2016" fields were provided by Dr. Qian Di of Harvard in the form of MATLAB
files (*.mat) of daily average PM2.5 on a national grid.

• The "VD2019" fields were provided by Dr. Aaron van Donkelaar in the form of netCDF
files (*.nc) of annual average concentration. These files are also available at:
http://fizz.phvs. dal.ca/~atmos/martin/?page_id=140.

A-4

-------
• The "downscaler" files were developed in terms of daily average Downscaler predictions
on a national grid following methods described in the risk assessment appendix.

A.4.2 Data Averaging and Coefficient of Variation

• PM2.5 concentration fields were loaded into R version 3.4.4, and daily fields were

averaged to the annual period. Concentrations for each method at prediction points were
then averaged to the corresponding CMAQ grid cells to enable consistent comparisons
for Figure 2-28, Figure 2-29, and Table 2-2.

The coefficient of variation (CoV) was calculated for each grid cell using the following
formula

, , 100

CoV(%) = -=-
F N

if=1(Pi-py

where P is the prediction for each of the four methods (i.e., N=4).
A.5 ANALYSES OF BACKGROUND PM

Data sources for Figure 2-38: Smoke and fire detections observed by MODIS in August
2017

- Image was produced using the NASA Worldview platform

(https://worldview.earthdata.nasa.eov/). Layers selected were 1) Corrected
Reflectance and 2) Fires and Thermal Anomalies, both from Aqua/MODIS. Day
selected was August 4, 2017.

• Data sources for Figure 2-39: Fine PM mass time series during 2017 from North Cascades
IMPROVE site

- Image was archived from the IMPROVE website

(http://views.cira. colostate.edu/fed/SiteBrowser/Default.aspx?appkey=SBCF_Pm
HazeComp; hosted by CIRA/CSU and sponsored by NPS and USFS) for the
North Cascades (NOCA1) site in 2017.

• Data sources for Figure 2-40: Speciated annual average fine PM mass from IMPROVE at
select remote monitors in 2004 and 2016

- Speciated IMPROVE data from 2004 and 2016

(http://views.cira. colostate.edu/fed/SiteBrowser/Default.aspx?appkey=SBCF_Pm
HazeComp) were averaged annually for each monitor. Corresponding monitor
locations are shown in Figure 2-41.

A.6 COMPARISON OF PM2.5 FIELDS IN ESTIMATING EXPOSURE AND
RELATIVE TO DESIGN VALUES: METHODS

Section 2.3.3.2.4 outlines analyses comparing the PM2.5 concentrations in estimating
exposure relative to design values. Below details the data sources and methods used.

To calculate annual average concentrations over the U.S. for 2000-2016, gridded
concentration fields were obtained based on the DI2019 (Di et al., 2019) and the HA2020

A-5

-------
(Hammer et al., 2020) and (Van Donkelaar et al., 2019) methods. The DI2019 concentrations
were acquired from a Google Drive and the HA2020 concentrations (version V4.NA.03) were
acquired from a web link. To identify grid cells that fall within the contiguous U.S. and Core
Based Statistical Areas (CBSAs) boundaries, cartographic boundary shapefiles
("cb_2017_us_state_5m" and "cb_2017_us_cbsa_5m") were downloaded from the census.gov
website. The concentration data and shapefiles were read into R version 3.62 (R Core Team,
2019), and grid cells within the contiguous U.S. and CBSAs were identified using the Simple
Features package version 0.8-0 (Pebesma, 2018) in R. Average concentrations were then
calculated for each year and for each region (i.e., contiguous U.S. and CBSAs within the
contiguous U.S.) using the dplyr package version 0.8.3 (Wickham et al., 2019) in R.

To generate the population-weighting for the DI2019 and HA2020 PM2.5 concentrations,
2015 gridded population counts at 0.05x0.05° from the fourth version of the Gridded Population
of the World (GPWv4; https://sedac.ciesin.columbia.edu/data/collection/gpw-v4) were spatially-
collocated with the PM2.5 concentrations surfaces after conversion to latitude-longitude
coordinates. A similar CBSA filtering was performed for the gridded population and spatially-
collocated PM2.5 surfaces from DI2019 and HA2020 and the fractional population for each grid
was multiplied by the PM2.5 concentrations within each CBSA.

Regulatory design values were calculated using the data handling described by 40 CFR
Appendix N to Part 50 - Interpretation of the National Ambient Air Quality Standards for PM2.5,
by CBSA, for each 3-year period of available hybrid modeling surface data from the EPA's Air
Quality System (AQS, https://www.epa.gov/aqs). Within each CBSA, by each 3-year period, the
ratio of design values to estimated PM2.5 concentrations was calculated.

A-6

-------
REFERENCES

Hammer, MS, van Donkelaar, A, Li, C, Lyapustin, A, Sayer, AM, Hsu, NC, Levy, RC, Garay,
MJ, Kalashnikova, OV and Kahn, RA (2020). Global estimates and long-term trends of
fine particulate matter concentrations (1998-2018). Environmental Science &
Technology 54(13): 7879-7890.

Pebesma, EJ (2018). Simple features for R: standardized support for spatial vector data. R J.
10(1): 439.

R Core Team (2019). R: A language and environment for statistical computing R Foundation for
Statistical Computing Vienna, Austria.

Van Donkelaar, A, Martin, RV, Li, C and 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. Environmental science &
technology 53(5): 2595-2611.

Wickham, H, Francis, R, Henry, L and Miiller, K (2019). dplyr: A Grammar of Data
Manipulation. R package version 0.8.3.

A-7

-------
APPENDIX B. SUPPLEMENTAL STUDY
INFORMATION: SELECTION CRITERIA, STUDY
METHODS AND DETAILS

-------
TABLE OF CONTENTS

B.l ForestPlots B-l

B.2 Monitored PM2.5 Concentrations in Key Epidemiologic Studies B-l

B.3 Hybrid Model Predicted PM2.5 Concentrations in Key Epidemiologic Studies B-3

B.4 Details of Key Epidemiologic Studies, Including Study Design, Exposure Metric, and
Statistical Analysis B-7

References B-92

-------
This appendix presents supplemental information on the key epidemiologic studies
evaluated in section 3.3.3 of this PA. Section B.l provides supplemental information on the
forest plots presented in Figures 3-4 to 3-7. Section B.2 provide supplemental information on the
study-reported PM2.5 concentrations presented in Figure 3-8, Figure 3-9, while section B.3
provides supplemental information on studies presented Figure 3-10 and Figure 3-11. Section
B.4 provides details on key elements of epidemiologic studies, including the study design and
details on the statistical analyses employed, including control for confounding effects.

B.l FOREST PLOTS

Figure 3-4 through 3-7 in Chapter 3 present forest plots that include the effect estimates
and 95% confidence intervals from epidemiologic studies that were assessed in the 2019 ISA and
ISA Supplement that have the potential to be most informative in reaching conclusions on the
adequacy of the current primary PM2.5 standards. Epidemiologic studies included in these figures
support "causal" or "likely to be causal" relationships with PM exposures in the 2019 ISA and
include mortality (all-cause mortality, cardiovascular (CVD) mortality, respiratory mortality,
lung cancer mortality), and morbidity (asthma incidence, lung cancer incidence, lung function
and lung development, CVD and respiratory emergency room visit or hospital admission) health
endpoints. Further, studies included in Figure 3-4 to Figure 3-7 were restricted to multi-city
studies in the United States or Canada. Multi-city studies within a single State were not included,
with the exception of respiratory morbidity endpoints, where multi-city studies were limited
(U.S. EPA, 2019). For some of the major cohort studies included in the 2009 ISA, like the
American Cancer Society (ACS) cohort, we included more recent studies that reanalyze
epidemiologic associations for multiple mortality endpoints (e.g. lung cancer mortality and IHD
mortality) and an extension of follow-up periods (e.g., Pope et al. (2015), Turner et al. (2016),
Jerrett et al. (2016), and Thurston et al. (2016b)), as well as a reanalysis (Krewski et al. (2009) of
the original ACS dataset, including an extended follow-up period, that was evaluated in the 2009
ISA (U.S. EPA, 2009)).

B.2 MONITORED PM2.5 CONCENTRATIONS IN KEY EPIDEMIOLOGIC
STUDIES

Based on the key studies identified in Figure 3-4 to Figure 3-7, a subset of studies are
depicted in Figure 3-8 and Figure 3-9 and includes key epidemiologic studies that report an
overall study mean or median concentration of PM2.5 (as opposed to a study mean/median range
across study area locations) and based on ambient PM2.5 monitored data. The plots include
studies that report significant effect estimates and studies that report non-significant effect
estimates. Further, to be included, only key studies for which the years of air quality data used to

B-l

-------
estimate exposures overlap entirely with the years during which health events are reported were
included. The PM2.5 concentrations reported by studies that estimate exposures from air quality
corresponding to only part of the study period, often including only the later years of the health
events1 are not likely to reflect the full ranges of ambient PM2.5 concentrations that contributed to
reported associations.2

Some of the key epidemiologic studies assessed in the 2019 ISA also provide city-
specific study mean concentrations and city-specific health events, but this information was not
available in studies evaluated in the ISA Supplement. PM2.5 exposure estimates corresponding to
the 10th and 25th percentiles of those events were calculated in the following manner. City-
specific cases and PM2.5 concentrations were input in ascending order by PM2.5 concentration.
The city-specific percent of cases was calculated as a proportion of the total study cases and the
cumulative percent of cases was determined. The PM2.5 concentration associated with the
cumulative percent closest to the 10th and 25th percentiles are presented in Figure 3-8 and Figure
3-9 and the cumulative percent values closest to the associated 10th and 25th percentile values are
shown in Table B-l.3 Data for Bell et al. (2008) and Zanobetti and Schwartz (2009) were
previously provided by the study authors, as described in Rajan (2011).

Table B-l. PM2.5 concentrations corresponding to the 25th and 10th percentiles of estimated
health events.

Citation

10th Percentile PM2.5
(|jg/m3) (Cumulative
percent value closest)

25th Percentile PM2.5
(|jg/m3) (Cumulative
percent value closest)

Bell et al. (2008)

9.8

11.5

Franklin et al. (2007)

10.4(11.1%)

12.9 (25.3%)

Stieb et al. (2009)

6.7(16.5%)

6.8 (20.5%)

Szyszkowicz (2009)

6.4(4.1%)

6.5(18.6%)

Zanobetti and Schwartz (2009)

10.3

12.5

1 The following studies do not have an overlap between the years of PM2 5 air quality data and the years during
which health effects are reported: Miller et al., 2007 ; Hart et al., 2011 ; Thurston et al., 2013; Weichenthal et al.,
2014; Pope et al., 2015 ; Villeneuve et al., 2015; Turner et al., 2016; Weichenthal et al., 2016a;; Parker et al.,
2018; Pope etal., 2019; and Bevan et al., 2021.

2 This is an issue only for some studies of long-term PM2 5 exposures. While this approach can be reasonable in the

context of an epidemiologic study evaluating health effect associations with long-term PM2 5 exposures, under the
assumption that spatial patterns in PM2 5 concentrations are not appreciably different during time periods for
which air quality information is not available (e.g., Chen et al., 2016), our interest is in understanding the
distribution of ambient PM2 5 concentrations that could have contributed to reported health outcomes.

3 That is, 25% of the total health events occurred in study locations with mean PM2 5 concentrations (i.e., averaged

over the study period) below the 25th percentiles identified in Figure 3-8 and Figure 3-9 and 10% of the total
health events occurred in study locations with mean PM2 5 concentrations below the 10th percentiles identified.

B-2

-------
B.3 HYBRID MODEL PREDICTED PM2.5 CONCENTRATIONS IN KEY
EPIDEMIOLOGIC STUDIES

Figure 3-10 and Figure 3-11 focus on multicity/multistate studies in the U.S. and Canada,
that are part of the evidence supporting "causal" or "likely to be causal" determinations in the
2019 ISA and that use hybrid modeling methods to estimate PM2.5 exposures, as well as studies
assessed in the ISA Supplement. In addition, as detailed in section 3.2.3.2.1, for studies included
in Figure 3-10 and Figure 3-11 we also consider the approach used to estimate PM2.5
concentrations and the approach used to validate hybrid model predictions when determining
those studies that we identify as key epidemiologic studies. Such studies are identified as those
that use hybrid modeling approaches for which recent methods and models were used (e.g.,
recent versions and configurations of the air quality models); studies that are fused with PM2.5
data from national monitoring networks (i.e., FRM/FEM data); and studies that reported a
thorough model performance evaluation for core years of the study.4

Figure 3-10 and Figure 3-11 present overall means of hybrid model-predicted PM2.5
concentrations for key studies, and the concentrations corresponding to the 25th and 10th
percentiles of estimated exposures or health events, when available. For Di et al. (2017b), we
present 25th and 10th percentiles of annual PM2.5 concentrations by ZIP code corresponding to
long-term exposure estimates, while for Di et al. (2017a), we present daily air pollution
concentrations (short-term exposure estimates) corresponding to the 25th and 10th percentiles of
deaths at the ZIP code level. These values, along with other percentiles, are illustrated in Figure
B-l and Figure B-2 (Jenkins, 2019a, Jenkins, 2019b). The study authors for Di et al. (2017b)
additionally provided information on population weighted percentile values corresponding to
long-term PM2.5 exposure (Chan, 2019). These are presented in Table B-2. For other studies
included in Figure 3-10 and 3-11 (Bai et al., 2019, Erickson et al., 2019, Kloog et al. (2012),
Kloog et al. (2014), Shi et al. (2016), U.S. EPA, 2021, and Wang et al. (2017)), 25th percentiles
of exposure estimates were derived from study manuscripts of air quality descriptive statistics
and can be found in Table B-3.

4 The following studies do not meet these criteria: Bravo et al., 2017, Crouse et al., 2015; Puett et al., 2009, Puett et
al., 2011, Hystad et al., 2012; Hystad et al., 2013, Hayes et al., 2020; Elliott et al., 2020; Lefler et al., 2019;
Pappin et al., 2019; Cakmak et al., 2018; Fisher et al., 2019; Sun et al., 2019; McClure et al., 2017; Loop et al.,
2018 ; and Honda etal., 2017.

B-3

-------
Percentiles of PM2.5 By Zip Code

Threashoids defining percentiles of PM2 5 exposure for each zip code.

Percentile of PM2.5, Based on ZIP code PM2.5 Value

0.0209025

6.1962803

10%

7.2742546

15%

8.0043245

20%

8.5892973

25%

9.0612931

30%

9.4644903

35%

9.8273901

40%

10.1797192

45%

10.5371831

50%

10.9015790

55%

11.2791073

G0%

11.6666804

65%

12.0707952

70%

12.4916270

75%

12.9386305

80%

13.4294338

85%

13.9765291

90%

14.6375324

95%

15.6106067

100% 32.5759482

Figure B-l. Percentiles of annual PM2.5 concentrations by ZIP code corresponding to long-
term exposure estimates in Di et al., 2017b.

B-4

-------
Table B-2. Population weighted percentiles of annual PM2.5 concentrations by ZIP code
corresponding to long-term exposure estimates in Di et al., 2017b.

Percentile

Population Weighted PM2.5

(Hg/m3)

0.0

5.0

7.1

10.0

7.9

15.0

8.6

20.0

9.1

25.0

9.5

30.0

9.9

35.0

10.3

40.0

10.6

45.0

11.0

50.0

11.4

55.0

11.7

60.0

12.1

65.0

12.5

70.0

12.9

75.0

13.4

80.0

13.9

85.0

14.4

90.0

15.1

95.0

16.1

100.0

32.6

B-5

-------
Percentiles of PM2.5 By Zip Code

Threasholds defining percentiles of Daily PM2.5 exposure for each zip code.

Percentile of Daily PM2,5, Based on ZIP code PM2.5 Value

0% 0.0006378

5% 3.8286960

10% 4.7224770

15% 5.4309290

20% 6.0727840

25% 6.6863868

30% 7.2922285

35% 7.9031599

40% 8.5292050

45% 9.1836408

50% 9.8740436

55% 10.6124979

60% 11.4111824

65% 12.2910351

70% 13.2835707

75% 14.4301324

80% 15.8159815

85% 17.5894591

90% 20.0959732

95% 24.4759063

100% 201.3071287

Figure B-2. Daily air pollution concentrations (short-term exposure estimates)

corresponding to various percentiles of deaths at the ZIP code spatial resolution level in
Di et al., 2017a.

B-6

-------
Table B-3. PM2.5 concentrations corresponding to the 25th and 10th percentiles of
estimated exposures in Figure 3-8.

Citation

10th Percentile PM2.5 (HEJ/m3)

25th Percentile PM2.5 (HE)/™3)

Di et al. (2017a)

4.7

6.7

Di et al. (2017b)

7.3

9.1

Kloog et al. (2012)

6.4

Kloog et al. (2014)

7.9

Shi et al. (2016)

4.6

Shi et al. (2016)

6.2

Wang et al. (2017)

9.1

Bai et al. (2019)

7.9

Christidis etal. (2019)

4.3

Shinet al. (2019)

B.4 DETAILS OF KEY EPIDEMIOLOGIC STUDIES, INCLUDING
STUDY DESIGN, EXPOSURE METRIC, AND STATISTICAL
ANALYSIS

Table B-4 below summarizes additional details related to the designs of the U.S. and
Canadian epidemiologic studies included in Figures 3-4 to 3-7, and Figure 3-8 to Figure 3-11, as
well as studies included in the risk assessment (Table 3-13).

B-7

-------
Table B-4. Study characteristics from key studies.

Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Bai et al., 2019

CHF and

AMI
incidence
based on
hospital
discharge
data

Ontario,
Canada

Cohort
study
(ONPHEC
)

Annual estimates of PM2.5
were derived using AOD,
GEOS-CHEM combined
with a geographically
weighted regression
model for North America.
The PM2.5 estimates were

available at 1-km2
resolution for the period of

1998-2012. PM2.5
estimates in 1998 were
rescaled to 1994-1997
and in 2012 to 2013-2015.
Annual estimates of PM2.5
concentrations
were assigned to the
participants' postal code
of residence and were
then used to calculate a 3-
year moving average
PM2.5 concentration for
each year of follow-up for
study participants in the
study.

Cox Proportional Hazard
models were used to estimate
associations of incidence of
CHF and AMI with PM2.5. The
shape of the concentration-
response relationships between
the pollutants and outcomes
were assessed using Shpe
Constrained Health Impact
Functions (SCHIF) in the fully
adjusted Cox models.

Models were adjusted for
individual-level covariates:
neighborhood-SES, urban/rural
residency, and various other
factors in sensitivity analysis
(comorbidities, neighborhood
deprivation, minority, health
care access).

The 3-year
moving average

for study
participants at
the baseline
residence
location was
used to calculate
overall mean
PM2.5
concentrations at
the beginning of
the follow-up
period in 2001.

B-8

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Baxter et al.,
2017

All-cause
mortality

77 U.S. Cities

Time
Series
study
(NCHS)

EPA's National and State
Local Ambient Monitoring
Stations providing
integrated daily
measurements and
operated more than 6
months or had more than
30 observations (2001-

2005) considered.
Monitors representing the
general population
exposure in the cities
were selected. For this
correlation was assessed

between each pair of
monitors within the county
and the ones uncorrelated

(coefficients.8 with
majority of other monitors)

were excluded. Once
appropriate valid monitors
were identified the
summary measure of
PM2.5 concentration over
the county was calculated.

2-day moving average
(lag 0-1 days) of PM2.5
cone included in the
model.

Poisson regression model and
meta-regression

In stage 1, ran single city
Poisson time-series models;
adjusted for temperature and

dew point temperature,
including variables for previous
day temperature, temporal
trends, and trends by age.
In stage 2, meta-regression with
cluster analysis (5 clusters)
based on characteristics of
residential infiltration.

Average daily
PM2.5 values
were calculated
for each city.
First, a global

mean and
variance were
created within
each city for the

entire time
period. Using the
valid monitor
measurements.
Next, all values

were
standardized and
average PM2.5
within a given
day in each city
was calculated.
Finally, the
standardized
daily value was

reversed to
calculate average
daily PM2.5 for
each city.

B-9

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Bell et al., 2008

CVD HA
Age 65+

202 U.S.
Counties with
populations>
200,000

Time
Series
study
(MEDICA

RE
enrollees)

PM2.5 concentrations
obtained from EPA
monitors providing data
daily or every 3 days for
the period 1999-2005.
Used 10% trimmed mean
to calculate daily average
across monitors after
correction for yearly
monitor averages (to
protect against outliers as
applied in Dominici et al.
2006).

Used lagO PM2.5 in the
model.

2-stage Bayesian hierarchical
model

In stage 1, adjusted for
temperature and dew point
temperature, including variables
for previous day's conditions,
day-of-the-week, temporal
trends, and differential temporal

trends by age. In stage 2,
county-specific estimates were
combined, accounting for their
statistical uncertainty.

Average daily

PM2.5
concentrations
for each county
used to calculate
overall mean for
the study area
and duration.

B-10

-------
Bell et al., 2014

CVD,
Asthma,

and
COPD HA
Age 65+

4 Counties in
MA and CT

Time-
series
study
(MEDICA

RE
enrollees)

PM2.5 Teflon filter samples
(measuring PM2.5 total
mass) obtained from CT
and MA DEP for the
period of 2000-2004.
Used data from five
monitoring locations
(providing daily or every
third day data) within four
county regions. Assigned
daily PM2.5 concentration
from a single monitor to

three counties. For
Fairfield County with two
monitors: daily PM2.5
concentration was
calculated by using
population-weighted
averaging of census tract

PM2.5 concentrations.
First, each census tract in
the Fairfield County (209
tracts in total) was
assigned the PM2.5
exposure of the nearest
monitor. Then, PM2.5
exposures for all tracts

were averaged and
weighted by each tract's

2000 U.S. census
population to calculate a
county-level exposure for
the Fairfield County.

Log-linear Poisson regression
analysis

Adjusted for temperature and
dew point temperature,
including previous day's
temperature and dew point
temperature, day-of-the-week
temporal trends, and region.

Daily PM2.5
concentrations

for all four
counties (three

with single
monitor and one
with two monitors
that used
population
weighted
approach) over
the period of
2000-2004 were
used to calculate
the overall mean
PM2.5for the
study location
and period.

B-ll

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Explored various lags and
presented lagO PM2.5
model.

Bell et al., 2015

HF HA
65+

213 U.S.
Counties

Time-
series
study
(MEDICA

RE
enrollees)

Daily monitored PM2.5
data from the U.S. EPA
AQS monitors for the
period of 1999-2010. On

average, county-level
PM2.5 data was available
for 56.5% of study days
(range: 7.8%-99.9%; no

imputation done for
missing data). For each
county, daily PM2.5
measurement was
calculated by averaging
the PM2.5 values from all
monitors within a county
in a given day.

Explored various lags and
presented lagO PM2.5
model.

2-stage Bayesian hierarchical
model

The stage 1 model included
county-specific model adjusted
for weather (temperature, dew

point, previous days'
temperature, and dew point),
day-of-the-week, and temporal

trends. In stage 2 county-
specific effect estimates were
pulled together to present
overall association.

Daily PM2.5
concentrations
for 213 counties
over the period of
1999-2010 were
used to calculate
region-specific
mean PM2.5, and
overall mean
PM2.5 for the
study location
and period.

B-12

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Bravo etal.,
2017

CVD HA
Age 65+

418 U.S.
Counties

Time-
series
study
(MEDICA

RE
enrollees)

Daily (24-hr) monitored
PMzsdata from the U.S.

EPA AQS monitors
(NAMS/SLAMS) obtained

for the period of 2002-
2006. Approximately 80%
of PM2.5 monitors
recorded observation
once every 3 days. For
each county (>=50K
population), daily (24-hr)
PM2.5 concentration was
calculated by averaging

multiple monitor
measurements for the
same day.

Explored various lags and
distributed lags of PM2.5
exposure.

2-stage Bayesian hierarchical
model

The stage 1 included log-linear
Poisson regression models with

over-dispersion fit at county-
level. Model adjusted for same-
day temperature and dew point
temperature, 3-day moving
average of temperature and
dew point temperature,
temporal trends in
hospitalizations, day-of-the-

week, and age. Fitted
distributed lag model with
multiple lags (0- to 7-day lags)
of PM2.5 cone simultaneously in
the county-specific model.
The stage 2 estimated the
association for the entire study
area using two-level normal

independent sampling
estimation with priors thus
allowing to combine risk
estimates across counties while
accounting for within county SE
and between-county variability
in the true RR.

Daily PM2.5
concentrations
for 418 counties
over the period of
2002-2006 were
used to calculate
overall mean
PM2.5 for the
study location
and period.

B-13

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Bravo etal.,
2017

CVD HA
Age 65+

708 U.S.
Counties

Time-
series
study
(MEDICA

RE
enrollees)

Daily PM2.5 concentrations
were estimated at census
tract centroids using the
downscaler method (input
from the U.S. EPAAQS
NAMS/SLAMS monitoring
data, and gridded 12x12
km CMAQ) for the period
of 2002-2006. County-
level daily PM2.5
exposures were
calculated from a pop-
weighted averages of
PM2.5 concentrations
predicted at census tract
within each county using
2000 U.S. Census data.
CMAQds was generated
for all days in the study

period 2002-2006.
CMAQds-subset was
calculated by taking
population-weighted
county level exposures
only for counties and days
with monitoring data
(n=418 counties.

Explored various lags and
distributed lags of PM2.5
exposure.

2-stage Bayesian hierarchical
model

The stage 1 included log-linear
Poisson regression models with

independent sampling
estimation with priors thus
allowing to combine risk
estimates across counties while
accounting for within county SE
and between-county variability
in the true RR.

24-hr average

PM2.5
concentrations
for 708 counties
over the period of
2002-2006 were
used to calculate
overall mean
PM2.5 for the
study location
and period.

B-14

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Burnett and
Goldberg, 2003

All-cause
mortality

8 Canadian
Cities

Time-
series
study

PM2.5 data obtained from
dichotomous sampler with
Teflon filters operating on
6-day schedule for the
period of 1986-1996.

Each city had one
sampler and two cities
have two samplers. If two
samplers then data was
averaged between the
samplers and assigned to
the city.

Lag 1 explored.

Generalized additive model
(GAM) analysis to generate
pooled estimate of air pollution
effect among the eight cities.

The model adjusted for day-of-
the-week, temporal trends, and
weather variables (daily
average temperature, daily
average relative humidity, and
barometric pressure lagged 0
and 1 days).

Daily PM2.5
concentrations
(day before the

death) for 8
Canadian cities
over the period of
1986-1996 were
averaged to get
overall mean for
the study area
and period

Burnett et al.,
2004

All-cause
mortality

12 Canadian
Cities

Time-
series
study
(data from
Statistics
Canada)

Monitoring data available

for 12 cities from the
Statistics Canada for the

period of 1981-1999.
PM2.5 data available every
6th-day sampling
schedule. Daily PM2.5
concentrations were
calculated for each city by
averaging data over all
monitors with each city.

Explore various lags and

moving average and
presented data for lag 1
for PM2.5.

Random-effects regression
model.

Adjusted for temporal trends in
mortality and effects of weather
using humidex index at lag 0

and lag 1 (a measure of
combined effect of temperature
and humidity)

Daily PM2.5
concentrations
for all 12 cities
over the period of
1981-1999 were
used along with

population
information to
calculate an
overall population
weighted PM2.5
concentration for
the study location
and period.

B-15

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Cakmak et al.,
2018

Non-
accidental

, CVD,
respiratory
and lung
cancer
mortality

Canada
Nationwide

Cohort
study
(CanCHE
C)

PM2.5 estimates obtained

from median satellite-
derived concentrations for
the period of 1998-2011.
The concentration was
determined at 10 km2
resolution as detailed in
(van Donkelaar, 2010).

Changes in PM2.5
between 1998 and 2006
was inferred using
satellite instruments,
MISR and SeaWiFS
(Boys, 2014). Annual
estimates of PM2.5
concentration was
assigned to participants
based on postal code of
residence and was used
to calculate 7-year moving
average (at least 4 out of
7 years of data is
available) PM2.5
concentration for each
year of follow-up in the
study.

Cox proportional hazards
models to estimate the
relationship between long-term
exposure and date of death
accounting for residential
mobility.

Model adjusted for individual-
level covariates (aboriginal
ancestry, minority status,
marital status, education,
immigrant status and income)

The 7-year
moving averages

for study
participants were

then used to
calculate overall

mean PM2.5
concentration (all
and by
geographic
zones)

B-16

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Chen et al.,
2020

CVD
mortality,

CVD
morbidity

Ontario,
Canada

Cohort
study
(ONPHEC
)

PM2.5 concentration
estimated 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, 2019 at the
spatial resolution of 1
km2. Annual estimates of
exposure to PM2.5 and the

composition for each
participant was estimated
by interpolating the annual
mean concentrations of

PM2.5 and the
corresponding proportion
of PM2.5 attributed to the
seven major components

to the centroid of their
residential postal code for

that year, thereby
accounting for residential
mobility.

Component-adjusted approach
that jointly estimated the health
impacts of PM2.5 and its major
components while allowing for a
potential nonlinear
PM2.5-outcome relationship.
Compared this approach with
three traditional approaches
using Cox Hazard models.

Adjusted for individual-level
covariates, four time-varying
variables for neighborhood-level
SES, area-level indicators.

Annual PM2.5
concentrations
across all postal
code areas in the
Ontario region
were used to
calculate overall

mean PM2.5
concentration for
the study location
and period.

B-17

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Chi et al., 2016

CVD first
events
(Ml,
stroke,
death
from CHD
and
CBVD)

Nationwide

Cohort

study

(WHI)

Likelihood-based ambient

point-specific PM2.5
predictions at participant
residences were obtained

using a regionalized
national universal kriging
model that included over

200 geographic
covariates reduced via

partial least squares
techniques (Sampson et
al. 2013). Only the year
2000 data was used to
represent the exposure for
the entire study follow up

period considering
consistent concentrations
of the particulate pollution.
For each participant,
PM2.5 exposure was
calculated as an average

of the current and all
previous PM2.5 predictions
weighted by time spent at

each residence. The
exposure was only time-
varying as it incorporated
residential history, but not
time-varying in calendar
time as all predictions
were estimated for the
year 2000.

Cox proportional hazards
models to estimate the
relationship between long-term
annual average PM2.5 exposure

and first CVD events. Effect
modification by each individual-
level SES and Neighborhood
SES indicator was
investigated by fitting
multiplicative interaction terms
for different levels of the
SES variable with PM2.5.

Model adjusted for individual-
level covariates (age,
race/ethnicity, diabetes,
hypertension,
hypercholesterolemia, smoking
and BMI)

Annual average

PM2.5
concentrations
for the study
participants were
used to calculate
overall mean
PM2.5
concentration for
the study location
and period.

B-18

-------
Christidis et al.,

Non-

Canada

Cohort

PM2.5 exposures derived

Cox proportional hazard models

The 3-year

2019

accidental

Nationwide

study

from AOD retrievals using

to assess the relationship

moving average

mortality

(mCHHS)

GEOS-Chem calibrated to
surface measurements by
GWR (van Donkelaar,
2015). Spatial variation
from modeled surface
used with simulate PM2.5
and constrained with local
ground-based monitors to

estimate PM2.5
concentrations through
2015 (Meng, 2019).
Linked postal codes to
PM2.5 concentrations
using points of latitude
and longitude. When
multiple points of latitude
and longitude was
available for a single
urban postal code, equal
weighting of the multiple
air pollutant values was
used to provide a singular

value. In rural
communities, population-

weighted average of
the values associated with
duplicate postal codes

was used. Used
population-weighing to
average multiple values to
create inputs for partial
postal codes (2 to 5 digit).
For each individual and
year of follow-up, PM2.5

between PM2.5 exposure and
non-accidental death in low-
exposure environment.

C-R relationship observed using

Shape constrained health
impact function (SCHIF Model)
Adjusted for socio-economic,
behavioral, and time-varying
contextual covariates

PM2.5
concentrations
for the study
participants used
to calculate
overall mean
PM2.5 concentrati
on for the study
periodr

B-19

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

estimates was calculated
as 3-year moving average
with one-year lag.

B-20

-------
Crouse et al.
2012

All-cause
mortality

11 Canadian
Cities

National
Cohort
study
(Subset of
Canadian
census
mortality
follow-up

study;
43%; non-
immigrant
population

Monitor data from ground-
based stations available
for 11 cities for the 15-yr
period including the 5-yr
prior to baseline and 10-yr
of follow-up (1987-2001)
from Statistics Canada.
PM2.5 data available every
6th-day sampling
schedule. To address
missing monthly PM2.5
data for some stations,
data from all stations
within 6-km of each other
were pooled to calculate
monthly, seasonal, annual
and five-yr (1987-1991,
1992-1996, 1997-2001)
means at each monitored
location. Mean annual
concentration (averaged
over 1987-2001) from
ground-based monitors
was then assigned to the
cohort member based on
the 11 census divisions of

their residence.
A second set of exposure
(10x10 km) was created
using estimates of PM2.5

from remote sensing
during period 2001-2006
to calculate 6-yr average.
The mean concentration
of PM2.5 within boundaries
of each enumeration area

2 different modelling approach.
Approach 1: Cox proportional
hazards model, and Approach

2: nested, spatial random-
effects Cox model with spatial
clusters.

Models adjusted for individual-
level covariates, urban/rural

indicator, and ecological
covariates (% unemployed, %
without high school diploma,
lowest income quintile, and
rural/urban indicator).

AnnualPM2.5
concentrations
for the study
participants were
used to calculate
overall mean
PM2.5 for the
study population
and duration.

B-21

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

was calculated by
overlaying PM surface

over the surface of
enumeration area across
country. Satellite derived
PM2.5 estimate was then
assigned to participants
based on their
enumeration area of
residence in 1991.

B-22

-------
Crouse et al.,

Non-

Canada

Cohort

PM2.5 concentrations

Cox Hazard model to assess

The moving

2019

accidental

Nationwide

study

derived from AOD

the relationship between PM2.5

average PM2.5

, CVD,

(CanCHE

retrievals using GEOS-

exposure at different temporal

concentrations at

respiratory

Chem calibrated to

and spatial scales.

the baseline

mortality,

surface measurements by

Adjusted for individual-level

residence

and lung

GWR (van Donkelaar,

variables (aboriginal identity,

location were

cancer

2015). Spatial variation

visible minority status, marital

then used to

from modeled surface

status, highest level of

calculate overall

used with simulated PM2.5

education, employment status,

mean PM2.5

and constrained with local

and household income

concentration at

ground-based monitors to

adequacy quintiles)

the beginning of

estimate PM2.5

the follow-up

concentrations through

period in 2001 at

2015 as detailed in

various temporal

(Meng, 2019). Linked

and spatial

postal codes to PM2.5

scales.

concentrations from grid

cells. Annual PM2.5

estimates from the postal

code and assigned to

study participants based

on the postal code for

residence was used to

calculate moving

average at various

temporal (1-, 3-, 8-year)

and spatial scales (1-, 5-,

10-km) based on the

location and year of

follow-up.

Dai et al., 2014

All-cause,

75 U.S. Cities

Time-

Monitored data obtained

Two stage: Stage 1. City-

Daily PM2.5

CVD, and

(with

series

from U.S. EPA AQS for

specific season-stratified time-

concentrations

available

the period of 2000-2006.

for all 75 cities

B-23

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Respirator
y mortality

daily mortality

data and
PM2.5 data for
at least 400
days between
2000 and
2006)

study
(NCHS)

Daily PM2.5 concentrations

from each monitor
assigned to corresponding
city. For cities with more
than one sampling site,
concentration data were

averaged across all
monitors within the city.

Used average of 2-day
lag (lag 01) PM2.5.

series analysis using Poisson
regression in GAM

Model adjusted for 24-hr
average temperature from
closest weather station to the
city center at lagO and Iag1,
temporal trends, and day-of-
the-week. Stage 2. Multivariate
random effects meta-analysis to
combined 300 (i.e., 75 cities * 4
seasons) effect estimates to
obtain overall association.

over the period of
2000-2006 were
used to calculate
an overall mean

PM2.5
concentration for
the study location
and period.

B-24

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

deSouza et al.,
2021

First CVD
HA

U.S.
Nationwide

Time-
stratified
case-
crossover
design (M
EDICAID)

PM2.5 concentration were
derived for 1 km2 grid
cells in the continental
United States by
integrating remote
sensing, outputs from a
chemical transport model,
and other variables such
as meteorological and
land-use variables (Di et

al. 2019); from an
ensemble-based model
that integrated multiple

machine learning
algorithms for the period
of 2000-2012. Daily PM2.5
estimates of grid
cells averaged at ZIP
code were assigned to
study participants based
on the ZIP code of

residence.

Used Iag01 average
exposure in the model.

Conditional logistic regression

models to estimate the
associations between short-
term exposure to PM2.5 and
CVD hospitalization rates.
Adjusted for individual-level
covariates, air and dew-point
temperature.

Daily PM2.5
concentration fro

m case days
were then used
to calculate
overall case
day mean
PM2.5 concentra
tion for the study
location and
period.

B-25

-------
Diet al.,2017b

All-cause
mortality
65+

U.S.
Nationwide

Cohort
(MEDICA

RE
enrollees)

Artificial neural network
that incorporated satellite-

based measurements,
simulation outputs from a
chemical transport model,

land-use terms,
meteorological data, and
other data to predict daily
concentrations of PM2.5.
The neural network was fit
with monitored PM2.5 data
and daily PM2.5
concentrations were
predicted for nationwide
grids that were 1x1 km.
ZIP code-level PM2.5
concentrations were
estimated by taking the
inverse-distance averages
of the four nearest grid
cells to the ZIP code
centroid and the annual
average was calculated.
For each calendar year
during which a person
was at risk of death the
annual average PM2.5
concentration was
assigned according to the
ZIP code of the person's
residence. As part of a
sensitivity analysis,
monitored PM2.5 data was
matched with each person
in the study within a

Two-pollutant Cox proportional
hazards model with generalized
estimating equation to account
for correlation between ZIP
codes.

Accounted for individual
variables, (sex, race, Medicaid
eligibility, and average age at
study entry), ZIP code-level
variables (% Hispanic, % Black,
median household income,
median value of housing, % >
65 living below poverty level, %
> 65 with less than high school
education, % of owner-occupied
housing units, and population
density), county-level variables
(county-level BMI and % ever
smokers), hospital service area-
level variables (% low-density
lipoprotein level measured, %
glycated hemoglobin level

measured, and % >1
ambulatory visits), 32 km2
gridded weather and 1 km2
gridded pollution variables
(annual average PM2.5
concentration, annual average

temperature, and annual
average humidity), monitor level

air pollution variables (PM2.5
monitored data), and a regional
dummy variable.

Average PM2.5
concentrations
over all ZIP
codes in the
continental U.S.
were used to
calculate overall
mean PM2.5 for
the study location
and period.

B-26

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

distance of 50 km of the
nearest monitoring site.

Dietal., 2017b
(< 12 ug/m3)

Analysis restricted to
persons-years with PM2.5
exposures lower than 12
ug/m3

B-27

-------
Di et al„ 2017a

All-cause
mortality
65+

US
Nationwide

Case-
crossover

study
(MEDICA

RE
enrollees)

Artificial neural network
that incorporated satellite-

based measurements,
simulation outputs from a
chemical transport model,

land-use terms,
meteorological data, and
other data to predict daily
concentrations of PM2.5.
The neural network was fit
with monitored PM2.5 data
and daily PM2.5
concentrations were
predicted for nationwide
grids that were 1x1 km.
For each case day (date
of death) and its control
days, the 24-hour PM2.5
concentrations were
assigned ZIP code by
taking the inverse-
distance mean of the four
nearest grid cells to the
ZIP code centroid of the
residence.

As part of a sensitivity
analysis, monitored PM2.5
data was matched with
each person in the study
within a distance of 50 km
of the nearest monitoring
site, and cross-validation
was performed between
predicted and monitored
concentrations.

Conditional logistic regression.

"Case Day" defined as death.
For the same person, compared
daily air pollution exposure on

the case day vs. daily air
pollution exposure on "control

days." Control days were
chosen (1) on the same day of
the week as the case day to
control for potential confounding

effect by day of week; (2)
before and after the case day to
control for time trend; and (3)
only in the same month as the
case day to control for seasonal
and sub-seasonal patterns.
Individual-level covariates and
ZIP code-level covariates that
did not vary day to day (e.g.,
age, sex, race/ethnicity, SES,
smoking, and other behavioral

risk factors) were not
considered to be confounders
as they remain constant when
comparing case days vs control
days.

The regression model adjusted
for air and dew point
temperature.

Daily PM2.5
concentrations
for case and
control days
assigned to
participants
based on ZIP

code of
residence were
used to calculate
overall mean
PM2.5 for the
study location
and period

B-28

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Used average of 2-day lag
(lag 01) PM2.5-

Dominici et al.,
2006

HF and
COPD HA
65+

204 Urban
U.S. counties

Time-
series
study
(MEDICA

RE
enrollees)

Monitored PM2.5
concentrations available
from U.S. EPA AQS for
the period of 1999-2002.

Of the 204 counties
(>200,000 population), 90
counties had daily PM2.5
data across the study
period and the remaining
counties had PM2.5 data
collected once every 3
days for at least 1 full
year. To protect against
consequences of outliers,
used 10% trimmed mean
to calculate daily average
across monitors after
correction for yearly
averages for each
monitor.

Various lags (lag 0,1, 2
days) and distributed lags
assessed and presented.

2-stage Bayesian hierarchical
models to estimate county-
specific, region-specific, and
national-average associations.

Stage 1 model included single
lag and distributed lag over-
dispersed Poisson regression
models to estimate county-
specific risk. Models adjusted
for temperature and dew point

on the same day and the 3
previous days, calendar time to
control for seasonality and other
time-varying influences, daily
numbers of individuals at risk,
and day-of-the-week. In Stage

2, to produce a national
average estimate, Bayesian
hierarchical models were used

to combine RRs across
counties and accounting for
within-county statistical error

and for between-county
variability or heterogeneity. To
produce regional estimates.

The Stage 2 hierarchical
models described above was
used for 7 regions separately.

Daily PM2.5
concentrations
for all 204 U.S.
counties were
used to calculate
an overall mean

PM2.5
concentration for
the study regions
and period.

B-29

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Dominici et al.,
2019

Non-
accidental
mortality

Nationwide

Cohort
study
(MEDICA
RE)

Artificial neural network
that incorporated satellite-

based measurements,
simulation outputs from a
chemical transport model,

land-use terms,
meteorological data, and
other data was used to
predict daily
concentrations of PM2.5 (Di
et al. 2017). Daily PM2.5

concentrations were
predicted for nationwide
grids at1 km2 resolution
for the period 2000-2012.

Survival analyses using the
Andersen-Gill method, a variant
of the traditional Cox
proportional hazards model
C-R relationship assessed
fitting a log-linear model with

thin-plate splines.
Adjusted for individual-level
covariates, county-and ZIP

code-level variables,
meteorological variables, and
other area-level variables.

Average PM2.5
concentrations
over all ZIP
codes in the
continental U.S.
were used to
calculate overall
mean PM2.5 for
the study location
and period.

B-30

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Elliott et al.,
2020

Incident
CVD (Ml
or Stroke)
and all-
cause
mortality

Nationwide

Cohort
study
(Nurses'
Health)

PM2.5 monthly exposures

derived from
spatiotemporal prediction
model (Yanosky, 2014) for
the period of 1988-2008.
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). 24-month
moving average ambient
PM2.5 exposures were
estimated at residential
addresses.

Cox proportional hazard models
used to assess relationship
between 24-month average
PM2.5 exposure and incident
CVD (risk of Ml, Stroke or both)
and mortality. Also modeled
multiplicative interactions
between quintiles of PM2.5
exposure and quartiles of
physical activity (MET-hours)
using stratified Cox hazard
models.

Adjusted for individual-level

covariates including
demographics, education and
SES, marital and retirement
status, behavioral,
comorbidities, diet etc., as well
as area-level SES variables.

Annual (24-
month) average

PM2.5 for the
study participants

was used to
calculate overall
mean for the
study period.

B-31

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Erickson et al.

Non-

Canada

Cohort

PM2.5 exposures derived

Cox proportional hazard models

Average annual

2019

accidental

Nationwide

study

from AOD retrievals using

for the relationship between

PM2.5

, CVD,

(CanCHE

GEOS-Chem calibrated to

PM2.5 and mortality.

concentrations of

and lung

surface measurements by

study participants

cancer

CCHS, m

GWR (van Donkelaar,

Adjusted for individual-level

were used to

mortality

CCHS)

2015). Linked postal

variables including socio-

calculate overall

codes to PM2.5

demographics, SES, education

mean PM2.5

concentrations from grid

and employment, minority and

concentration for

cells. Annual PM2.5

aboriginal identity, and

the study period.

concentrations estimated

contextual ecological

from the postal code and

covariates.

assigned to study

participants based on the

postal code for residence

was used to calculate 3-

year moving average

based on the location and

year of follow-up for years

1998-2012.

B-32

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Erickson et al.,

Non-

Canada

Cohort

PM2.5 exposures derived

Cox proportional hazards

Average annual

2020

accidental

Nationwide

study

from AOD retrievals using

models to examine the

PM2.5

, CVD,

(CanCHE

GEOS-Chem calibrated to

associations between ambient

concentrations of

respiratory

surface measurements by

PM2.5 exposure and non-

study participants

mortality,

GWR (van Donkelaar,

accidental and cause-specific

were used to

and lung

2015). Linked postal

mortality.

calculate overall

cancer

codes to PM2.5

Adjusted for individual-level and

mean PM2.5

concentrations from grid

contextual-level covariates.

concentration for

cells. Annual PM2.5

the study

estimates from the postal

period by

code and assigned to

immigrant status

study participants based

and duration in

on the postal code for

Canada.

residence was used to

calculate 3-year moving

average based on the

location and year of

follow-up for years 1998 -

2016.

B-33

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Eum et al., 2018

All-cause

Geographic

Cohort

PM2.5 concentration

Age-stratified log-linear model

Annual average

mortality

regions:

study

obtained from U.S. EPA's

including offset terms for the

PM2.5 concentrati

"East" of the

(MEDICA

AQS for the period of

size of the population as a base

ons were used to

Mississippi

RE)

2000-2012. Monitoring

model. Also included the

calculate overall

River,

sites with daily

temporal and spatiotemporal

mean

"Center"

measurements for at least

components. Ran base model

concentration for

between the

8 calendar years with

using data for entire 13-year

the study location

Mississippi

each year having 9+

study period (2000-2012) and

(all and by study

River and the

months and with 4+daily

for shorter periods ranging

region) and study

Sierra

measurements included.

between 3 and 12 years and

period.

Nevada

798 sites then were used

compared MRRs to assess

Mountain

to calculate long-term

temporal confounding. In

range, and

concentration (yearly

addition to base model, also

"West" of the

moving average with 350+

assessed temporal confounding

Sierra

days of valid data) using

using three approaches

Nevada

Greven et al. Annual

(decomposition-based, residual-

Mountain

average assigned to

based, and spline models)

range

individuals that lived in

Adjusted for individual-

ZIP codes with centroids

covariates, as well as county-

within 6 miles of a valid

level behavioral covariates, %

monitor.

of non-whites, smoking status,

comorbidities, access to health

care, income, and BMI.

B-34

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Fisher et al.,

Stroke

Nationwide

Time-

Validated national-scale,

Conditional logistic regression

Daily PM2.5

2019

(Self-

stratified

log-normal ordinary

models

concentration on

reported

case-

kriging model for PM2.5

Adjusted for mean daily

the case day

stroke

crossover

were used to estimate

temperature, and stratified

were used to

adjudicate

study

daily PM2.5 concentration.

models to examine effect

calculate overall

d by

(HPFS)

U.S. EPA's AQS data

modification by individual-level

case day PM2.5

physician

used to calculate monitor

characteristics.

mean for the

medical

specific daily averages

study period.

record)

(monitors >=18 hours

measures). These inputs

were then used to

produced kriged surfaces

of daily mean PM2.5

concentrations at the

geocoded residential

addresses of all HPFS

participants for the period

1999-2010.

Lag periods up to 3 days

prior to the stroke event

and a 4-day average used

in model.

B-35

-------
Franklin et al.
2007

All-cause,
CVD, and
Respirator
y mortality

27 U.S.
communities
(with PM2.5
monitoring
and daily
mortality data
for at least 2
years of 6-
year study
period 1997-
2000)

Case-
crossover
study
(NCHS)

Monitored daily PM2.5
concentrations available
from U.S. EPAAQS
(NAMS/SLAMS) for the
period of 1997-2000. Data
for Boston area available
from Harvard University.

To determine which
monitors in the county are
representative of
exposure for a general
population in the county,
correlation was assessed
between monitor pairs
and excluded the monitors
with r<0.8 for 2 or more

monitor pairs. Once
appropriate monitors were
identified then a summary
measure of PM2.5 cone for
the county was calculated
using alternate averaging

method described in
Schwartz 2000 to account
for data availability
variation (daily vs 3-6
days for each monitors in
the county) and calculate
daily average PM2.5 cone
for each of the 27
counties and
corresponding
communities.

2-stage time-stratified analysis:

1) Conditional logistic
regression analysis to generate
community specific estimates;
2) Meta-regression analysis to
combined community specific
estimates to generate overall
pooled effect estimate.

Stage 1 of the model adjusted
for day-of-the-week, as well as
apparent temperature at lagO
and lagl Cases were defined
as "deaths" and control days for
a particular subject were
chosen to be every third day
within the same month and year

that death occurred. Effect
modification of age and gender
was examined using interaction
terms in stage 1, while effect
modification of community-
specific characteristics including

geographic location, annual
PM2.5 concentration > 15 ug/m3
and central AC prevalence was
used in stage 2.

Daily PM2.5
concentrations
for all 27 U.S.
communities over

the period of
1997-2000 were
used to calculate
an overall mean

PM2.5
concentration for
the study location
and period.

B-36

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Calculated and presented
various lags and averages
for PM2.5.

B-37

-------
Franklin et al.
2008

All-cause,
CVD, and
Respirator
y mortality

25U.S
communities
(with PM2.5
monitoring
and daily
mortality data
for at least 4
years of 6-
year period
between
2000-2005)

Case-
crossover
study
(NCHS)

Monitored daily PM2.5
concentrations available
from U.S. EPAAQS
(NAMS/SLAMS) for the
period of 2000-2005. Data
for Boston area available
from Harvard University.

monitor pairs. Once
appropriate monitors were
identified then a summary
measure of PM2.5 cone for
the county was calculated
using alternate averaging

2-stage time-stratified analysis:

1) Conditional logistic
regression analysis to generate
community specific estimates;
2) Meta-regression analysis to
combined community specific
estimates to generate overall
pooled effect estimate.

that death occurred. Effect
modification of age and gender
was examined using interaction
terms in stage 1.

Daily PM2.5
concentrations
for all 25 U.S.
communities over

the period of
2000-2005 were
used to calculate
an overall mean

PM2.5
concentration for
the study location
and period
(overall and by
seasons).

B-38

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Calculated and presented
various lags and averages
for PM2.5.

Gharibvand et
al., 2016

Lung
cancer
incidence

US
Nationwide

Cohort
study
(AHSMOG
-2 study)

Monitor data obtained
from U.S. EPA AQS for
the period of 2000-2001
(2-year prior to start of the
study). Using monitored
PM2.5 data, inverse
distance weighted
interpolations methods,
monthly pollution surfaces
for PM2.5 were created for
the US. Monthly exposure
averages were based on
daily PM2.5
measurements. Only
months with at least 75%
valid data were included

in the exposure
estimation. Participants
were assigned monthly
exposure based on their
baseline residential
address.

Cox proportional hazards model

Covariates included sex, race,
smoking status, years since

participant quit smoking,
average number of cigarettes
per day during all smoking
years, and education level.
Additional covariates included

calendar time, alcohol
consumption, family income,
BMI, physical activity, and
marital status. 3 variables
identified a priori as either as
confounders or effect modifiers:
hours/day spent outdoors,
years of pre-study residence
length at enrollment address,
and moving distance from
enrollment address during
follow-up.

Monthly PM2.5
concentrations

for study
participants were
used to calculate
overall 2-yr mean
PM2.5 for the
study period
2000-2001.

B-39

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Hart et al., 2015

All-cause

Cohort

Monitored data obtained

Cox proportional hazards

Monthly PM2.5

(monitored)

mortality

Nationwide

study

from U.S. EPA AQS for

model.

concentrations

(Nurses'

the period 1999-2006.

for the previous

Health

Monthly average PM2.5

Information on potential

12-months that

study)

concentration calculated

confounders was available

were assigned to

from the nearest

every two years (4 years for diet

study participants

monitoring location for all

information) and each woman

at residence

addresses. The monthly

was assigned updated

locations during

data was again averaged

covariate values for each

the study follow-

to get the previous 12-

questionnaire cycle.

up period were

month moving average at

Confounders examined include

used to calculate

each residential address

age, race, region, season,

overall mean for

prior to mortality.

physical activity, BMI,

the study

Nearest monitor

hypercholesterolemia, family

participants

exposures were validated

history of Ml, smoking history,

included in the

against personal

Current smoking status, diet,

study.

exposures to PM2.5 of

SES (education level,

ambient origin.

occupation of both of the

B-40

-------
Hart et al., 2015
(modeled)

All-cause
mortality

US
Nationwide

Cohort
study
(Nurses'
Health
study)

Spatio-temporal models

(developed using
monitored data from U.S.
EPAAQS, the IMPROVE

network, and also
included meteorological

and GIS-derived
covariates, such as urban
land use within 1 km,
elevation, tract- and
county-level population
density, distance to the
nearest road for road
classes A1-A3 and point-
source emission density
within 7.5 km) was used
to estimate monthly PM2.5
exposures at each
geocoded address.
The monthly data was
again averaged to get the

previous 12-month
moving average prior to
mortality for each
residential address.
Modeled exposures were
validated against personal
exposures to PM2.5 of
ambient origin.

Previous 12-month
moving average of
exposure either from
nearest monitor or spatio-
temporal models were

nurses' parents when she was
16, marital status, and
husband's education if
applicable). Also adjusted for
area-level SES (census tract
level median income and house
value), and long-term temporal
trends.

Risk set regression calibration
for time-varying exposures was
used to correct for bias due to
exposure measurement error in
the hazard ratios of all-cause
mortality using the personal
exposure validation data.

B-41

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

assigned to study
participants.

Hayes et al.,
2020

CVD
mortality

6 U.S. states
(California,
Florida,
Louisiana,
New Jersey,

North
Carolina, and
Pennsylvania
)

and 2 urban

areas
(Atlanta, GA,
and Detroit,
Ml,)

Cohort
study
(NIH-
AARP
Diet and
Health
study)

Modelled (Hybrid land use
regression geostatistical
model developed by Kim
et al. 2017) for the period
of 1980-2010. Mean
annual estimates of
PM2.5 for each census
tract in the U.S. from
spatio-temporal model
were used till 1998. For
period 1999-2010,
monitored U.S.
EPA monitor and
IMPROVE network was
used to derive annual
average estimates.
Annual average PM2.5
concentrations assigned

at census tract level
lagged by 1 year in time-

dependent manner.
Annual PM2.5 exposure
analyzed as continuous
and categorical <8, 8-<12,
12-<20, and 20+ ug/m3
variables.

Cox regression modelling with
time-dependent covariates.

Adjusted for individual-level
variables (age, race/ethnicity,
education, marital status, BMI,
alcohol, and smoking status),
as well as census tract
variables.

Annual PM2.5
concentrations of

the study
participants for
the year 2000
was used to
calculate overall

mean PM2.5
concentration for
the period 2000.

B-42

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Ito et al., 20135

All-cause
mortality

150 U.S.
cities

Time-
series
study

24-hr average PM2.5 mass
data in a given city, and
when data from multiple
monitors were available in
a given city, computed the

average of the daily
values after standardizing
each site's data using the

mean and standard
deviation of the sites data.

Pollutant concentration is
expressed in the model as

a deviation from the
monthly mean to reduce

the influence of the
seasonal cycles of the
pollutants on the overall
associations and help
focus on the short-term
associations.

Poisson regression analysis

First city- and season-specific
Poisson regression was run,
and then city-specific estimates
were combined using random
effects approach

Adjusted for temporal trends
(annual cycles and influenza
epidemics), immediate and
delayed temperature, and day-
of-week pattern, for entire years
(2001-2006) and for warm
(April-September) and cold
(October-March) seasons.

In second stage, assessed
effect modification using land-
use variables and average air
pollution levels.

5 This study is not referenced individually in the 2019 ISA, but is study 3 of the National Particle Component Toxicity (NPACT) Initiative published in HEI
(Lippmann et al., 2013).

B-43

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Jerrett et al.,
2016

IHD
mortality
30+

U.S.
Nationwide

Cohort
study
(ACS
Cancer
Preventio
n Study II)

Multiple exposure
estimation approaches
evaluated within the study
- risk assessment uses

results based on an
ensemble approach that
incorporates chemical
transport modeling, land
use data, satellite data,
and data from ground-
based monitors

Cox proportional hazards
regression

Covariates included current and
former smoking status as well
as smoking duration, amount,
age started, secondhand
cigarette smoke (hours/day
exposed), exposure to PM2.5 in
the workplace for each of the
subject's major lifetime
occupation, self-reported
exposure to dust/fumes at work,
marital status, level of
education, BMI, alcohol
consumption, dietary
vegetable/fruit/fiber index,
dietary fat index, missing
nutrition information. Ecologic
characteristics included median
household income, percentage
of people with < 125% of

poverty-level income,
percentage of persons > 16
who are unemployed,
percentage of adults with < 12th

grade education, and
percentage of population who
were Black or Hispanic.

B-44

-------
Kioumourtzoglo

All-cause

207 U.S.

Open

Monitored data available

2-stage approach for modelling.

Annual PM2.5

u et al., 2016

mortality

cities

Cohort

from U.S. EPA AQS for

concentrations

65+

study
(MEDICA

RE
enrollees)

the period of 2000-2010.
City-specific annual and
2-year PM2.5 averages
was calculated using data
from all available monitors
in each city.

In Stage 1, Cox proportional
hazards model was fit for each
city stratified by age, gender,
race, and follow-up time in
study. Control for slowly varying

potential confounders (e.g.,
SES) and confounders that vary
across subjects, city, and time.

City-characteristics for:
proportion of city population >
65, median household income,
proportion in poverty, proportion

of city families in poverty,
proportion of white, black, and
Asian residents, proportion of
residents with/without high-
school degrees and a college
degree, and city-specific
smoking and obesity rates.
Population-weighted city
averages were developed
based on census data at the
county level. Also included
average annual temperature in

the model.

In stage 2, combined the city-

specific estimates using a
random effects meta-analysis to
generate region-specific effects.
Assessed effect modification by
annual temperature levels, and
population and city
characteristics (greenness,

for 207 cities
during the period
of 2000 to 2010
were averaged to
calculate overall

mean PM2.5
exposure for the
study location (all
and region
specific) and
study period.

B-45

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

poverty, racial composition,
etc.).

Klemm and
Mason, 2003

All-cause
mortality

Harvard Six-
City study
reanalysis

Time-
series
study

24-hour PM2.5
concentration obtained

from Dichotomous
samplers placed at the
central residential
monitoring sites in each of
the six cities. Integrated
24-hour samples were
collected daily for part of
the study periods but were
collected at least every
other day until the late
1980s.

Generalized additive and
generalized linear models

Model adjusted for temporal
trends, day-of-the-week,
weather (average daily
temperature and average daily
dew point temperature).

Daily PM2.5
concentration of
six cities over the
period of 1979-
1988 were used
to calculate
overall mean,
median and
percentiles of
PM2.5 exposure
for the study
location (all and
by study center)
and period.

B-46

-------
Kloog et al.,
2012

ST, LT

CVD HA
Age 65+

New England
Area with 6
U.S. States

Mixed
study
design
(with time
series and

cohort
componen
ts)

Spatiotemporal model:

Used day-specific
calibrations of aerosol
optical depth (AOD) data,
using ground PM2.5

measurements.
Incorporated land use

regressions and
meteorological variables
(temperature, wind speed,
visibility, elevation,
distance to major road,
percent of open space,
point emissions and area
emissions) for the period
of 2000-2006. Model
predicted daily PM2.5
concentrations at a 10 x
10 km spatial resolution.
The PM2.5 concentration
then was matched to ZIP
codes based on spatial

location and date.
Short-term exposure:
used the mean of PM2.5on
the day of admission and
day before admission.
Long-term exposure:
calculated as the mean
exposure in each zip-code
across the 7-year study

period. Short term
exposure was defined as
the difference between
the two-day average and
the long-term average.

Equivalence between Poisson
regression and the piecewise
constant proportional hazard
model to model the time to a
hospital admission as a function
of both long-term and short-
term exposure simultaneously
and enabling simultaneously
examination of short term and
long-term associations with

hospital admissions
(Hierarchical mixed Poisson
regression model).

The model adjusts for
temperature, age, percent
minorities, median income, and
percent of people with no high
school education.

Daily PM2.5
concentration of
all grids within
the NE area for
the acute (0 day
lag) and chronic
(365 day moving
average) were
used to calculate
overall mean
short- and long-
term PM2.5
exposure
respectively, for
the study location
and period.

B-47

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Kloog et al.,
2014

CVD and
COPD HA
Age 65+

7 U.S. Mid-

Atlantic
States and
D.C.

Case-
crossover
design
(MEDICA

RE
enrollees)

Spatiotemporal model:

Used day-specific
calibrations of aerosol
optical depth (AOD) data,
using ground PM2.5

measurements.
Incorporated land use
regression (elevation,
distance to major roads,
percent of open space,
point emissions and area

emissions) and
meteorological variables
(temperature, wind speed,

relative humidity, and
visibility) for the period of
2000-2006. Model used

to predict daily PM2.5
concentrations at a 10 x
10 km spatial resolution.
Daily predicted PM2.5
exposure estimates at
grids were matched to ZIP
codes.

Average of 2-day lag (lag
0 and 1) PM2.5 used.

Conditional logistic regression
analysis

Temperature with the same
moving average as PM2.5 was
included in the model as a

potential confounder.
Study design samples only
cases and compares each
subject's exposure experience
in a time period just before a
case-defining event with the
subject's exposure at other
times, eliminating confounding
(unmeasured or measured) that
do not vary over time. Cases
were matched on day of the
week and defined the relevant
exposure time window as the
mean exposure of the day of
and day before the patient's
hospital admission. Effect
modification: 1) assessed
whether subject residence
within 30 km of a monitor or
farther modified the PM2.5
association; 2) examined
interaction between exposure
and income level and gender.

2-day moving
average of PM2.5
concentration of
all grids within
the mid-Atlantic
states were used

to calculate
overall mean (all
area and
rural/urban
areas) PM2.5
exposure for the
study location
and period.

B-48

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Krall et al., 2013

All-cause
mortality

72 Urban
U.S.
Communities

Time-
series
study
(NCHS)

Monitored data available
from U.S. EPA AQS for
the period of 2000-2005.

Excluded data from
source-oriented monitors

that may not be
representative of typical
population exposures.
Daily community-level
pollutant exposure as the
arithmetic mean of daily
monitor observations
within the community. For
communities with single

monitor pollutant
concentration represented
concentrations recorded
by that monitor.

Used lag 1 PM2.5 in
model.

Log-linear Poisson Regression
Model

Model adjusted for temperature
and previous day's
temperature, long-term and
seasonal trends, age, and day-
of-the-week. Also included
interaction term for pollutant
concentration and seasons.

Daily PM2.5
concentration of
72 U.S. urban
communities
were used to
calculate overall

mean PM2.5
exposure for the
study location
and period.

B-49

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Krall et al., 2018

ED Visits

Multi-city (5

Time-

PM2.5 concentrations

Poisson time- series regression

Daily (24-hr) PM

for CVD

Metropolitan

series

obtained from ambient

model accounting for over-

concentrations

(CHF,

areas)

study

monitoring stations

dispersion (Peng et al. 2009;

for specific cities

Cardiac

(Electronic

located within each of the

Krall et al. 2013) to calculate

were used to

dysrhythm

billing of

metropolitan areas were

city specific associations. To

calculate overall

ia, IHD,

ED visits)

fused with Community

calculate overall and posterior

PM2.5

Stroke) or

Multi-Scale Air Quality

city-specific associations,

concentration by

model estimates (Friberg

applied Bayesian hierarchical

city for the study

(asthma/w

et al, 2016, 2017) to

models (Everson and Morris

period.

heeze,

obtain population-

2000).

COPD,

weighted average

pneumoni

estimates of the 24-hour

Adjusted for weekday, season,

a, URI)

average PM2.5

holidays, metrology, temporal

concentrations.

trends.

B-50

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Lavigne et al.,

Non-

24 Canadian

Case-

Daily (24-hour) average

Conditional logistic regression

Daily PM2.5

2018

accidental

cities

crossover

PM2.5 concentrations

analysis.

concentrations in

, CVD,

study

obtained from monitors in

Performed stratified analyses

24 Canadian

and

Canada's NAPS network

examining the relationship

cities were used

respiratory

and were used to estimate

between PM2.5 and mortality

mortality

PM2.5 concentrations for

across tertiles of Oxidant

calculate overall

the period of 1998-2011.

capacity.

mean PM2.5

Exposure estimates were

concentration

assigned to each study

over the study

participant based on the

location and

monitoring station(s)

period.

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.

B-51

-------
Lee et al„ 2015

All-cause,
Cardiovas

cular,
respiratory
mortality

3 U.S.
Southeast
States

Case-
crossover
design
(Dept. of
Pub
Health
data)

Spatio-temporal model
that used satellite AOD
data to predict daily PM2.5
at 1X1 km resolution for
the period of 2007-2011.
Daily PM2.5 concentration

at 1km grids were
aggregated into the ZIP
code level. For this, 1 km
grid cells were matched to

ZIP code area by
assigning the centroid of
each 1 km grid cell to the
centroid of the closest ZIP
code. ZIP code areas that
contained one or more 1
km grid cells were given
the averaged PM2.5 and

ZIP codes that were
smaller than 1 km2 were
given the predictions from
the closest grid cell.

Finally, PM2.5
concentrations from ZIP
codes were assigned to
the study participants
based on their residence
ZIP code and for specific
days.

Conditional logistic regression

Model adjusted for temperature
and day of the week

Also ran stratified analysis by
age, sex, race, education, and
primary cause of death.

Analysis also restricted for ZIP
codes where annual average of
PM2.5 <12 or daily average <35
separately.

Sensitivity analysis: potential
non-linear relationship between
temp and mortality modelled
using natural spline to the
temperature term.

Daily PM2.5
concentrations
for ZIP codes in

the 3
Southeastern
states were
averaged to
calculate the
overall mean
PM2.5
concentration (all
states and by
state)

For sensitivity: Daily
monitored PM2.5
concentrations from the

nearest EPA and
IMPROVE monitors from

B-52

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

resident ZIP code (no
distance limit) were
identified and assigned to
individuals.

Used lagO and Iag1 in
model.

B-53

-------
Lefler et al.
2019

All-cause
mortality,
Cardiopul
monary
mortality

Nationwide

Cohort

study

(NHIS)

Annual average PM2.5 was
modeled using regulatory

monitors and land use
data as described in (Kim,
2018). PM2.5 exposure

prior to 1999 were
estimated using PM10
data. Estimates for each
pollutant-year through
2015

were generated at the
census-block level using

year-2010
Census block centroids.
Tract-level estimates for
year 2000 Census tracts
and year-2010 Census
tracts were estimated by

mapping year-2010
Census blocks to census
tracts and then calculating
a population-weighted
average of the census
blocks within a census
tract. PM2.5 exposure
estimates were assigned
to home census tracts as
either 2-year (i.e., cohort
year and previous year) or
5-year (i.e., cohort year
and previous 4 years)

average PM2.5
concentrations, 17-year

average PM2.5
concentrations (1999 -
2015), or 28-year average

Cox hazard model 2 versions:
Basic PH model, and complex
PH model using
SURVEYPHREG.

Basic model adjusted for age,

sex, and race/ethnicity.
Complex model adjusted for
complex survey design. Both
models controlled for marital
status, household income,
education, smoking status, BMI,
urban/rural, census regions and
survey year.

Annual PM2.5
concentration
were for
participants were
used to calculate

overall mean
concentration for
thel7-year study
period 1999-
2015.

B-54

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

PM2.5 concentrations
(1988-2015).

Lepeuleet al.,
2012

All-cause,
Cardiovas
cular, lung
cancer
mortality

HARVARD 6
cities

Prospectiv
e

Cohort/Lo
ngitudinal
follow-up

study
(HARVAR
D 6 cities
data)

PM2.5 data from monitors
in the participant's city.

PM2.5 data 1979-
1986/1988 from monitors,
end of monitoring to 1998

estimated from PM10
using U.S. EPA monitors,
1999-2009 direct PM2.5
measurement from U.S.
EPA monitors. 1-yror 1-
3yr or 1-5 yr. moving
PM2.5 averages were
assigned to participants
based on city of
residence.

Cox proportional hazard
models, Poisson survival
analysis

Stratified analysis by sex, age,

and time in the study (1-yr
interval). Confounders included:

Baseline information on
smoking status, smoking pack-
years, education, linear and
quadratic term for BMI.

Also explored effect
modification of PM2.5 on
mortality by smoking status at
enrollment, as well as time
period in study.

B-55

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Liu et al., 2019

All-cause

and
cause-
specific
mortality

107 U.S.
Cities

Time-
series
study
(MCC
Collaborat

ive
Research
Network)

Monitored PM2.5
concentration obtained
from MCC database for
the period of 1987-2006.
Hourly data was used to
calculate 24-hr daily
average. Daily PM2.5
concentrations were
averaged across stations
within each city. Finally, 2-
day moving average for
the city was calculated.
2-day moving average
(Iag01) was used in
model.

Used two-stage analytic
protocol, which had been
developed and widely applied in
previous
multicity time-series studies.
First stage estimated city-
specific association using
quasi-Poisson generalized
additive models. Second stage
used random-effects models to
pool the estimates of the city-
specific associations. Two-
stage regional analysis was
also performed by WHO
regions.

Also explored the shape of the
relationship using C-R curves
with PM term appearing with a
B-spline function with two knots
at 25th and 75th percentiles.

Daily PM2.5
concentration
were used to
calculate overall

mean
concentration for
the study location
and period.

Loop et al., 2018

Incident
total CHD,
incident
CHD
deaths,
incident
nonfatal
Ml

Southeastern
States (the
Stroke Belt
Region)

Cohort
study
(REGARD
S)

Daily PM2.5 concentration
estimated for 10 km2 grids
using ground-level
monitoring data and
satellite measurements of

AOD. Annual average
PM2.5 concentrations were

linked to geocoded
residential addresses at
baseline.

Cox proportional hazards
models

Model adjusted for 1-year mean
temp, season, race, region,
urbanicity, income, education,
age, gender, pack-years, BMI,
alcohol use, physical activity,
calendar year, medication use,
diabetes.

Annual PM2.5
concentrations
assigned to study
participants were
used to calculate

overall PM2.5
concentration for
the study period.

B-56

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Malig et al.,

Respirator

35 CA

Case-

PM2.5 data obtained from

County-level conditional logistic

2013

counties

crossover

California Air Resources

regression analysis. Overall

morbidity

(9 counties

design

Board. Same day lag and

estimate was then calculated by

(Asthma

included for

(CA Office

various days lags average

combining county-level

and

PM2.5

were calculated for PM2.5.

estimates using a random-

COPD ED

analysis)

Statewide

Participants were

effects meta-analysis

and HA)

Health

assigned exposure from

Planning

the closest monitor from

Time-invariant confounders and

and

the residential population-

seasonal trends were controlled

Developm

weighted ZIP code

for given the study design.

ent Data)

centroid. Only participants

living in ZIP codes within

Other confounders included in

20 km of PM2.5 monitors

the models were: other gaseous

were included to increase

pollutants including ozone,

validity of pollution

linear and squared term for

exposure metrics.

daily average temperature.

Stratified analysis also by

distance to monitor within 10

km vs. 10-20 km

B-57

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Malik et al.,
2019

All-cause
mortality
(in people
presenting
with Ml)

Multi-city (31
US hospitals)

Cohort
study
(TRIUMP

H and
PREMIER
cohorts)

Daily average PM2.5
concentration estimated

using U.S. EPA's
downscaled CMAQ, which
is a Bayesian space-time
fusion model that bias
corrects modeled output
data with monitored for
the period of 2002-2007
(Berrocal, 2012). PM2.5
concentration at census
tract centroid of patient
was then estimated and
annual average PM2.5
concentrations calculated
for the year prior to
myocardial infarction.

Generalized additive models
with Gaussian errors for SAQ
and SF-12 PCS scores and
using a proportional odds
logistic regression model for
dyspnea scores. Smoothing or
restricted cubic splines were
used to allow for non-linear
associations. Also examined
the association of 12-month avg
air quality parameters prior to
Ml with all-cause mortality for 5
years follow-up using Kaplan-
Meier methods and Cox hazard
models.

Adjusted for individual-level
variables (age, race/ethnicity,
sex, smoking, SES, date of
enrollment to account for
temporal or seasonal effects,
comorbidities).

The annual
average PM2.5
concentrations

for the
participants were
used to calculate

overall PM2.5
concentration for
the study period.

B-58

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

McConnell et al.,
2010

Asthma
Incidence

13 CA
communities

Cohort
Study
(CHS)

PM2.5 concentration data
measured in central site

monitors in each
community (for 9 of 13
communities since 1994
and others different time

period). This study
considered 2003-2004
PM2.5 measurements at
each community monitor.
Average annual PM2.5
concentration from each
community was assigned

to study participants
based on their community
of residence.

Multi-level Cox proportional
hazard model accounting for
residual variation in time to
asthma onset and clustering of
children around schools and
communities

Models adjusted for:
secondhand smoke, pets in
home, race/ethnicity, age at
study entry, sex, and random
effects for community and
school.

Average annual

PM2.5
concentrations
assigned to study
participants were
used to calculate
overall mean
PM2.5 exposure
for the study
location and
period.

B-59

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Miller et al.,

CVD

Nationwide

Cohort

PM2.5 data obtained from

Cox proportional hazard model

Average annual

2007

incidence

study

U.S. EPA's Aerometric

was used to assess the

PM2.5

(Ml,

(WHI)

Information Retrieval

relationship between long-term

concentrations

revascular

System AirData for the

PM2.5 exposure and the first

assigned to study

ization,

year 2000 (exposure year

CVD event.

participants were

stroke,

of interest). Participants

used to calculate

death

were assigned annual

Models adjusted for: age, BMI,

overall mean

from CHD,

exposure from the closest

smoking status, SBP,

PM2.5 exposure

CBVD)

monitor to the residential

educational level, household

for the study

address.

income, race/ethnicity,

location and

presence of diabetes,

period.

hypertension or

hypercholesterolemia, also

tested for other additional

confounders in extended

models. Stratified analysis

diabetes, age and BMI.

B-60

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Ostro et al.,

Asthma

Case-

PM2.5 (24-hour average)

County-level conditional logistic

Daily PM2.5

2016

and

metropolitan

crossover

data obtained from U.S.

regression analysis. Overall

concentrations

COPD ED

areas/countie

design

EPA provided by

estimate was then calculated by

for all 8

s in CA

(CA Office

California Air Resources

combining county-level

metropolitan

Board for the period of

estimates using a random-

counties over the

Statewide

2005-2009. Participants

effects meta-analysis

period of 2005-

Health

were assigned exposure

2009 were used

Planning

from the closest monitor

Time-invariant confounders and

to calculate an

and

from the residential

seasonal trends were controlled

overall mean

Developm

population-weighted ZIP

for given the study design.

PM2.5

ent Data)

code centroid. Only

concentration for

participants living in ZIP

Other confounders included in

the study location

codes within 20 km of

the models were: linear and

and period.

PM2.5 monitors were

squared term for lagO

included to increase

temperature, day of the week.

validity of pollution

exposure metrics.

Used lagO, Iag1 and Iag2

in model.

B-61

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Pappin et al.,
2019

Non-
accidental
mortality

Canada
Nationwide

Cohort
study
(CanCHE
C)

PM2.5 exposures derived
from AOD retrievals using
GEOS-Chem calibrated to
surface measurements by
GWR (van Donkelaar,

2015). 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
(Meng, 2019). Annual
M2.5 estimates from the
postal code and assigned

to study participants
based on the postal code
for residence was used to
calculate 3-year moving
average based on the
location and year of
follow-up.

Cox Hazard model and DAG
approach. Also performed C-R
analysis using a 3 step
approaches: (1) fit the data
using restricted cubic splines
(RCS) with a large number of
knots; (2) smooth potential
erratic predictions from the
large number of knots using
monotonically increasing
smoothing splines (MISS); and
(3) fit the shape constrained
health impact function (SCHIF)

to the MISS predictions.
Cox model stratified by age,
sex, and immigration status
separately by CanCHEC
cohorts. Two covariate
adjustment models. First based

on DAG and controlled for
airshed, urban form, CMA/CA
size. Second model "full" model
adjusted for individual-level
variables (income, education,
occupation, marital status).

The annual
average PM2.5
concentrations
were used to
calculate overall

mean PM2.5
concentration for
the study cohorts
and periods.

B-62

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Peng et al.,
2009

CVD HA
Age 65+

119 U.S.
Urban
counties>150,

000
populations

Time-
series
analysis
(MEDICA

RE
enrollees)

PM2.5 data (daily or every

3 days) obtained from
U.S. EPA'sAQSand STN
for the period of 2000-
2006. Countywide PM2.5
total mass concentration

was calculated by
averaging the daily PM2.5
values from all the
monitors in a county.

Used lagO, Iag1 and Iag2
in model.

Log-linear Poisson Regression
analysis

Adjusted for potential
confounders including weather,
day of the week, unobserved
seasonal factors. In county-
specific regression model,
following indicators were
included: indicator for the day of
the weeks, a smooth function of

time per calendar year to
control for seasonality and long-
term trends, a smooth function
of current-day temperature, a
smooth function of the 3-day
running mean temperature, a
smooth function of current-day
dew-point temperature, and a
smooth function of the 3-day

running mean dew-point
temperature. To model smooth
functions, we used a natural
spline basis.

Daily PM2.5
concentrations

for all 119
counties over the
period of 2000-
2006 were used
to calculate an
overall median

PM2.5
concentration for
the study location
and period.

B-63

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Pinaultet al.,
2016

All-cause,
CVD and
lung
cancer
mortality

Canada
Nationwide

Cohort
Study
(mCCHS)

PM2.5 exposures at 1 km2
resolution from 2004-2012
derived from a national
model that relates AOD
retrievals (MODIS) to
near-surface PM2.5 using
a GEOS-Chem model and

further calibrated to
surface measurements by
GWR (van Donkelaar,

2015). PM2.5
concentrations extended
back to 1998 (1998-2003)
by applying interannual
variation of a previously
published PM2.5 dataset
(Boys, 2014). Annual
PM2.5 estimates from the
postal code were
assigned to study
participants based on the
postal code for residence
and used to calculate a 3-

year moving average
based on the location and
year of follow-up for years
1998-2012.

Cox proportional hazards
models

Models were stratified by age
(5-yr interval) and sex. Models

adjusted for individual
socioeconomic covariates and
behavioral (BMI, smoking and
alcohol consumption, fruit, and

vegetable consumption)
covariates, ecological variables
including neighborhood
socioeconomic status (both
social and material deprivation).

Annual 3-year
PM2.5 average
concentration for

the study
participants were
used to calculate

overall PM2.5
concentration for
the study period.

B-64

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Pinaultet al.,
2017

Non-
accidental

, CVD,
respiratory
and lung
cancer
mortality

Canada
Nationwide

Cohort
study
(CanCHE
C)

PM2.5 exposures at 1 km
resolution from 2004-2012
derived from a national
model that related AOD
retrievals (MODIS) to
near-surface PM2.5 using
GEOS-Chem model and

further calibrated to
surface measurements by
GWR (van Donkelaar,

year moving average
based on the location and
year of follow-up for years
1998-2012.

Cox survival models. Also
estimated Shape Constrained
Health Impact Functions (a
concentration-response
function) for selected causes of
death.

Adjusted for individual
demographic and
socioeconomic variables at
baseline (on Census day):
Aboriginal identity, visible
minority status, marital status,
educational attainment, income
adequacy quintile, and labor
force status, and contextual
variables at the census division
scale.

The annual 3-
year moving
average PM2.5
concentrations

for study
participants were
used to calculate
overall mean
PM2.5
concentration for
the study period

B-65

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Pinaultet al.,
2018

CVD
mortality

Canada
Nationwide

Cohort
study
(CanCHE
C, mCCH
S)

PM2.5 exposures at 1 km2
resolution from 2004-2012
derived from a national
model that related AOD
retrievals (MODIS) to
near-surface PM2.5 using
GEOS-Chem model and

further calibrated to
surface measurements by
GWR (van Donkelaar,

to study participants
based on the postal code
for residence and used to
calculate a 3-year moving
average based on the
location and year of
follow-up for years 1998 -
2012.

Cox proportional hazard
models.

Considered co-occurring
diabetes with and without other
contributing causes of death:
hypertension, dementia or
Alzheimer's disease, and
chronic kidney disease, as

these comorbidities are
medically related to diabetes.
Also considered diabetes status
at baseline as effect modifier
using CCHS-mortality cohort.
Adjusted model for individual-

level variables (aboriginal
identity, visible minority status,
education, labor force status
and income adequacy), and
neighborhood-level variables.

The annual 3-
year moving
average PM2.5
concentrations

for study
participants were
used to calculate
overall mean
PM2.5
concentration for
the study period

B-66

-------
Pope et al.,

All-cause,

U.S.

Cohort

Monthly exposure to PM2.5

Cox proportional hazards

Monthly mean

2015

IHD

Nationwide

study

was estimated by linking

models

PM2.5

mortality

(ACS

geocoded home

concentration for

(30+)

Cancer

addresses

The individual-level covariates

study participants

Preventio

of the study participants to

incorporated in the models

were used to

n Study II)

ambient PM2.5

included 13

calculate overall

concentrations derived

variables that characterized

mean

using

current and former smoking

concentration for

a national-level hybrid

habits (including

the study period.

land use regression (LUR)

smoking status of never,

and Bayesian

former, or current smoker,

Maximum Entropy (BME)

linear and

interpolation model (LUR-

squared terms for years

BME) that incorporated

smoked, and cigarettes smoked

data from ground-based

per day, indicator

monitors for the study

for starting smoking at aged

period of 1999-2004.

<18 years, and pipe/cigar

smoker).

1 continuous variable that

assessed exposure to second-

hand cigarette

smoke (hours/d exposed); 7

variables that reflected

workplace PM2.5

exposure in each subject's main

lifetime occupation; a variable

that

indicated self-reported

exposure to dust and fumes in

the workplace;

variables that represented

marital status

(separated/divorced/widowed

or single versus married);

variables that characterized the

level

B-67

-------
B-68

of education (high school, more
than high school versus less
than

high school); 2 body mass
index variables (linear and
squared terms
for body mass index); variables
that characterized the
consumption
of alcohol (beer, missing beer,
wine, missing wine, liquor, and
missing
liquor); and variables that
indicated quartile ranges of

dietary fat
index and quartile ranges of a
dietary vegetable/fruit/fiber
index.

Ecological covariates included
median household income;

percentage
of people with <125% of
poverty-level income;
percentage of
unemployed individual aged
>16 years; percentage of adults
with

<12th grade education; and
percentage of the population
who were
black or Hispanic. These
ecological covariates were

included in the
models using both ZIP code
level data and ZIP code
deviations from

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

the county means.

Pope et al.,
2019

All-cause

and
cause-
specific
mortality

Nationwide

Cohort
study

PM2.5 concentration
estimated for census
block using regulatory
monitoring data from
1999-2015 within a
universal kriging
framework employing
land-use regression
methods and other
variables (Kim 2018).
Pop-weighted annual
averages were
calculated for all 17 years
for each 2000 and 2010
census tract. Individual

were assigned air
pollution cone based on

their census tract of
residency at the time of
survey, e.g.: using year-
2000 census tract for
individuals surveyed
1986-2010 and using
year-2010 census tract for

individuals surveyed
2011-2014. For primary
analysis: PM2.5 exposure

is an average
concentration over the 17
yrs.

Cox Hazard models. Ran 2
models: one accounting for
complex survey design and
sampling strategy including
sample weights
(SURVEYPHREG) and another
without accounting for complex
survey design (PHREG). Ran
model using full-cohort and sub-
cohort with additional data on
BMI and smoking. The shape of
the PM2 5—mortality relationship

was also explored using an
integrated modeling approach.

Adjusted for age, sex, and race-
ethnicity, income inflation-
adjusted to 2015, education
levels, marital status, urban vs
rural, U.S. regions, survey
years, smoking status.

Annual PM2.5

average
concentrations
over the 17-years
(1999-2015)
were used to
calculate overall

PM2.5
concentration.

B-69

-------
Puett et al.
2011

All-cause
mortality

and
incidence
(Ml and
stroke)

13 Northeast
and Midwest
US States

Cohort
study
(Health
Profession
als)

PM2.5 monthly exposures

derived from
spatiotemporal prediction
model (Yanosky, 2008,
2009) for the period of

1988-2002. Model
predictions used GIS-
based spatial smoothing
model and include
pollution data from
monitoring sites in the
U.S. EPAAQS (U.S. EPA
2009), Visibility
Information Exchange
Web System (VIEWS
2004), the Interagency
Monitoring of Protected
Visual Environments
(IMPROVE) network,
Stacked Filter Unit (a

predecessor to
IMPROVE), Clean Air
Status and Trends
(CASTNet) networks, and
Harvard research studies,
as well as GIS-derived
covariates including
population density,
distance to nearest road,
elevation, urban land use,
point-source and area-
source primary emissions,
and meteorological
information. 24-month
moving average ambient
PM2.5 exposures were

Time-varying Cox proportional
Hazard model to assess the
relationships between all-cause
mortality and CVD incidence
with annual average PM2.5
exposure.

Models were adjusted for year
(linear term), season (indicator

variables), and for state of
residence (indicator variables).

Annual PM2.5

average
concentrations
for the study
participants for
the baseline
study year (i.e.,
1989) were used
to calculate
overall PM2.5
concentration for
the study
location.

B-70

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

estimated at residential
addresses.

Shi et al., 2016

ST and LT

Total
mortality
(65+)

New England
Area with 6
U.S. States

Open
Cohort
study
(MEDICA

RE
enrollees)

Daily PlVtafor the New

England area was
predicted at 1-km2 spatial
resolution from novel 3-
stage statistical models
for the period of 2003-
2008.

365-day moving average
(for long-term exposure)
and average Iag0-1 (for
short-term exposure) were
calculated for each grid
cell. The long-term and
short-term averages at
grid-cells were matched to
ZIP codes by linking the
ZIP code centroid to the
nearest PM2.5 grid.
Participants were
assigned PM2.5
concentrations based on
the ZIP codes of
residence.

Used Iag0-1 average for
short-term exposure
analysis in model.

Chronic effects of air pollution

assessed using Cox
proportional hazard models.
Acute effects of air pollution
assessed using Poisson log-

linear models.

Both acute and chronic effects
were assessed using Poisson

survival analysis. Analysis
performed in full-cohort as well
as low exposure cohorts.

Poisson survival models were
adjusted for smooth function of
time, temporal covariates such
as temperatures and day of the
week, spatial covariates such
as ZIP code-level socio-
economic variables.

Long-term
average:
Average annual

PM2.5
concentrations of
all grid cells in
the study area
were used to
calculate overall

mean PM2.5
exposure for the
study location
and period.

Short-term
average: Lag01

PM2.5
concentrations of
all grid cells in
the study area
were used to
calculate overall

mean PM2.5
exposure for the
study location
and period.

B-71

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Shin et al., 2019

AF and
Stroke (1st
HA)

Ontario,
Canada

Cohort
study
(ONPHEC
)

PM2.5 concentrations
estimated using AOD and
PM2.5 simulated by the
GEOS-Chem chemical
transport model (i.e.,
individual's exposure in
2001 was estimated as
mean exposure from
1996-2000). Final surface
with 1 x 1 km resolution

was generated for
Ontario. Annual PM2.5
estimates was calculated
for the postal code and

assigned to study
participants based on the
postal code for residence.
PM2.5 concentration was
used to calculate 5-year
moving average based on
the location and year of
follow-up.

Cox proportional hazards
models.

Adjusted for individual-level
variables (age and sex),
neighborhood-level SES
variables, and geographic
indicators.

The 5-year
moving average

PM2.5
concentrations

for study
participants were
used to calculate
overall mean
PM2.5
concentration for
the study period

B-72

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Shin et al., 2021

All-cause
hospitaliza
tion and
all-cause
mortality

22 Canadian
cities

Time
series
study
(Statistics
Canada)

Daily (24-hour) average
PM2.5 concentrations were
calculated for each study

city using ambient
monitoring data available
from Canada's NAPS for
the period 2001-2012.
Daily PM2.5 concentrations
were averaged across
monitors within a city
when multiple monitoring
sites were present.

Generalized additive Poisson
model and Bayesian
hierarchical model. Static
approach to estimate the

nationwide overall
associations between air
pollution and health outcomes
for all

years combined. A two-stage
hierarchical model was
employed: firstly, a
generalized additive Poisson
model for city-specific
associations
between individual health
outcomes and individual air

pollutants,
respectively, and secondly a
Bayesian random effects model
to

combine the city-specific
associations to obtain
nationwide associations.

Daily PM2.5
concentrations of

22 Canadian
cities were used
to calculate
overall mean
OM2.5
concentration for
the study location
and period.

B-73

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Stieb et al.,
2009

Cardiac

and
Respirator
y ED visits

Seven
Canadian
Cities

Time
series
study
(Hospital
cases)

PM data obtained from
National Air Pollution
Surveillance (NAPS)
system for the period of

the 1990s and early
2000s. City averages of
the PM2.5 exposure were
calculated by averaging
all monitoring stations
within the city.

Used lag 0, land 2, in
model.

Generalized Linear Models with
natural spline functions of time
to adjust for seasonal cycles in
air pollution and health

Confounders included: Mean
daily temperature and relative
humidity at lag 0,1, and 2 days,
day of the week and holidays.

Daily PM2.5
concentrations of
the cities were
used to calculate
the overall mean
PM2.5 exposure
for the study
location (by site)
and study period.

Sun et al., 2019

Incident
stroke:
Total, HS
and IS
(self-
reported)

Nationwide

Time-
stratified
case-
crossover
(WHI)

PM2.5 concentration
obtained from log-normal
ordinary kriging model as
previously described (Liao
et al., 2006). This model

estimates daily air
pollutants at each address

based on weighted
average of measurement
from nearby monitors
(Legendre and Fortin,
1989). Daily mean PM2.5
concentrations were
estimated atgeocoded
participant address for the
period of 1993-2012.

Conditional logistic regression.

Adjusted for time-varying
variables (daily mean ambient
temperature, dew point
temperature, and relative
humidity)

Various lags assessed (1-day
moving average to 6-day
moving average)

Daily PM2.5
concentrations
on the case days

for the
participants were
used to calculate
the overall mean

PM2.5
concentration for
the study period.

B-74

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Szyszkowicz,
2009

Angina
ED

Seven
Canadian
Cities

Time
series
study
(Hospital
cases)

PM data obtained from
National Air Pollution
Surveillance (NAPS)
system. City averages of

the exposure were
calculated by averaging
stations within the city.

Used lag 0,1 and 2, in
model.

Generalized Linear Mixed
models

Models adjusted for
meteorological variables such

as relative humidity,
temperature, and atmospheric
pressure (a daily 24-hr average

measurements were
calculated). Temperature and
relative humidity in models were
represented by natural splines.
Stratified analysis by season as
well as combined for the whole
period.

Daily PM2.5
concentrations of
the cities were
used to calculate
the overall mean
PM2.5 exposure
for the study
location (all and
by cities) and
study period.

B-75

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Thurston etal.,
2016a

All-cause,
CVD and
respiratory
mortality

6 U.S. States
and 2 MSAs

Cohort
study
(NIH_AAR
P cohort)

PM data obtained from
U.S. EPA AQS for the
period of 2000-2008.
Census-tract estimates
were generated using
hybrid LUR and BME
models, which were
combined to generate
monthly estimates of
PM2.5. Participants
exposure was estimated

at census-tract of
residence and included
annual mean
concentration in prior year
of mortality.

Cox proportional hazard models

Stratified analysis by age, sex,
regions (6 states and 2 MSAs).
Confounders adjusted included:
race, education, marital status,
BMI, alcohol consumption,
smoking history, contextual
variables such as median
household income and % pop

with less than high school
education. Several interactions

between PM2.5 and socio-
demographics were also tested.

Average annual

PM2.5
concentrations in
the prior year at
the census tract
of residence over

the follow-up
period 2000-2008

were used to
calculate overall

mean PM2.5
exposure for the
study
participants.

B-76

-------
Turner et al.
2016

Lung
cancer
mortality
(30+)

U.S.
Nationwide

Cohort
study
(ACS
Cancer
Preventio
n Study II)

Estimated PM2.5
concentrations were
obtained using a national-
level hybrid land use
regression (LUR) and
Bayesian
maximum entropy (BME)
interpolation
model. Monthly PM2.5
monitoring data were
collected from 1,464
sites from 1999 through
2008, with 10%
reserved for cross-
validation. The base LUR
model that predicted PM2.5

concentrations
included traffic within 1 km
and green space within
100 m3. Residual
spatiotemporal variation in
PM2.5 concentrations was
interpolated with
a BME interpolation
model. The two estimates
were then combined. The
cross validation
R2 was approximately
0.79. Mean PM2.5 (1999-
2004) concentrations
were used here.

Cox proportional hazards model

Models were adjusted for
education; marital status; BMI
and BMI squared; cigarette
smoking status; cigarettes per

day and
cigarettes per day squared;
years smoked, and years
smoked squared; started
smoking at younger than 18
years of age; passive smoking
(hours); vegetable, fruit, fiber,
and fat intake; beer, wine, and

liquor consumption;
occupational exposures; an
occupational
dirtiness index; and six
sociodemographic
ecological covariates at both
the postal code and postal code
minus county-level mean
derived from the 1990 U.S.
Census (median household

income and percentage
of African American residents,
Hispanic residents, adults with

postsecondary education,
unemployment, and poverty).

Potential confounding examined
by elevation, MSA size, annual

average daily maximum air
temperature, mean county-level
residential radon

B-77

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

concentrations, and 1980
percentage of air conditioning.

Urman et al.,
2014

Lung-
function
decline

8 Southern
CA

communities/
counties

Cohort

study

(CHS)

Central monitors in each
community provided data
on air pollutants. Each

child was assigned
exposure based on the
child's resident
community.

Linear Regression model

Models were adjusted for
demographic, socio-economic
and anthropometric variables
(BMI, height), study community.

Wang et al.,
2017

Total
mortality
(65+)

7 U.S.
Southeast
States

Cohort
study
(MEDICA

RE
enrollees)

Three stage Hybrid model
to predict daily PM2.5
concentration at 1km X
1km resolution for the
period 2000-2013. These
daily concentrations were

used to calculate the
annual average of PM2.5
for each grid cell. The
annual average PM2.5
concentrations of all grid
cells in each of the ZIP
code tabulation area
(ZCTA) were then used to
calculate ZCTA specific
arithmetic means PM2.5.

Participants were
assigned annual averages
of PM2.5 based on their
ZCTA of residence.

Cox Proportional hazard
models

Models were stratified by age
groups, sex, race. Adjusted for
variables: year of enrollment,
previous admission due to CHF,

COPD, Ml and diabetes,
numbers of days spent in ICU
and CCU, state, ZCTA level
socio-demographic variables
such as % pop below poverty,
urbanicity, lower education,
median income and median
home value, and behavioral
variables such as % smokers
and obesity at county level.
Further model also included

yearly mean summer
temperature at ZCTA level.

Average annual

PM2.5
concentrations of

ZCTAs were
used to calculate
overall median
PM2.5 exposure
for the study
location (overall
and by state),
and period
(overall and by
year).

B-78

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Wang et al.,
2020

Non-
accidental
cause-
specific
mortality
(Resp,
CVD,
cancer)

Nationwide

Cohort
study
(MEDICA
RE)

Daily PM2.5 was estimated
on a 6-km grid using a
spatio-temporal model
described in (Yanosky,
2014) for the period of
2000-2008. Model inputs
included monitored PM2.5,

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-
month period prior to
death.

Cox hazard models. Also fit
models using restricted cubic
splines (RCS) with three knots
to characterize non-linearity.
Effect-modification assess for
age, sex, race and urbanicity.
Adjusted for SES variables.

Annual average

PM2.5
concentration for
participants were
used to calculate
overall annual m

ean PM2.5
exposure for the
study period.

B-79

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Weichenthal et
al., 2016c

Asthma

and
COPD ED

15 cities in
Ontario

Case-
crossover
Design
(cases
extracted

from
NACRS
database)

Daily average
concentration of PM2.5
collected from fixed-
monitoring stations for the
period of 2004-2011 in
Ontario, which is part of
Canada's National Air
Pollution Data. PM data
obtained from 19 sites
located in 15 cities. 2
years of data available for
3 cities and remaining had

5-8 years of daily air
pollution data. Case and
control days of study
participants were
assigned PM2.5
concentration based on
the city of residence and
based on monitoring
station closest to the
population-weighted
centroid of each subject's
3-digit postal code (if

multiple monitors
available in participants
city such as Toronto and
Hamilton).

Various lags assessed:
lagO, Iag1, Iag2 and mean
of lagO-2.

Conditional logistic regression
models

Models adjusted for 3-day
mean temperature and relative
humidity using cubic splines.

Daily PM2.5
concentrations in
Ontario over the
period of 2004-
2011 were used
to calculate the
overall mean
PM2.5 exposure
for the study
location and
period.

B-80

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Weichenthal et

Ml ED

16 cities in

Case-

Daily average

Conditional logistic regression

Daily PM2.5

al., 2016b

Ontario

crossover

concentration of PM2.5

models

concentrations in

Design

collected from fixed-

Ontario over the

(cases

monitoring stations for the

Models adjusted for 3-day

period of 2004-

extracted

period of 2004-2011 in

mean temperature and relative

2011 were used

from

Ontario, which is part of

humidity using cubic splines.

to calculate the

NACRS

Canada's National Air

overall mean

database)

Pollution Data. PM data

PM2.5 exposure

obtained from 20

for the study

provincial monitoring sites

location and

located in 16 cities. Case

period.

and control days of study

participants were

assigned PM2.5

concentration based on

the monitoring station

closest to the population-

weighted centroid of each

subject's 3-digit postal

code.

Various lags assessed:

lagO, Iag1, Iag2 and mean

of lagO-2.

B-81

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Wu et al., 2020

All-cause
mortality
(65+)

US
Nationwide

Cohort
study
(MEDICA
RE)

Annual PM2.5 exposure.
Modeled PM2.5 exposure
at 1km2 grid cells across
the U.S. using well-
validated ensemble
models (Di et al. 2019a,
Di 2019b) for the period of

2000-2016. Daily
concentration in grid cells
were then averaged
to estimate
annual concentration at

ZIP code and then
assigned to individual
based on ZIP code of
residence.

Five statistical approaches: 2

regression approach (Cox
Hazard, Poisson reg); 3 causal
inference approach (GPSs)

Stratified by individual-level
characteristics. Further adjusted

for community-level factors
such as smoking and BMI, ZIP
code-level census variables and
meteorological variables,
geographic regions, and
calendar years (2000-2016).

Annual average

PM2.5
concentration for
participants were
used to calculate
overall mean
PM2.5
concentration for
the study period.

B-82

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Wyatt etal.,
2020

All-cause,
CVD, RD
30-day
hospital
readmissi
ons

530 U.S.
counties

Case-
crossover

and
Cohort stu
dy design
s

(USRDS h
emodialysi
s patients)

PM2.5 concentration
estimates from AOD
integrated with chemical

transport model
predictions, meteorology,
land use variables for 1
km grid cells (Di,2016).
Gridded PM2.5 estimates

were subsequently
converted to population-

weighted
county-level estimates
using 2010 Census tract
population values. Daily
PM2.5 was linked to patient
hospitalizations based on
the county of their last
dialysis visit.

Examined Lag 0 and
unconstrained distributive
lag model.

The relative risks of hospital
admissions associated with
daily PM2.5 were estimated with
conditional Poisson models for
each of the three health
outcomes separately. Cox
proportional hazards models

were used to assess the
relative risk of early (1-7 days
post discharge) and late (8-30

days post discharge)
readmission associated with
daily PM2.5 following all-cause
and cause-specific index
hospitalizations.
Cox model adjusted for time-
dependent (daily PM2.5, daily
temperature, daily RH, and day

of the week) and time-
independent (patient-specific
hospitalization event and county
SES) risk factors.

Daily
estimates at
county-level were
used to calculate

overall PM2.5
concentration for
the study location
and period.

B-83

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Yap et al., 2013

Asthma
HA

12 CA
counties

Time
Series
study
(Hospital
admission

PM2.5 data for the period
of 2000-2005 obtained

from California Air
Resources Board that
maintains information
from the National Air
Monitoring Stations.
PM2.5 reported was 24-hr

average mass
concentration based on
measurements taken
every 1, 3, or 6 days. For
counties with more than 1
monitoring site, daily
average PM2.5 was
calculated by taking the
average across monitors
within the county. Missing
values were computed
based on data from other
monitoring stations.

PM at various lags lagO-
Iag6 were assessed.

Generalized Additive Poisson
Regression analysis were run at
county-level

Models adjusted for: long-term
time trends and seasonality,
day of the week and smoothing
splines within different lags for
temperature. Effect modification
by single or composite area-
based SES assessed.

Daily PM2.5
concentrations in
12 CA counties
over the period of
2000-2005 were
used to calculate
the overall mean
PM2.5 exposure
for the study
location and
period.

B-84

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Yazdi et al.,
2019

First HA:
Stroke,
COPD,
pneumoni
a, Ml, lung
cancer,
and HF

Southeastern
states: FL,
AL, MS, GA,
NC, SC, and
TN

Cohort
study
(MEDICA
RE)

PM2.5 concentration
estimated from spatio-
temporal prediction model
at 1-km2 grid cell (Di et al.
2017) for the period of
2000-2012. Daily PM2.5
concentrations for grid
cells were averaged to
create annual PM2.5
concentration at ZIP code

level and assigned to
study participants based
on the ZIP code of
residence

Marginal structural Cox
proportional hazards models
which was weighted with
stabilized IPWs (to approximate
a causal model).
Adjusted for individual-level
variables (sex, race, year, state,
Medicaid eligibility), as well as
census SES.

B-85

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Zanobetti et al.,
2009

Heart
Failure
and Ml HA
65+

26 U.S.
communities

Time
Series
study
(MEDICA

RE
enrollees
data)

PM2.5 data obtained from
U.S. EPA AQS for the
period of 2000-2003. For
majority of cities,
metropolitan counties
encompassed the city and
its suburbs but some
cities like Boston,
Minneapolis-St Paul
included multiple counties.
Daily PM2.5 data available
for various monitors were
averaged over the county
and community (Monitors
ranged from 1-4). Before

averaging, however,
monitors were tested for
correlations and those
with correlation <0.8 with
2 or more monitor pairs
within a county were
excluded considering it
does not represent
exposure for general
population.

Generated 2-day moving
average (Iag01)
concentration

Poisson regression analysis

Models stratified by season.
Controlled for long-term trend
with natural cubic spline for
each season and year, day of
the week, three-day average
temperature and dew point
temperature.

B-86

-------
Zanobetti and
Schwartz, 2009

All-cause,
CVD and
respiratory
mortality

112 U.S.
cities

Time
Series
study
(NCHS
data)

PM25 data obtained from
U.S. EPAAQS (NAMS
and SLAMS providing
daily PM2.5 concentration)
for the period of 1999-
2005. For majority of
cities, counties
encompassed the city but
some cities like Boston,
Atlanta, Washington DC,
the city included multiple
counties. Daily PM2.5 (24-
hr) data available for
various monitors were
averaged over the county
and city. Before
averaging, however,
monitors were tested for
correlations and those
with correlation <0.8 with
2 or more monitor pairs
within a county were
excluded considering it
does not represent
exposure for general
population. Used
standardized method to fill
in the missing data in
some monitors with at
least 265 days of data in
at least one year.

Poisson regression analysis

First city- and season-specific
Poisson regression was run,
and then city-specific estimates
were combined using random
effects approach in total by
season and region.

Controlled for long-term trend
with natural cubic spline for
each season and year, day of
the week, same day, and
previous day temperature.

Daily PM2.5
concentrations in
112 U.S. cities
over the period of
1999-2005 were
used to calculate
the overall mean
PM2.5 exposure
for the study
location and
period???

Generated 2-day moving
average (lag 01)
concentration

B-87

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Zanobetti et al.,
2014

All-cause
mortality
65+

121 U.S.
communities/
cities

Case-
Crossover
Design
(MEDICA

RE
enrollees)

PM2.5 data obtained from
U.S. EPAAQS. Daily
PM2.5 data available for
various monitors were
averaged over the
communities. Participants
were assigned 2-day
moving average (lag 0

and 1) based on
community of residence.

Conditional logistic regression
models at community level. In a
second stage of analysis, the
community specific results were
combined using the multivariate
meta-analysis techniques

Conditional logistic regression
controlled for confounders such
as average temp for the same
and previous day. Temperature
was modelled using spline to

account for nonlinear
relationship. Effect modification

tested for cause of prior
admission due to neurological
disorders or diabetes, primary
or secondary hospitalization for

other disease conditions.
Stratified analysis by sex, age,
or race.

B-88

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Zeger etal.,
2008

All-cause
mortality
65+

668 U.S
Urban
counties

Retrospec
tive
Cohort
Study of
MEDICAR
E

enrollees
(MCAPS)

PM2.5 data (every 6th day

at many locations)
available from U.S. EPA's
AirData Database for the

period of 2000-2005.
Calculated mean annual
PM2.5 concentration for all
4,568 ZIP code centroids
within 6 miles of a monitor
with >10 months of data
per year. Given the focus
of study on long-term
exposure, ZIP code 6-
year average of PM2.5 was
calculated and assigned
to study participants living
within a ZIP code both
during the 6 years of
follow-up and some time
before cohort enrollment.

Log-linear Regression model
ran for specific U.S. regions
separately

Models adjusted for individual
socio-demographic variables

and ZIP code level SES
variables (education, income,
poverty etc.). Also included
standardized mortality ratio for
COPD as a surrogate indicator
of long-term smoking pattern of
its residents.

Average annual

PM2.5
concentrations of
ZIP codes were
used to calculate
overall mean
PM2.5 exposure
for the study
location (all and
by region) for the
study period
2000-2005.

B-89

-------
Citation

Long-term
(LT)/Short
-term (ST)

Health
Endpoint

Geographic
Area

Study
Design

Exposure Metric

Statistical Analysis Including
Confounding Variables
Addressed

Calculation of
study reported

mean PM2.5
concentrations

Zhang et al.,
2021

Non-
accidental
, CVD and
respiratory
mortality

Ontario,
Canada

Cohort
study
(Ontario
Health
Study)

PM2.5 exposures derived
from AOD retrievals using
GEOS-Chem calibrated to
surface measurements by

GWR (van Donkelaar,
2015). PM2.5 estimates at
1 km2 were used to
estimate annual PM2.5
average and then 3-year

and 5-year moving
averages. These annual

estimates were then
assigned to participants
based on postal code of

residence (updated
annually to account for
residential mobility).

Cox proportional hazard
models.

Basic model stratified by age,
sex, ethnicity, enrollment year
to control for baseline risks.
Models were adjusted for born
in Canada, education, marital
status, household income, BMI,

fruits, and vegetable intake,
smoking and drinking, physical
activity, urban/rural, and various
neighborhood level SES
indicators.

The 5-
year average

PM2.5
concentrations
were used to
calculate overall

mean PM2.5
concentration for
the baseline
year.

B-90

-------
Zigler et al.,
2018

All-cause

Eastern US

Causal

mortality

inference

and

methods

hospital

and a

admission

spatial

sfor

hierarchic

COPD;

cardiovas

regression

cular

model

stroke;

heart

failure;

heart

rhythm

disorders;

IHD;

peripheral

vascular

disease;

respiratory

tract

infection;

all

respiratory

hospitaliza

tions; and

all

cardiovas

cular

hospitaliza

tions

PM2.5 monitoring data was

used to estimate the
average annual ambient
concentrations of PM2.5
from
2010-2012

Study 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 outcomes.

The study design considers
years before 2005 (2000-2005)
as the

"baseline period," with a "follow-
up" period of 2010-2012. The
study 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 |jg/m3 or
the daily standard of 65 |jg/m3,
led to a corresponding
reduction in ambient PM2.5
concentrations and
hospitalization admission rates
for mortality, cardiovascular-
related and respiratory related
outcomes among Medicare
beneficiaries in the eastern U.S.
from 2009 to 2012 in areas
designated as nonattainment
areas versus areas designated
as attainment areas.

B-91

-------
REFERENCES

Bell, ML, Ebisu, K, Peng, RD, Walker, J, Samet, JM, Zeger, SL and Dominic, F (2008).

Seasonal and regional short-term effects of fine particles on hospital admissions in 202
U.S. counties, 1999-2005. American Journal of Epidemiology 168(11): 1301-1310.

Bell, ML, Son, JY, Peng, RD, Wang, Y and Dominici, F (2015). Brief report: Ambient PM2.5 and
risk of hospital admissions: do risks differ for men and women? Epidemiology 26(4):
575-579.

B-92

-------
Chan, E. (2019). Email and attachment from Elizabeth Chan, EPA, Question re: Request for
additional PM2.5 information from Di et al 2017. Available in PM NAAQS Review
Docket (EPA-HQ-OAR-2015-0072). December 3, 2019. Office of Air Quality Planning
and Standards Research Triangle Park, NC.

Chi, GC, Hajat, A, Bird, CE, Cullen, MR, Griffin, BA, Miller, KA, Shih, RA, Stefanick, ML,
Vedal, S, Whitsel, EA and Kaufman, JD (2016). Individual and neighborhood
socioeconomic status and the association between air pollution and cardiovascular
disease. Environmental Health Perspectives 124(12): 1840-1847.

Christidis, T, Erickson, AC, Pappin, AJ, Crouse, DL, Pinault, LL, Weichenthal, SA, Brook, JR,
van Donkelaar, A, Hystad, P, Martin, RV, Tjepkema, M, Burnett, RT and Brauer, M
(2019). Low concentrations of fine particle air pollution and mortality in the Canadian
Community Health Survey cohort. Environmental Health: A Global Access Science
Source 18(1): 84.

Crouse, DL, Peters, PA, Hystad, P, Brook, JR, van Donkelaar, A, Martin, RV, Villeneuve, PJ,
Jerrett, M, Goldberg, MS, Pope, CA, Brauer, M, Brook, RD, Robichaud, A, Menard, R
and Burnett, RT (2015). Ambient PM 2.5, 03, and N02 exposures and associations with
mortality over 16 years of follow-up in the Canadian Census Health and Environment
Cohort (CanCHEC). Environmental Health Perspectives 123(11): 1180-1186.

Crouse, DL, Pinault, L, Balram, A, Brauer, M, Burnett, RT, Martin, RV, van Donkelaar, A,

Villeneuve, PJ and Weichenthal, S (2019). Complex relationships between greenness, air
pollution, and mortality in a population-based Canadian cohort. Environment
International 128: 292-300.

B-93

-------
deSouza, P, Braun, D, Parks, RM, Schwartz, J, Dominici, F and Kioumourtzoglou, MA (2021).
Nationwide Study of Short-term Exposure to Fine Particulate Matter and Cardiovascular
Hospitalizations Among Medicaid Enrollees. Epidemiology 32(1): 6-13.

Di, Q, Dai, L, Wang, Y, Zanobetti, A, Choirat, C, Schwartz, JD and Dominici, F (2017a).

Association of short-term exposure to air pollution with mortality in older adults. JAMA:
Journal of the American Medical Association 318(24): 2446-2456.

Dominici, F, Schwartz, J, Di, Q, Braun, D, Choirat, C and Zanobetti, A (2019). Assessing

adverse health effects of long-term exposure to low levels of ambient air pollution: Phase
1. Health Effects Institute. Boston, MA. Available at:
https://www.healtheffects.org/svstem/files/dominici-rr-200-report.pdf.

Elliott, EG, Laden, F, James, P, Rimm, EB, Rexrode, KM and 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. Environmental Health
Perspectives 128(12): 127012.

Eum, KD, Suh, HH, Pun, VC and Manjourides, J (2018). Impact of long-term temporal trends in
fine particulate matter (PM2.5) on associations of annual PM2.5 exposure and mortality:
An analysis of over 20 million Medicare beneficiaries. Environmental Epidemiology
2(2): e009.

Fisher, JA, Puett, RC, Laden, F, Wellenius, GA, Sapkota, A, Liao, D, Yanosky, JD, Carter-

Franklin, M, Koutrakis, P and Schwartz, J (2008). The role of particle composition on the
association between PM2.5 and mortality. Epidemiology 19(5): 680-689.

B-94

-------
Franklin, M, Zeka, A and Schwartz, J (2007). Association between PM2.5 and all-cause and
specific-cause mortality in 27 US communities. Journal of Exposure Science and
Environmental Epidemiology 17(3): 279-287.

Gharibvand, L, Shavlik, D, Ghamsary, M, Beeson, WL, Soret, S, Knutsen, R and Knutsen, SF
(2016). The association between ambient fine particulate air pollution and lung cancer
incidence: Results from the AHSMOG-2 study. Environmental Health Perspectives
125(3): 378-384.

Hart, JE, Garshick, E, Dockery, DW, Smith, TJ, Ryan, L and Laden, F (2011). Long-term

ambient multi-pollutant exposures and mortality. American Journal of Respiratory and
Critical Care Medicine 183(1): 73-78.

Hart, JE, Liao, X, Hong, B, Puett, RC, Yanosky, JD, Suh, H, Kioumourtzoglou, MA,

Hayes, RB, Lim, C, Zhang, YL, Cromar, K, Shao, YZ, Reynolds, HR, Silverman, DT, Jones,

RR, Park, Y, Jerrett, M, Ahn, J and Thurston, GD (2020). PM2.5 air pollution and cause-
specific cardiovascular disease mortality. International Journal of Epidemiology 49(1):
25-35.

Hystad, P, Demers, PA, Johnson, KC, Carpiano, RM and Brauer, M (2013). Long-term

residential exposure to air pollution and lung cancer risk. Epidemiology 24(5): 762-772.

Jenkins, S. (2019a). Email and attachment from Scott Jenkins, EPA, Question re: PM2.5 LT

estimates Di et al. (2017) Studies. Available in PM NAAQS Review Docket (EPA-HQ-
OAR-2015-0072-0022). May 8, 2019. Office of Air Quality Planning and Standards
Research Triangle Park, NC.

B-95

-------
Jenkins, S. (2019b). Email and attachment from Scott Jenkins, EPA, Question re: PM2.5 ST

estimates Di et al. (2017) Studies. Available in PM NAAQS Review Docket (EPA-HQ-
OAR-2015-0072-0025). May 9, 2019. Office of Air Quality Planning and Standards
Research Triangle Park, NC.

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 and
Burnett, RT (2016). Comparing the health effects of ambient particulate matter estimated
using ground-based versus remote sensing exposure estimates. Environmental Health
Perspectives 125(4): 552-559.

Klemm, RJ and Mason, R (2003). HEI Special Report: Revised Analyses of Time-Series Studies
of Air Pollution and Health, Part II Replication of reanalysis of Harvard Six-City
mortality study. Health Effects Institute. Boston, MA. Available at:
https://www.healtheffects.org/svstem/files/TimeSeries.pdf.

Kloog, I, Coull, BA, Zanobetti, A, Koutrakis, P and Schwartz, JD (2012). Acute and chronic
effects of particles on hospital admissions in New-England. PLoS ONE 7(4): e34664.

Krall, JR, Chang, HH, Waller, LA, Mulholland, JA, Winquist, A, Talbott, EO, Rager, JR,
Tolbert, PE and Sarnat, SE (2018). A multicity study of air pollution and
cardiorespiratory emergency department visits: Comparing approaches for combining
estimates across cities. Environment International 120: 312-320.

Krewski, D, Jerrett, M, Burnett, RT, Ma, R, Hughes, E, Shi, Y, Turner, MC, Pope, CA, III,
Thurston, G, Calle, EE, Thun, MJ, Beckerman, B, Deluca, P, Finkelstein, N, Ito, K,
Moore, DK, Newbold, KB, Ramsay, T, Ross, Z, Shin, H and Tempalski, B (2009).
Extended follow-up and spatial analysis of the American Cancer Society study linking
particulate air pollution and mortality. HEI Research Report 140. Health Effects Institute.
Boston, MA. https://www.healtheffects.org/svstem/files/Krewskil40Statement.pdf.

B-96

-------
Lee, H, Honda, Y, Hashizume, M, Guo, YL, Wu, CF, Kan, H, Jung, K, Lim, YH, Yi, S and Kim,
H (2015). Short-term exposure to fine and coarse particles and mortality: A multicity
time-series study in East Asia. Environ Pollut 207: 43-51.

Lippmann, M, Chen, LC, Gordon, T, Ito, K and Thurston, GD (2013). National Particle

Component Toxicity (NPACT) Initiative: Integrated epidemiologic and toxicologic
studies of the health effects of particulate matter components. 177. Health Effects
Institute. Boston, MA. https://www.healtheffects.org/svstem/files/RR177-Lippmann.pdf.

Malig, BJ, Green, S, Basu, R and Broadwin, R (2013). Coarse particles and respiratory

emergency department visits in California. American Journal of Epidemiology 178(1):
58-69.

Malik, AO, Jones, PG, Chan, PS, Peri-Okonny, PA, Hejjaji, V and 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(4): e005598.

McClure, LA, Loop, MS, Crosson, W, Kleindorfer, D, Kissela, B and Al-Hamdan, M (2017).

Fine Particulate Matter (PM2.5) and the Risk of Stroke in the REGARDS Cohort. Journal
of Stroke & Cerebrovascular Diseases 26(8): 1739-1744.

McConnell, R, Islam, T, Shankardass, K, Jerrett, M, Lurmann, F, Gilliland, F, Gauderman, J,

Avol, E, Kiinzli, N, Yao, L, Peters, J and Berhane, K (2010). Childhood incident asthma

B-97

-------
and traffic-related air pollution at home and school. Environmental Health Perspectives
118(7): 1021-1026.

Peng, RD, Dominici, F and Welty, LJ (2009). A Bayesian hierarchical distributed lag model for
estimating the time course of risk of hospitalization associated with particulate matter air
pollution. J Roy Stat Soc Ser C (Appl Stat) 58: 3-24.

Pinault, LL, Weichenthal, S, Crouse, DL, Brauer, M, Erickson, A, Donkelaar, AV, Martin, RV,
Hystad, P, Chen, H, Fines, P, Brook, JR, Tjepkema, M and Burnett, RT (2017).
Associations between fine particulate matter and mortality in the 2001 Canadian Census
Health and Environment Cohort. Environmental Research 159: 406-415.

Pope, CA, Lefler, JS, Ezzati, M, Higbee, JD, Marshall, JD, Kim, SY, Bechle, M, Gilliat, KS,
Vernon, SE, Robinson, AL and Burnett, RT (2019). Mortality Risk and Fine Particulate
Air Pollution in a Large, Representative Cohort of U.S. Adults. Environmental Health
Perspectives 127(7): 77007.

B-98

-------
Pope, CA, Turner, MC, Burnett, R, Jerrett, M, Gapstur, SM, Diver, WR, Krewski, D and Brook,
RD (2015). Relationships between fine particulate air pollution, cardiometabolic
disorders and cardiovascular mortality. Circulation Research 116(1): 108-U258.

Rajan, P, Schmidt, M., Hassett-Sipple, B. (2011). Memorandum to PM NAAQS Review Docket
(EPA-HQ-OAR-2007-0492). PM2.5 Distributional Statistical Analyses. April 7, 2011.
Docket ID No. EPA-HQ-OAR-2007-0492. Office of Air Quality Planning and Standards
Research Triangle Park, NC. Available at:

https://www3.epa.gov/ttn/naaqs/standards/pm/data/Raian040711.pdf.

Stieb, DM, Szyszkowicz, M, Rowe, BH and Leech, JA (2009). Air pollution and emergency

department visits for cardiac and respiratory conditions: A multi-city time-series analysis.
Environmental Health: A Global Access Science Source 8(25): 25.

Szyszkowicz, M (2009). Air pollution and ED visits for chest pain. American Journal of
Emergency Medicine 27(2): 165-168.

Thurston, GD, Ahn, J, Cromar, KR, Shao, Y, Reynolds, HR, Jerrett, M, Lim, CC, Shanley, R,
Park, Y and Hayes, RB (2016a). Ambient particulate matter air pollution exposure and
mortality in the NIH-AARP Diet and Health Cohort. Environmental Health Perspectives
124(4): 484-490.

B-99

-------
Thurston, GD, Burnett, RT, Turner, MC, Shi, Y, Krewski, D, Lall, R, Ito, K, Jerrett, M, Gapstur,
SM, Diver, WR and Pope, CA, III (2016b). Ischemic heart disease mortality and long-
term exposure to source-related components of U.S. fine particle air pollution. Environ
Health Perspect 124(6): 785-794.

Turner, MC, Jerrett, M, Pope, A, III, Krewski, D, Gapstur, SM, Diver, WR, Beckerman, BS,

U.S. EPA (2021). Supplement to the 2019 Integrated Science Assessment for Particulate Matter
(External Review Draft). U.S. Environmental Protection Agency, Office of Research and
Development, Center for Public Health and Environmental Assessment. Research
Triangle Park, NC. U.S. EPA. EPA/600/R-21/198. December 2019. Available at:
https://www.epa.gov/naaqs/particulate-matter-pm-standards-integrated-science-
assessments-current-review.

Urman, R, McConnell, R, Islam, T, Avol, EL, Lurmann, FW, Vora, H, Linn, WS, Rappaport,
EB, Gilliland, FD and Gauderman, WJ (2014). Associations of children's lung function
with ambient air pollution: Joint effects of regional and near-roadway pollutants. Thorax
69(6): 540-547.

B-100

-------
Wang, Y, Shi, L, Lee, M, Liu, P, Di, Q, Zanobetti, A and Schwartz, JD (2017). Long-term
exposure to PM2.5 and mortality among older adults in the Southeastern US.

Epidemiology 28(2): 207-214.

Weichenthal, S, Hatzopoulou, M and Goldberg, MS (2014). Exposure to traffic-related air

pollution during physical activity and acute changes in blood pressure, autonomic and
micro-vascular function in women: A cross-over study. Particle and Fibre Toxicology
11(1): 70.

Wyatt, LH, Xi, YZ, Kshirsagar, A, Di, Q, Ward-Caviness, C, Wade, TJ, Cascio, WE and

Rappold, AG (2020). 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. British Medical Journal Open 10(12): e041177.

Yap, P, Gilbreath, S, Garcia, C, Jareen, N and Goodrich, B (2013). The influence of

socioeconomic markers on the association between fine particulate matter and hospital
admissions for respiratory conditions among children. American Journal of Public Health
103(4): 695-702.

B-101

-------
Zanobetti, A and Schwartz, J (2009). The effect of fine and coarse particulate air pollution on
mortality: A national analysis. Environmental Health Perspectives 117(6): 1-40.

Zigler, CM, Choirat, C and Dominici, F (2018). Impact of National Ambient Air Quality

Standards nonattainment designations on particulate pollution and health. Epidemiology
(Cambridge, Mass.) 29(2): 165.

B-102

-------
APPENDIX C. SUPPLEMENTAL INFORMATION
RELATED TO THE HUMAN HEALTH RISK

ASSESSMENT

-------
TABLE OF CONTENTS

C. 1 Additional Technical Detail on the Risk Assessment Approach C-l

C. 1.1 Selection of Key Health Endpoints and Specification of Concentration-Response
Functions from Epidemiologic Studies C-2

C. 1.2 Specification of Demographic and Baseline Incidence Data Inputs C-l 1

C. 1.3 Study Area Selection C-l 1

C. 1.4 Generation of Air Quality Inputs to the Risk Assessment C-17

C. 1.5 Ri sk Modeling Approach C-46

C.2 Supplemental Risk Results C-48

C.2.1 Results from Full Set of 47 Study Areas C-50

C.2.2 Results from Set of 30 Study Areas controlled by the Annual Standard C-55

C.2.3 Key Observations from the Suppmental Risk Results C-56

C.3 Additional Technical Detail on the At-Risk Analysis C-57

C.3.1 Race/Ethnicity C-58

C.3.2 Concentration-Response Functions C-59

C.3.3 Age C-61

C.3.4 Baseline Incidence Rates C-61

C.3.5 Selection of Air Quality Simulation Approach C-63

C.4 Supplemental At-Risk Results C-63

C.5 Characterizing Variability and Uncertainty in Risk Estimates C-70

C. 5.1 Quantitative Assessment of Uncertainty C-72

C.5.2 Qualitative Uncertainty Analysis C-73

C.5.3 Conclusion C-81

C.6 PM2.5 Design Values for the Air Quality Projections C-82

References C-l 06

-------
This appendix provides supplemental information related to the risk assessment described
in section 3.4 of this PA for the reconsideration of the 2020 final decision on the particulate
matter (PM) National Ambient Air Quality Standards (NAAQS), including:

• Additional technical detail on the risk assessment approach, including sources and
derivation of key inputs to the risk modeling process (section C. 1).

• Supplemental risk results (section C.2) intended to provide additional context for the
summary risk estimates presented in sections 3.4.2.1-3.4.2.3.

• Additional technical detail on the at-risk analytic approach, including sources and
derivation of key inputs to the risk modeling process (section C.3).

• Supplemental at-risk analytics (section C.3.4.2) intended to provide additional context for
the summary risk estimates presented in section 3.4.2.4.

• Characterization of variability and uncertainty related to the risk assessment (section C.5)
intended to provide additional context for section 3.4.2.5.

C.l ADDITIONAL TECHNICAL DETAIL ON THE RISK ASSESSMENT
APPROACH

As discussed in section 3.4, our general approach to estimating PIVh.s-associated human
health risks in this reconsideration utilizes concentration-response (CR) functions obtained from
epidemiologic studies to link ambient PM2.5 exposure to risk in the form of mortality incidence
(counts). The derivation and use of this type of CR function in modeling PM2.5-attributable risk
is well documented both in previous PM NAAQS-related risk assessments (section 3.1.2 of U.S.
EPA, 2010) and section C.l.l of this appendix. Inputs required to model risk using CR functions
are identified below (Figure C-l) and include

(1) the CR functions themselves, obtained from epidemiologic studies (section C.l.l and
C.3.2),

(2) baseline health incidence data and information on population demographics (section 0
and C.3.4),

(3) study areas (section C.1.3), and

(4) modeled ambient PM2.5 concentrations corresponding to air quality scenarios of interest
(section C.1.4).

C-l

-------
Specification of
demographic
and baseline

incidence inputs

Selection of Health

Endpoints and
Specification of CR
functions (including
selection of epi studies)

Risk modeling
(including generation
of risk metrics)
Using BenMAP-CE
Model

Risk estimates
(metrics)

Study area
selection

Air quality
characterization

Figure C-l. Key inputs to the risk assessment.

C.l.l Selection of Key Health Endpoints and Specification of Concentration-Response
Functions from Epidemiologic Studies

In selecting specific CR functions for the risk assessment, we began by considering
health outcomes for which the 2019 Integrated Science Assessment (ISA) determined the
evidence supports either a "causal" or a "likely to be causal" relationship with short- or long-
term PM2.5 exposures (U.S. EPA, 2019). As discussed in Chapter 3 (Table 3-1), these outcomes
include the following:

• mortality (resulting from long- and short-term exposure),

• cardiovascular effects (resulting from long- and short-term exposure),

• respiratory effects (resulting from long- and short-term exposure),

• cancer (resulting from long-term exposure), and

• nervous system effects (resulting from long-term exposure).

We focused the risk assessment on short- and long-term PM exposure-related mortality,
reflecting its clear public health importance, the large number of epidemiologic studies available
for consideration, and the broad availability of baseline incidence data. The specific set of health
effect endpoints included in the risk assessment are:

• Long-term PM exposure-related mortality, all-cause

C-2

-------
• Short-term PM exposure-related mortality, all-cause and non-accidental

To identify specific epidemiologic studies for potential inclusion in the risk assessment,
we focus on U.S. multicity studies assessed in the 2019 ISA. These studies are identified in
section 3.4.1.5 of this PA. Of these, we used the following criteria to identify the specific set of
studies for inclusion in the risk assessment:

• National-scale geographic coverage: We focus on epidemiologic studies reporting
national-level CR functions. Epidemiologic studies that focus on individual cities or
regions were excluded. Focusing on national-level epidemiologic studies has the benefit
of characterizing PIVh.s-associted risks broadly across the U.S. and in relatively large
populations (compared with single-city or regional studies), which tends to improve
precision in the CR functions generated.

• Evaluation of relatively lower ambient PM concentrations: In selecting epidemiology
studies, to the extent possible, we focus on those studies which characterized the ambient
PM2.5-mortality relationship at levels at or near the current NAAQS, given that the risk
assessment would be focusing on evaluating risk associated with the current NAAQS.

• Populations with available baseline incidence data: For some populations (e.g., diesel
truck drivers), it can be challenging to model risk at the national-level given uncertainties
associated with specifying key inputs for risk modeling (i.e., baseline incidence rates for
mortality endpoints and detailed national-level demographics). For that reason, we focus
on those epidemiology studies providing CR functions for populations readily
generalizable to the broader U.S. population (e.g., specific age groups not differentiated
by additional socio-economic, or employment attributes).

• Estimates of long-term PM2.5 exposures based on hybrid modeling approaches: For long-
term PM2.5 exposures, we focus on epidemiologic studies that estimate exposures with
hybrid modeling approaches. The rationale for this decision is the agreement between the
design of these epidemiology studies (i.e., their use of hybrid-based modeling approaches
in characterizing ambient PM) and the hybrid air quality surfaces we are using in this risk
assessment. This general agreement between the air modeling surfaces used in long-term
mortality epidemiology studies and our air quality modeling reduces uncertainty in the
risk assessment.

• Estimates of short-term PM2.5 exposures based on composite monitor data: Short-term
mortality epidemiology studies utilizing hybrid modeling approaches, which are fewer in
number compared with long-term mortality studies, tend to be regional in scope and did
not meet the criterion of providing national-scale effect estimates. For that reason, in
modeling short-term mortality, epidemiology studies utilizing composite-monitor based
exposure surrogates were used as the basis for deriving CR functions. We recognize the
uncertainty introduced into the modeling of short-term mortality due to the use of CR
functions obtained from studies utilizing composite monitors. However, we felt these use
of national-scale epidemiology studies was a more important criterion for selection.

• Evaluation of potential confounders and effect modifiers: To the extent possible,
preference was given to studies which more fully address potential confounders and
effect modifiers and to those studies which utilize individual (rather than ecological)

C-3

-------
measures in representing those confounders/effect modifiers. Recognizing that both
single- and co-pollutant models have advantages and disadvantages in characterizing the
ambient PM-mortality relationship, to the extent possible, we include epidemiology
studies (and associated CR functions) based on either single- or co-pollutant models that
include ozone. Additional information available in the Estimating PM2.5 and Ozone-
Attributable Health Benefits TSD associated with the 2021 Revised Cross-State Air
Pollution Rule Update (RCU) (U.S. EPA, 2021).

• Exploration of multiple approaches for estimating exposures: For studies that estimate
PM2.5 exposures using hybrid modeling approaches, preference was given to studies that
also explore additional methods for estimating exposures (i.e., multiple hybrid methods
or hybrid methods plus monitor-based methods) and compare health effect associations
across approaches.

Application of the criteria listed above resulted in the selection of the epidemiology
studies presented in Table C-l for inclusion in the risk assessment as sources of effect estimates.
Table C-l includes summary information on study design, details on the selection of effect
estimates, the derivation of beta values, and specification of CR functional form based on those
effect estimates for use in the risk assessment. The procedure used to derive CR functions
(including specification of beta values and mathematical forms for those functions) is described
below.

The remainder of this section describes the method used in specifying the CR functions
used in the PM NAAQS HHRA. Information presented in this section is drawn from the EPA's
Environmental Benefits Mapping and Analysis Program - Community Edition (BenMAP-CE)
Manual, Appendix C.1 CR functions translate changes in ambient PM2.5 into changes in baseline
incidence rates for specific disease endpoints utilizing beta (P) values obtained from
epidemiology studies studying the association between ambient PM2.5 exposure and specific
health endpoints. P values (and associated standard errors) are based on effect estimates obtained
from the underlying epidemiology studies. In addition, the mathematical forms for the health
impact functions specified for use in this risk assessment reflect the models used in the
epidemiology studies providing those effect estimates. Consequently, derivation of the P values
based on effect estimates from underlying epidemiology studies (and specification of the form of
the health impact functions) represents a key step in the design of the HHRA.

The majority of the epidemiology studies providing effect estimates for this PM HHRA
utilized either Poisson or Cox proportional hazard models which result in exponential (or log-
linear) forms for the CR functions, where the natural logarithm of mortality incidence is a linear

1 https://www.epa.gov/benmap/benmap-ce-manual-and-appendices

C-4

-------
function of PM2.5.2 If we let xo denote the baseline (starting) PM2.5 level, and xi denote the
control (ending) PM2.5 level, yo denote the baseline incidences rate of the health effect, and Pop
the underlying population count for the applicable demographic group in the spatial unit of
analysis3 we can derive the following CR function specifying the relationship between the
change in x, Ax= (xo- xi) and the corresponding change in y, Ay (mortality incidence):

Ay = y0[ 1 — e~PAx] * Pop

Given that the epidemiology studies providing effect estimates for long-term exposure-
related mortality and short-term exposure-related mortality in the context of the current PM
HHRA (Table C-l) use different categories of models (Cox proportional hazard and
Poisson/Logistic, respectively) we describe the process of deriving the betas and specifying CR
functional forms separately for each of these endpoint categories. As noted earlier, the logit
model utilized in Zanobetti et al., 2014, is discussed at the end of the section covering short-term
PM2.5-related mortality.

Derivation of betas for long-term PM2 5 exposure-related mortality

Cox proportional hazard models used to evaluate mortality associated with long-term
PM2.5 exposure are designed to model effects on population survival. This class of epidemiology
model is based on a hazard function, defined as the probability that an individual dies at time t,
conditional on that individual having survived up to time t. As such, the hazard function
represents a time-specific snapshot of the rate of mortality (events per unit time) within a study
population. While the risk can vary over time, in the case of the Cox proportional hazard model,
it is assumed that the hazard ratio is constant. The proportional hazard model takes the form:

h(X, t) = h0(t)ex'P

Where X is a vector of explanatory variables, P is a vector of associated coefficients and
ho(t) is the baseline hazard (the risk when all covariates (X) are set to zero).

Epidemiology studies utilizing the Cox proportional hazard model in characterizing
ambient PM2.5-health effects typically report hazard ratios (HRs) as the effect estimate. HRs
represent the ratio of hazard functions for the baseline and control scenarios reflecting a specific

2 One study. Zanobetti et al., 2014, supporting the modeling of short-term PM2 5 exposure-related mortality provided

a logistic-based model form, which is discussed at the end of this section.

3 Spatial unit of analysis refers to the geographic scale at which the CR function is applied in generating a risk

(incidence) estimate (e.g., zip code, county, 12km grid cell). Typically, the spatial unit of analysis used in a REA
is based on the spatial scale reflected in the epidemiology study(s) supplying the effect estimates. For this REA,
the spatial unit of analysis is the 12km grid cell.

C-5

-------
difference in ambient PM2.5 exposure (often a 10 |ig/m3 increment). The HR simplifies as shown
(with the baseline hazard ratio dropping out), allowing us to readily derive the P value from this
effect estimate:

_ h(X0,t) _ h0{t)exo'i3 _ gAPM'P
h{Xc,t) h0{t)exc'P

It is then possible to calculate the beta as follows:

„ _ ln{HR)

" APM

As noted in Sutradhar and Austin, 2018, the HR associated with a Cox-proportional
hazard model may approximate the RR when the effect estimate (and consequently the P) is
relatively small. This is the case with the effect on mortality modeled for long-term exposure to
ambient PM2.5 (i.e., the size of the effect estimate supports an assumed equivalency between HR
and RR). The near equivalency between the HR and RR, allows us to utilize the P derived above
in a CR function based on a log-linear functional form of the type presented earlier, to model
changes in mortality related to changes in ambient PM.

Derivation of betas for short-term PM2 5 exposure-related mortality

The epidemiology studies selected for use in modeling short-term PM2.5 exposure-related
mortality utilize both the Poisson (log-linear) model form (Baxter et al., 2017) and the logit
model form (Zanobetti et al., 2014).4 In both cases, the epidemiology studies provide effects in
terms of percent increase in mortality.

The log-linear (Poisson) model is used to evaluate effects associated with continuous
(count) events. With the log-linear (Poisson) model, the relative risk is simply the ratio of the
two risks:

RR = ^ = eP'APM

The derivation of the beta with a Poisson model specified RR is as follows. Taking the
natural log of both sides, the beta coefficient in the CR function underlying the relative risk can
be derived as:

„ _ m{RR)

" APM

4 Note that the Ito et al., 2013 study also utilizes a Poisson model. However, that study provides beta values

(including standard errors) and for that reason the results of this study are directly applicable in modeling changes
in mortality without any of the derivations presented here for the other studies.

C-6

-------
The beta derived in this fashion can then be used with a log-linear functional form (as
presented earlier) to model changes in mortality related to changes in ambient PM.

The logistic model form is used to model dichotomous events. With the logistic model
form, when we are provided with a RR value, as is the case here, we can make a similar
assumption to that used above with the Cox proportional hazard function (i.e., that the OR and
RR approach equivalency under conditions of relatively small effect levels). That observation in
turn allows us to assume that

RR =?!= (l-y0)x e~APM'P + y0

Then, assuming (based on the relatively small size of the baseline incidence) that:

e-APM.p ^ _ y(j) X e~APM.p + yo

^ RR = e~APM'P

It is then possible to calculate the underlying beta coefficient as follows:

/n(RR) _ „

-APM ~ "

Since the derivation of the beta is based on the assumption of a log linear functional
form, we can apply the beta in a log-liner CR function of the form described earlier:

Ay = y0[l — e~PAx] * Pop

C-7

-------
Table C-l. Details regarding selection of epidemiology studies and specification of concentration-response functions for the
risk assessment.

Reference
and Title

Study Description

Exposure
Estimation
Approach

CR Function

Location of CR
Functions) in
Article

Additional
Notes on CR
Function(s)
Selection

Epidemio-
logic
Statistic

Mortality
Endpoint

Selected
Effect
Estimate

Selected
Beta

Selected

Beta
Standard
Error (SE)

Long-term exposure-related mortality studies

Di et al., 2017

Air Pollution
and Mortality
in the
Medicare
Population

Exploring relationship between
air pollution (ozone, PM2.5) and
mortality
Key details:

- Medicare population (65+)

- ecological control for
confounders

- all-cause mortality only

- provides CR function slopes
for areas above and below the
current PM NAAQS level (but
model for areas below current
standard only done for low
ozone cells)

Exposures estimated
at zip code of
residence based on
a neural network
model that
incorporates satellite
data, chemical
transport modeling,
land-use terms,
meteorology data,
monitoring data, and
other data

Cox proportional-
hazards model
with a
generalized
estimating
equation to
account for the
correlation
between ZIP
codes

Table 2
Risk of death
associated with an
increase of 10 |jg/m3
PM2.5 or an increase
of 10 ppb in ozone
concentration. Uses
single pollutant
model for full
analysis.

Using single
pollutant, full PM
range model
(model for <12
|jg/m3 applicable
to only low-
ozone days)5

Hazard ratio
(95 percent
CI)

All-cause

1.073
(1.071,
1.075)

7.0E-03

1E-04

Turner et al.,
2016

Long-Term
Ozone
Exposure and
Mortality in a

Large
Prospective
Study

Evaluates the relationship
between long-term exposure
to ambient PM2.5 and all-cause
and cause-specific mortality.
Also, estimated the
association between PM2.5,
regional PM2.5, and near-
source PM2.5 and mortality in
single-pollutant, copollutant
and multipollutant models.

- ACS (30+)

- Includes lung cancer
(otherwise similar results to
Pope et al., 2015)

- county-level assessment

Exposures estimated
at residential
locations based on
land use data and
ground-based
monitors

Cox proportional
hazard model

Table E4. Adjusted
HRs (95th percentile
CI) for all-cause and
cause-specific
mortality in relation to
each 10 unit increase
in PM2.5 LUR-BME
concentrations,
follow-up 1982-2004,
CPS-II cohort, United
States (n = 669,046).

Note that the
non-cancer
mortality
endpoints
provided in table
E4 appear to
mirror those
provided in Table
1 of Popeetal.,
2015 -so will use
long-cancer
effect estimate
from this study
only.

Hazard ratio
(95 percent
CI)

All-cause

1.06
(1.04-
1.08)

5.8E-03

9.6E-04

Short-term exposure-related mortality studies

5 We note that Di et al., 2017 does include a copollutant model-based effect estimate (HR 1.073, 95th%CI 1.071-1.075). Had this effect estimate been used in risk
modeling (which would translate into a beta value of 7.05E-3), we would anticipate the risk estimates for all-cause mortality to be slightly less f 13% lower based
on comparison of calculated betas) than those estimated based on the single-pollutant model used in this risk assessment.

C-8

-------
Reference
and Title

Study Description

Exposure
Estimation
Approach

CR Function

Location of CR
Functions) in
Article

Additional
Notes on CR
Function(s)
Selection

Epidemio-
logic
Statistic

Mortality
Endpoint

Selected
Effect
Estimate

Selected
Beta

Selected

Beta
Standard
Error (SE)

Baxter et al.,
2017

Influence of
exposure
differences in
city-to-city
heterogeneity
in PM2.5-
mortality
associations
in U.S. cities

Uses cluster-based approach
to evaluate the impact of
residential infiltration factors
on inter-city heterogeneity in
short-term PM-mortality
associations.

- Mortality data from NCHS -
77 U.S. CBSAs (all ages)

- non-accidental mortality

- CBSA-level assessment

Exposure estimates
based on data from
ground-based
monitors

Poisson (log-
linear) at city-
level then
aggregated

Obtained from
results section in the
text. After pooling the
city-specific effect
estimates into an
overall effect
estimate, short-term
PM2.5 exposure was
found to increase 24-
hr non-accidental
mortality by 0.33%
(95% CI:

0.13, 0.53). Based
on lag 2 (day 0-1)

Percent
increase in
24-hr

mortality (95
percent CI)

24-hr non-
accidental
mortality

0.33
(0.13-
0.53)

3.29E-04

1.02E-04

Ito et al., 2013

NPACT study
3. Time-series
analysis of
mortality,
hospitalization
s, and

ambient PM2.5
and its
components

Use factor analysis to
characterize pollution sources,
assess the association
between PM2.5 and PM2.5
components with morbidity
and mortality outcomes. Also
evaluates pollution levels,
land-use, and other variables
as modifiers that may explain
inter-city variation in PM-
mortality effect estimates.

- Mortality data from NCHS -
150 and 64 U.S. cities (two
analyses) (all ages)

- MSA-level assessment

Exposure estimates
based on data from
ground-based
monitors

Poisson GLM

Appendix G, Table
G.6 for Figure 4 - use
all-year lag 1 Beta:
Regression
coefficients (beta)
and their SE for air
pollutants at lag 0
through 3 days used
to compute percent
excess risks in
figures shown in the
main text and in
Appendices B and G
(corresponding
figures are noted).

Utilized lag-1 (all
year) beta
because that had
the strongest
effect for CVD
mortality and
wanted our all-
cause to reflect
that stronger lag-
association for
the CVD effect
(even though
focusing on all-
cause)

Betas with
SE (no
conversion
required)

24-hr all-

cause

mortality

Study
provided
beta and
SE

1.45E-04

7.47E-05

C-9

-------
Reference
and Title

Study Description

Exposure
Estimation
Approach

CR Function

Location of CR
Functions) in
Article

Additional
Notes on CR
Function(s)
Selection

Epidemio-
logic
Statistic

Mortality
Endpoint

Selected
Effect
Estimate

Selected
Beta

Selected

Beta
Standard
Error (SE)

Zanobetti et
al., 2014

A national
case-
crossover
analysis of
the short-term
effect of PM2.5
on

hospitalization
s and
mortality in
subjects with
diabetes and
neurological
disorders

Estimates the effect of short-
term exposure to PM2.5 on all-
cause mortality. Additionally,
assesses the potential for pre-
existing diseases to modify the
association between PM2.5 and
mortality (neurological
disorders and diabetes)

- Medicare cohort -121 U.S.
communities (65+)

- Community-level assessment
(community defined as the
county or contiguous counties
encompassing a city's
population)

Exposure estimates
based on data from
ground-based
monitors

Logistic
regression

Table 2. Percent
increase for 10 |jg/m3
increase in the two
days average PM2.5:
Combined across the
121 communities

Percent
increase (95
percent CI)

All deaths

0.64
(0.42-
0.85)

6.38E-04

1.09E-04

C-10

-------
C.1.2 Specification of Demographic and Baseline Incidence Data Inputs

This risk analysis requires both demographic and baseline-incidence data for the mortality
endpoint categories evaluated. For our analyses, these data are for the year 2015 since the hybrid
surfaces included in the analyses are based on a 2015 model year.6 The BenMAP-CE model7 is
used in this risk assessment and the relevant demographic and baseline incidence data for the
contiguous U.S., from the sources described below, is readily available within the current version
of BenMAP-CE:

• Demographic data: BenMAP-CE includes 2010 U.S. Census block-level age, race,
ethnicity, and gender-differentiated data which the program can aggregate to various
grid-level definitions selected by the user, including the 12 km grid coverage used for
risk modeling in this analysis. In addition, BenMAP-CE has the ability to project future
demographics using county-level projections provided by Woods & Poole, 2015. See
BenMAP-CE manual Appendix J and the Estimating PM2.5 and Ozone- Attributable
Health Benefits TSD associated with the 2021 RCU for additional detail (U.S. EPA,

2021).

• Baseline incidence data for mortality endpoints: County-level mortality and population
data from 2012-2014 for seven causes of death in the contiguous U.S. was obtained from
the Centers for Disease Control (CDC) WONDER database. To estimate values for 2015,
we applied annual adjustment factors, based on a series of Census Bureau projected
national mortality rates for all-cause mortality. See BenMAP-CE manual Appendix D for
additional detail.

C.1.3 Study Area Selection

In selecting U.S. study areas for inclusion in the risk assessment, we focus on the
following characteristics:

• Available Ambient Monitors: We have greater confidence in estimating and simulating air
quality concentrations over areas with relatively dense ambient monitoring networks, as
the modeled air quality surfaces can be compared with monitored concentrations (air
quality adjustments are described below in section C.1.4).

• Geographical Diversity. Risk assessments including areas that represent a variety of
regions across the U.S. and a substantial portion of the U.S. population can be more
representative.

6 The 2015 model year was the most recent CMAQ modeling platform available at the time of the design of the risk

assessment and represents the central year of the 2014-2016 design value (DV) period. A single modeling year
was used in the risk assessment, rather than modeling risk for the full three-year design value period, because
model inputs for the 2016 period were not available at the time of the study (section C. 1.4.3).

7 https://www.epa.gov/benmap

C-ll

-------
• Ambient PM2.5 Air Quality Concentrations: Based on 2014-2016 design values, 16 CBSA8
areas exceeded either or both the current annual and 24-hr PM2.5 NAAQS. To include a
larger portion of the U.S. in this risk assessment, we also identified CBSA areas with ambient
PM2.5 concentrations below, but near, the current annual and/or 24-hr PM2.5 NAAQS.
Inclusion of such areas in the risk assessment necessitates an upward adjustment to PM2.5 air
quality concentrations in order to simulate just meeting the current standards. Given
uncertainty in how such increases could potentially occur, we select areas requiring a
relatively modest upward adjustment (i.e., no more than 2.0 |j,g/m3 for the annual standard
and 5 |j,g/m3 for the 24-hour standard, based on the 2014-2016 design value period). Areas
that appeared to be strongly influenced by exceptional events were also excluded (section
C.1.4). Using these criteria, 47 urban study areas were identified (PA Figure 3-16 and
Appendix section C.1.3), including 30 study areas where just meeting the current standards is
controlled by the annual standard,9 11 study areas where just meeting the current standards is
controlled by the daily standard,10 and 6 areas where the controlling standard differed
depending on the air quality adjustment approach (PA Figure 3-16).11

Applying these criteria resulted in the inclusion of 47 core-based statistical areas
(CBSAs). These 47 study areas are identified in Figure C-2, with colors indicating whether they
meet either or both the design value cutoffs. Please note, meeting the criteria for inclusion does
not mean the areas exceed the current annual and/or 24-hr PM NAAQS standards. Green
indicates areas that only exceed a 24-hr design value of 30 |ig/m3, blue indicates areas that only
exceed an annual design value of 10 |ig/m3, and red indicates areas that exceed both the 24-hr
and annual design values.

8 CBSAs (core-based statistical areas) can include one or more counties. Each CBSA selected included at least one

monitor with valid design values and several CBSAs had more than 10 monitors. See Table C-3 in Appendix C.

9 For these areas, the annual standard is the "controlling standard" because when air quality is adjusted to simulate

just meeting the current or potential alternative annual standards, that air quality also would meet the 24-hour
standard being evaluated.

10 For these areas, the 24-hour standard is the controlling standard because when air quality is adjusted to simulate
just meeting the current or potential alternative 24-hour standards, that air quality also would meet the annual
standard being evaluated. Some areas classified as being controlled by the 24-hour standard also violate the
annual standard.

11 In these 6 areas, the controlling standard depended on the air quality adjustment method used and/or the standard
scenarios evaluated.

C-12

-------
Longitude

_MrtO dAtA Cj?019 Gooote. INFOI

Above 10 annual and 30 daily
| Above 30 daily
Above 10 annual

(D 40-

35-

Jk Gulf of

-100 -90

Figure C-2. Map of the areas modeled in the risk assessment, colored by 2014-2016 PM2.5
design values (DV).

These 47 urban study areas include many highly populated CBSAs (Figure C-3 and
Figure C-4). The population at or above the age of 30 in these areas includes roughly 58.4
million people, or approximately 30% of the total U.S. population above the age of 30.
Additional age-specific population information corresponding to each identified mortality study
can be found in Table C-2.

C-13

-------
Q>

o%M ¥

* ^T3

2018 Population

¦ 0 to 8,880

¦ 8.880 to 18.600
18.600 to 36,800
36,800 to 93.800

'¦ 93.800 to 10.300.000

Figure C-3. Map of the 2018 U.S. population by CBSA, with the selected urban study areas
outlined.

C-14

-------
CBSA

New York-Newark-Jersey City, NY-NJ-PA
Los Angeles-Long Beach-Anaheim, CA
Chicago-Naperville-Elgin, IL-IN-WI
Philadelphia-Camden-Wilmington, PA-NJ-DE-MD
Houston-The Woodlands-Sugar Land, TX
Atlanta-Sandy Springs-Rosweil, GA
Detroit-Warren-Dearborn, Ml
Riverside-San Bernardino-Ontario, CA
St. Louis, MO-IL
Pittsburgh, PA
Sacramento-Roseville-Arden-Arcade, CA
Cincinnati, OH-KY-IN
Las Vegas-Henderson-Paradise, NV
Indianapolis-Carmel-Anderson, IN
Cleveland-Elyria, OH
Louisville-Jefferson County, KY-IN
Birmingham-Hoover, AL
Salt Lake City, UT
Fresno, CA
Akron, OH
Bakersfield, CA
Little Rock-North Little Rock-Conway, AR
Stockton-Lodi, CA
McAllen-Edinburg-Mission, TX
Lancaster, PA
Ogden-Clearfield, UT
Modesto, CA
Visalia-Porterville, CA
Canton-Massillon, OH
Provo-Orem, UT
Evansville, IN-KY
South Bend-Mishawaka, IN-MI
San Luis Obispo-Paso Robles-Arroyo Grande, CA

Merced, CA
Macon, GA
Elkhart-Goshen, IN
Napa, CA
Madera, CA
El Centro, CA
Wheeling, WV-OH
Johnstown, PA
Hanford-Corcoran, CA
Altoona, PA
Lebanon, PA
Weirton-Steubenville, WV-OH
Logan, UT-ID
Prineville, OR

OM 1M 2M 3M 4M 5M 6M 7M 8M 9M 10M 11M 12M

Population

Figure C-4. Population counts for ages 30 and above from each of the 47 CBSAs included
in the risk assessment.

C-15

-------
Table C-2. Population of the 47 urban study areas by age range.

Population Age
Range (Years)

Study Area Groupings (Millions)

Studies Using Age Range

30 (Annual-
Controlled)

11 (24-hr-
Controlled)

0-99

Baxter et al., 2017 and Ito et al., 2013

98.5

82.5

7.2

30-99

Turner et al., 2016

58.4

49.5

3.9

65-99

Di et al., 2017 and Zanobetti et al., 2014

13.2

11.1

0.8

As noted in section 3.4 of the PA and illustrated in Figure C-5, the 47 urban study areas
include 30 study areas where just meeting the simulated standards is controlled by the current
annual standard (12.0 |ig/m3), 11 study areas where just meeting the simulated standards is
controlled by the current 24-hr standard (35 |ig/m3), and 6 study areas where just meeting the
simulated standards is controlled by either the annual or 24-hr standard, depending on the air
quality scenario and adjustment strategy (discussed more fully in section C.1.4).

C-16

-------
Number of Urban Study
Areas (CBSAs)

Controlling
Standard

Population (>30
years old)

30 Annual (Blue) ~50M

11 Daily (Green) ~4M

6 Mixed (Grey) ~5M

Total: 47 -60M

Figure C-5. Map of 47 Urban Study Areas Reflected in Risk Modeling Identifying Subsets
Reflected in Risk Modeling (population estimates in millions of people).

C.1.4 Generation of Air Quality Inputs to the Risk Assessment

As described in detail below, air quality modeling was used to develop gridded PM2.5
concentration fields for the risk assessment. A PM2.5 concentration field for 2015 was developed
using a Bayesian statistical model that calibrates chemical transport model (CTM) predictions of
PM2.5 to surface measurements (Chapter 2). The 2015 PM2.5 concentration field was then
adjusted to correspond to just meeting the existing and potential alternative standards using
response factors developed from CTM modeling with emission changes relative to 2015. The
modeling approach applies realistic spatial response patterns from CTM modeling to a
concentration field, similar to those used in a number of recent epidemiologic studies, to
characterize PM2.5 fields at 12 km resolution for study areas.

C-17

-------
The adjustments to simulate just meeting the current standards and alternative standards
are approximations of these air quality scenarios. In reality, changes in PM2.5 in an area will
depend on what emissions changes occur and the concentration gradients of PM2.5 will vary
across an area accordingly. For our analyses, two different adjustment approaches were applied
to provide two outcomes that could represent potential bounding scenarios of PM2.5
concentrations changes across the study area. The two adjustment approaches used to guide the
generation of these modeled surfaces were:

• Primary PM-based modeling approach (Pri-PM): This modeling approach simulates air
quality scenarios of interest by preferentially adjusting direct (i.e., primary, directly-
emitted) PM emissions. As such, the changes in PM2.5 tend to be more localized near the
direct emissions sources of PM. In locations for which air quality scenarios cannot be
simulated by adjusting modeled primary emissions alone, SO2 and NOx precursor
emissions are additionally adjusted to simulate changes in secondarily formed PM2.5.

• Secondary PM-based modeling approach (Sec-PM): This modeling approach simulates
air quality scenarios of interest by preferentially adjusting SO2 and NOx precursor
emissions to simulate changes in secondarily formed PM2.5. In this case, the reductions in
PM2.5 tend to be more evenly spread across a study area. In locations for which air quality
scenarios cannot be simulated by adjusting precursor emissions alone, a proportional
adjustment of air quality is subsequently applied.

The air quality surfaces generated using these two approaches are not additive. Rather, they
should be viewed as reflecting two different broad strategies for adjusting ambient PM2.5 levels.

In addition, we also employed linear interpolation and extrapolation to simulate air
quality under two additional alternative annual standard levels, 11.0, 9.0, and 8.0 |ig/m3,
respectively (section 3.4.1.3 of the PA, Figure 3-15). Interpolation and extrapolation were only
performed for grid cells in the subset of 30 urban study areas where the annual standard was
controlling in both Pri-PM and Sec-PM simulated air quality scenarios of both 12/35 and 10/30
standard combinations. The interpolation and extrapolation were completed at the grid-cell level
based on values simulated using hybrid air quality modeling to just meet the current annual
standard of 12.0 ug/m3 and alternative annual standard of 10.0 ug/m3 (section 3.4.1.3 of the PA,
Figure 3-15). A similar linear extrapolation/interpolation was not conducted for additional 24-hr
standards due to the weaker relationship between the 98th percentile of 24-hr PM2.5
concentrations, which are most relevant for simulating air quality that just meets the 24-hour
standard, and the concentrations comprising the middle portion of the PM2.5 air quality
distribution, which are most relevant for estimating risks based on information from
epidemiologic studies (i.e., discussed further in sections 3.1.2 and 3.2.3.2 in the PA).

The sections below provide more detailed information on the air quality modeling
approach used to adjust air quality to simulate just meeting the current or alternative primary

C-18

-------
PM2.5 standards. Tables containing PM2.5 DVs for the air quality projections can be found in
section C.6.

C.l.4.1 Overview of the Air Quality Modeling Approach

To inform risk calculations, recent PM2.5 measurements were analyzed to characterize the
magnitude and spatial distribution of PM2.5 concentrations. These data were then coupled with
air quality modeling data to project ambient air quality levels corresponding to just meeting the
existing and alternative PM2.5 NAAQS12 in specific areas. An overview of the approach is
provided in Figure C-6. The process starts by acquiring PM2.5 monitoring data from EPA's Air
Quality System (AQS)13 and simulating PM2.5 concentrations with the Community Multiscale
Air Quality (CMAQ)14 model for base case and emission-sensitivity scenarios (Figure C-6, Box
1). The monitored and modeled data are then fused using the Downscaler model and the
Software for Model Attainment Test-Community Edition (SMAT-CE)15 to develop a baseline
spatial field of PM2.5 concentrations and relative response factors (RRFs) for projecting PM2.5
concentrations, respectively (Figure C-6, Box 2). PM2.5 concentrations are projected in two main
steps using output from Downscaler and SMAT-CE (Figure C-6, Box 3). First, the PM2.5
concentrations measured at monitoring sites in an area are iteratively projected using the RRFs to
identify the percent change in anthropogenic emissions required for the highest monitored DV in
the area to just meet the controlling standard. Second, gridded spatial fields of PM2.5
concentrations are projected using the area-specific percent emission change16 that corresponds
to just meeting the standard at the controlling ambient data site. Additional details on the method
are provided in (Kelly et al., 2019a; application of the method to the PM NAAQS risk
assessment is described in the remainder of this appendix.

12 The phrase, "just meeting the PM2.5 NAAQS" is defined as the conditions where the highest design value (DV) for
the controlling standard in the area equals the existing or alternative NAAQS level under consideration. DVs are
statistics used in judging attainment of the NAAQS (www.epa.gov/air-trends/air-qnalitv-design-vaines').

13 www.epa.gov/aas

14 www.epa. gov/cmaq

15 www.epa.gov/scram/pfaotoefaemica 1-modeling-tooIs

16 Scenarios based on a statistical projection approach were also developed for certain cases as discussed below.

C-19

-------
Figure C-6. Overview of the system for projecting PM2.5 concentrations to correspond to
just meeting NAAQS. See section C. 1.4.6 and Kelly et al., 2019a for more details.

C.l.4.2 PM2.5 Monitoring Data and Area Selection

The 2014-2016 DV period was the most recent period having a complete set of total and
speciated PM2.5 observations available at the time of the study. PM2.5 concentrations from the
2014-2016 DV period were used in selecting study areas and as the starting point for air quality
projections (Figure C-6, Box 1, "AQS"). Total and speciated PM2.5 concentrations for the 2014-
2016 DV period were acquired from AQS. For sites in Los Angeles and Chicago, DVs were
invalid during the 2014-2016 period. Los Angeles and Chicago have large populations, recent
valid DVs for sites in Los Angeles are above existing standards, and Chicago is part of a CBSA
that includes sites with valid 2014-2016 DVs in Indiana. For these reasons, invalid data for sites
in these areas were replaced with valid data from other recent periods to enable DVs to be
approximated for inclusion in the assessment. Specifically, for sites in Los Angeles and Orange
Counties in California, observations from April - October 2014 were replaced with observations
from the same months in 2013. For sites in Cook, DuPage, Kane, McHenry, and Will Counties in
Illinois, observations from January to mid-July 2014 were replaced with observations from the
same months in 2015.

Of the 56 areas initially identified as above the 10/30 selection threshold17, DVs for seven
areas18 appeared to meet the threshold due to the influence of wildfires. The influence of

17 " 10/30" indicates an annual standard level of 10 ng/ m3 and a 24-hr standard level of 3 ng m~3

18 Butte-Silver Bow, MT; Helena, MT; Kalispell, MT; Knoxville, TN; Medford, OR; Missoula, MT; and Yakima,
WA

C-20

-------
wildfires on DVs for these areas was estimated in part by recalculating 2014-2016 DVs with
days removed that were clearly associated with summertime wildfires in the northwest. Since
wildfire influence is often excluded when judging NAAQS attainment, these seven areas were
excluded from further consideration. Additionally, the Eugene, OR CBS A was excluded. One
monitor in the Eugene CBS A has a 24-hr 2014-2016 DV slightly above the 10/30 selection
threshold19, but the monitor is in a small valley in Oakridge with very local high concentrations
of PM2.5 in winter that are distinct from conditions in the broader CBSA. Finally, the Phoenix-
Mesa-Scottsdale, AZ CBSA was excluded. This CBSA had one monitor slightly above the 10/30
DV threshold20, but projecting concentrations for the CBSA was judged to be relatively uncertain
because the annual DV is invalid at the only site that exceeded the threshold and the 24-hr DV is
just above the threshold.

The remaining 47 CBSAs were selected for the risk assessment. These areas are shown in
Figure C-7. The maximum 2014-2016 DVs and associated sites for each CBSA are provided in
Table C-3, and the counties associated with the CBSAs are listed in Table C-4. DVs were
calculated to an extra digit of precision for the air quality projections compared with official
DVs. This approach is consistent with DV calculations in previous air quality projections (e.g.,
USEPA, 201221) and provides a precise target for the iterative projection calculations.

19 The 410392013 monitor in Oakridge has a 24-hr 2014-2016 DV of 31 |ig m 3

20 The 040213015 monitor in the Phoenix-Mesa-Scottsdale, AZ CBSA has 24-hr 2014-2016 DV of 31 |ig m 3

21 USEPA (2012) Regulatory Impact Analysis for the Final Revisions to the National Ambient Air Quality
Standards for Particulate Matter. Office of Air Quality Planning and Standards, Health and Environmental
Impacts Division, Research Triangle Park, NC 27711. EPA-452/R-12-005 Available:
https://www3.epa.gov/ttn/ecas/regdata/RIAs/finairia.pdf

C-21

-------
2 ^ —Wan date l2"1H Geooifi. inegi

-120 -110

-100 -90

Longitude

Above 10 annual and 30 daily
Above 30 daily
Above 10 annual

Figure C-7. CBS As selected for the risk assessment. Colors indicate whether the maximum
2014-2016 DVs in the CBS A are above the annual (10 (ig/m3) and/or 24-hr (30 |ig/m3)
selection criteria.

C-22

-------
Table C-3. Maximum annual and 24-hr PM2.5 DVs for 2014-2016 and associated sites for
selected CBSAs.

CBSA Name

# of

Annual

Annual
Max 14-16
DV

24-hr Max

Sites

Max Site

Site

14-16 DV

Akron, OH

391530017

10.99

391530017

23.7

Altoona, PA

420130801

10.11

420130801

23.8

Atlanta-Sandy Springs-Roswell, GA

131210039

10.38

131210039

19.7

Bakersfield, CA

060290016

18.45

060290010

70.0

Birmingham-Hoover, AL

010732059

11.25

010730023

22.8

Canton-Massillon, OH

391510017

10.81

391510017

23.7

Chicago-Naperviiie-Elgin, IL-IN-Wia

170313103

11.10

170310057

26.8

Cincinnati, OH-KY-IN

390610014

10.70

390170020

24.2

Cleveland-Elyria, OH

390350065

12.17

390350038

25.0

Detroit-Warren-Dearborn, Ml

261630033

11.30

261630033

26.8

El Centra, CA

060250005

12.63

060250005

33.5

Elkhart-Goshen, IN

180390008

10.24

180390008

28.6

Evansville, IN-KY

181630023

10.11

181630016

22.0

Fresno, CA

060195001

14.08

060190011

53.8

Hanford-Corcoran, CA

060310004

21.98

060310004

72.0

Houston-The Woodlands-Sugar Land, TX

482011035

11.19

482011035

22.4

Indianapolis-Carmel-Anderson, IN

180970087

11.44

180970043

26.0

Johnstown, PA

420210011

10.68

420210011

25.8

Lancaster, PA

420710012

12.83

420710012

32.7

Las Vegas-Henderson-Paradise, NV

320030561

10.28

320030561

24.5

Lebanon, PA

420750100

11.20

420750100

31.4

Little Rock-North Little Rock-Conway, AR

051191008

10.27

051191008

21.7

Logan, UT-ID

490050007

6.95

490050007

34.0

Los Angeles-Long Beach-Anaheim, CAa

060371103

12.38

060371103

32.8

Louisville/Jefferson County, KY-IN

180190006

10.64

180190006

23.9

Macon, GA

130210007

10.13

130210007

21.2

Madera, CA

060392010

13.30

060392010

45.1

McAllen-Edinburg-Mission, TX

482150043

10.09

482150043

25.0

Merced, CA

060470003

11.81

060472510

39.8

Modesto, CA

060990006

13.02

060990006

45.7

Napa, CA

060550003

10.36

060550003

25.1

New York-Newark-Jersey City, NY-NJ-PA

360610128

10.20

340030003

24.5

Ogden-Clearfield, UT

490570002

8.99

490110004

32.6

Philadelphia-Camden-Wilmington, PA-NJ-DE-
MD

420450002

11.46

421010055

27.5

Pittsburgh, PA

420030064

12.82

420030064

35.8

Prineville, OR

410130100

8.60

410130100

37.6

Provo-Orem, UT

490494001

7.74

490494001

30.9

Riverside-San Bernardino-Ontario, CA

060658005

14.48

060658005

43.2

Sacramento-Roseville-Arden-Arcade, CA

060670006

9.31

060670006

31.4

Salt Lake City, UT

490353006

7.62

490353010

41.5

San Luis Obispo-Paso Robles-Arroyo Grande,
CA

060792007

10.70

060792007

25.9

C-23

-------
CBSA Name

# of
Sites

Annual
Max Site

Annual
Max 14-16
DV

24-hr Max
Site

24-hr Max
14-16 DV

South Bend-Mishawaka, IN-MI

181410015

10.45

181410015

32.5

St. Louis, MO-IL

290990019

10.12

295100007

23.7

Stockton-Lodi, CA

060771002

12.23

060771002

38.7

Visalia-Porterville, CA

061072002

16.23

061072002

54.0

Weirton-Steubenville, WV-OH

390810017

11.75

390810017

27.2

Wheeling, WV-OH

540511002

10.24

540511002

22.5

a DVs for Chicago-Naperville-Elgin, IL-IN-WI and
described in section C. 1.4.2.

.os Angeles-Long Beac

i-Anaheim, CAwere approximated as

Table C-4. Counties associated with selected CBSAs

CBSA Name

Associated Counties

Akron, OH

Portage, Summit

Altoona, PA

Blair

Atlanta-Sandy Springs-Roswell, GA

Barrow, Bartow, Butts, Carroll, Cherokee, Clayton, Cobb, Coweta,
Dawson, DeKalb, Douglas, Fayette, Forsyth, Fulton, Gwinnett,
Haralson, Heard, Henry, Jasper, Lamar, Meriwether, Morgan,
Newton, Paulding, Pickens, Pike, Rockdale, Spalding, and Walton

Bakersfield, CA

Kern

Birmingham-Hoover, AL

Bibb, Blount, Chilton, Jefferson, St. Clair, Shelby, and Walker

Canton-Massillon, OH

Carroll, Stark

Chicago-Naperville-Elgin, IL-IN-WI

Cook, DeKalb, DuPage, Grundy, Kane, Kendall, Lake, McHenry,
Will, Jasper, Lake, Newton, Porter, and Kenosha

Cincinnati, OH-KY-IN

Dearborn, Ohio, Union, Boone, Bracken, Campbell, Gallatin,
Grant, Kenton, Pendleton, Brown, Butler, Clermont, Hamilton, and
Warren

Cleveland-Elyria, OH

Cuyahoga, Geauga, Lake, Lorain, and Medina

Detroit-Warren-Dearborn, Ml

Lapeer, Livingston, Macomb, Oakland, St. Clair, and Wayne

El Centra, CA

Imperial

Elkhart-Goshen, IN

Elkhart

Evansville, IN-KY

Posey, Vanderburgh, Warrick, and Henderson

Fresno, CA

Fresno

Hanford-Corcoran, CA

Kings

Houston-The Woodlands-Sugar Land, TX

Austin, Brazoria, Chambers, Fort Bend, Galveston, Harris,
Liberty, Montgomery, and Waller

Indianapolis-Carmel-Anderson, IN

Boone, Brown, Hamilton, Hancock, Hendricks, Johnson, Madison,
Marion, Morgan, Putnam, and Shelby

Johnstown, PA

Cambria

Lancaster, PA

Lancaster

Las Vegas-Henderson-Paradise, NV

Clark

Lebanon, PA

Lebanon

Little Rock-North Little Rock-Conway, AR

Faulkner, Grant, Lonoke, Perry, Pulaski, and Saline

C-24

-------
CBSA Name

Associated Counties

Logan, UT-ID

Franklin, Cache

Los Angeles-Long Beach-Anaheim, CA

Los Angeles and Orange

Louisville/Jefferson County, KY-IN

Clark, Floyd, Harrison, Scott, Washington, Bullitt, Henry,
Jefferson, Oldham, Shelby, Spencer, and Trimble

Macon, GA

Bibb, Crawford, Jones, Monroe, and Twiggs

Madera, CA

Madera

McAllen-Edinburg-Mission, TX

Hidalgo

Merced, CA

Merced

Modesto, CA

Stanislaus

Napa, CA

Napa

New York-Newark-Jersey City, NY-NJ-PA

Bergen, Essex, Hudson, Hunterdon, Middlesex, Monmouth,
Morris, Ocean, Passaic, Somerset, Sussex, Union, Bronx,
Dutchess, Kings, Nassau, New York, Orange, Putnam, Queens,
Richmond, Rockland, Suffolk, Westchester, and Pike

Ogden-Clearfield, UT

Box Elder, Davis, Morgan, and Weber

Philadelphia-Camden-Wilmington, PA-NJ-
DE-MD

New Castle, Cecil, Burlington, Camden, Gloucester, Salem,
Bucks, Chester, Delaware, Montgomery, and Philadelphia

Pittsburgh, PA

Allegheny, Armstrong, Beaver, Butler, Fayette, Washington, and
Westmoreland

Prineville, OR

Crook

Provo-Orem, UT

Juab and Utah

Riverside-San Bernardino-Ontario, CA

Riverside and San Bernardino

Sacramento-Roseville-Arden-Arcade, CA

El Dorado, Placer, Sacramento, and Yolo

Salt Lake City, UT

Salt Lake, and Tooele

San Luis Obispo-Paso Robles-Arroyo
Grande, CA

San Luis Obispo

South Bend-Mishawaka, IN-MI

St. Joseph and Cass

St. Louis, MO-IL

Bond, Calhoun, Clinton, Jersey, Macoupin, Madison, Monroe, St.
Clair, Franklin, Jefferson, Lincoln, St. Charles, St. Louis, Warren,
and St. Louis city

Stockton-Lodi, CA

San Joaquin

Visalia-Porterville, CA

Tulare

Weirton-Steubenville, WV-OH

Jefferson, Brooke, and Hancock

Wheeling, WV-OH

Belmont, Marshall, and Ohio

C.l.4.3 Air Quality Modeling

Air quality modeling was conducted using version 5.2.1 of the CMAQ modeling system
(Appel, 2018) to develop a continuous national field of PM2.5 concentrations and estimates of
how concentrations would respond to changes in PM2.5 and PM2.5 precursor emissions (Figure C-
6, "CMAQ"). The CMAQ modeling domain (Figure C-9) covered the contiguous U.S. with 12
km horizontal resolution and 35 vertical layers. Since 2015 was the most recent modeling

C-25

-------
platform available at the time of the study and represents the central year of the 2014-2016 DV
period, 2015 was selected as the baseline modeling year for the PM2.5 projections. A single
modeling year was used due to the time and resources needed to conduct photochemical grid
modeling, and because model inputs for the 2016 period were not available at the time of the
study.

Information on the CMAQ model configuration for the 2015 modeling is provided in
Table C-5. The 2015 model simulation and its evaluation against network measurements of
speciated and total PM2.5 has been described in detail previously (Kelly et al., 2019b). Model
performance statistics for PM2.5 organic carbon, sulfate, and nitrate were generally similar to or
improved compared to the performance for other recent national 12 km model simulations. One
exception to the generally good model performance was identified for the Northwest region (OR,
WA, and ID). Model performance statistics for this region were generally not as good as in our
recent modeling due to issues related to unusually high fire influences in 2015, atmospheric
mixing over sites near the Puget Sound, and other factors. However, model performance issues
in the Northwest have minimal influence on the risk assessment, because only two of the 47
CBSAs are in the Northwest region (i.e., Prineville, OR and part of the Logan, UT-ID, CBSA).
Also, the analysis uses ratios of model predictions rather than absolute modeled concentrations,
and systematic biases associated with mixing height and fire impact estimates may largely cancel
in the ratios. Moreover, fusion of monitor data with model predictions in developing PM2.5 RRFs
and the baseline concentration field helps mitigate the influence of biases in model predictions
(as discussed below). Overall, the model performance evaluation (Kelly et al., 2019b) indicates
that the 2015 CMAQ simulation provides concentration estimates that are generally as good or
better than in other recent applications and are reliable for use in projecting PM2.5 in the risk
assessment. Model performance statistics for PM2.5 by U.S. climate region and season are
provided in Table C-6 and statistic definitions can be found in Table C-7.

C-26

-------
Figure C-9. CiVIAQ modeling domain.
Table C-5. CMAQ model configuration.