Technical Support Document (TSD)
for the Final Revised Cross-State Air Pollution Rule Update
for the 2008 Ozone Season NAAQS

Docket ID No. EPA-HQ-OAR-2020-0272

Estimating PM2.5- and Ozone-Attributable

Health Benefits

U.S Environmental Protection Agency

Office of Air and Radiation
Research Triangle Park, North Carolina
March 2021

i


-------
Contents

1	Introduction	1

1.1	Benefits Analysis Background	1

1.2	The Relationship Between Identifying Health Endpoints for Valuation and WTP	2

1.3	Document Purpose and Overview	3

2	Approach to Identifying Studies and Risk Estimates	5

2.1	Study and Risk Estimate Identification Criteria	5

2.1.1	Minimum Criteria	5

2.1.2	Preferred Criteria Categories	6

2.2	Available Epidemiologic Literature	10

2.2.1	Identification of Exposure-Attributable Health Outcomes	10

2.2.2	Identification of Quantifiable Health Outcomes	21

2.2.3	Study Information Table	22

2.2.4	Methods for Presenting Health Benefits Estimates Using Multiple Risk Estimates for a
Single Endpoint	24

2.2.5	PM2.s	25

2.2.6	03	47

2.3	Identified Study and Risk Estimates for Benefits Assessments	57

2.3.1	Health Endpoints	57

2.3.2	Risk Estimates	59

3	Baseline Incidence and Prevalence Estimates	63

3.1	Mortality	65

3.1.1	Mortality Data for 2012-2014	66

3.1.2	Mortality Rate Projections 2015-2060	68

3.1.3	Race-Stratified Incidence Rates	69

3.2	Hospitalizations	69

3.3	Emergency Department Visits	71

3.4	Health Endpoint Onset/Occurrence	71

3.4.1	Acute Myocardial Infarctions (AMIs)	71

3.4.2	Asthma Onset and Symptoms	72

3.4.3	Allergic Rhinitis	74

3.4.4	Lung Cancer	74

3.4.5	Minor Restricted Activity Days (MRAD)	74

ii


-------
3.4.6	School Loss Days	75

3.4.7	Work Loss Days	75

4	Demographic Information	77

5	Health Endpoint Valuation	78

5.1	Mortality	81

5.1.1 Value of a Statistical Life (VSL)	81

5.2	Hospitalizations and Emergency Department Visits	82

5.3	Health Endpoint Onset/Occurrence	85

5.3.1	Acute Myocardial Infarctions (AMIs)	85

5.3.2	Allergic Rhinitis (Hay Fever)	86

5.3.3	Asthma Onset	86

5.3.4	Asthma Symptoms/Exacerbation	87

5.3.5	Cardiac Arrest	88

5.3.6	Lung Cancer	88

5.3.7	Minor Restricted Activity Days (MRADs)	90

5.3.8	School Loss Days	91

5.3.9	Stroke	91

5.3.10	Work Loss Days (WLDs)	92

5.4	Developing Income Growth Adjustment Factors for Health Endpoint Onset/Occurrence	92

6	Characterizing Uncertainty and Evaluating Sensitivity to Alternate Assumptions	96

6.1	Quantitative Characterization of PM2.5 Uncertainty and Evaluating Sensitivity to Alternate
PM2.5 Assumptions	96

6.1.1	Statistical Uncertainty Around the Risk Estimate (Monte-Carlo Assessment)	97

6.1.2	Adult All-Cause Mortality	97

6.1.3	Asthma Onset in Children	101

6.1.4	Cardiovascular Hospital Admissions	103

6.1.5	Respiratory Hospital Admissions	104

6.1.6	Effect Modification of Health Impacts in At-Risk Populations	106

6.2	Quantitative Characterization of 03 Uncertainties and Evaluating Sensitivity to Alternate 03
Assumptions	109

6.2.1	Statistical Uncertainty Around the Risk Estimate (Monte-Carlo Assessment)	109

6.2.2	Respiratory Mortality	109

6.2.3	All-Cause Mortality	110

6.2.4	Asthma Onset in Children	Ill

iii


-------
6.2.5	Understanding the Effect Modification of Health Impacts in At-Risk Populations	113

6.3	Quantitative Characterization of Baseline Incidence Rate Uncertainties	117

6.4	Quantitative Characterization of Economic Valuation Estimate Uncertainties	118

6.4.1	Mortality Cessation Lag	118

6.4.2	Lung Cancer Cessation Lag	118

6.4.3	Income Elasticity of Willingness to Pay	122

6.4.4	Statistical Estimates of VSL	123

6.4.5	Alzheimer's Disease and Parkinson's Disease Onset Lifetime Costs	123

6.5	Qualitative Characterization of Uncertainties	127

6.5.1	Applying Risk Estimates to Locations and Populations not Specified in the Epidemiologic
Study		127

6.5.2	Causality Determination	128

6.5.3	Estimating and Assigning Exposures in Epidemiology Studies	128

6.5.4	Differential Toxicity of PM2.5 According to Chemical Composition	129

6.5.5	Different Long-Term Exposure Windows	129

6.5.6	Discounting Future Benefit Estimates	129

6.5.7	Statistical Estimates of WTP	130

6.5.8	Confounding by Individual Risk Factors	130

6.5.9	Confounding by Other Pollutants	130

6.5.10	Risk Attributable to Long-Term and Short-Term Exposures	130

6.5.11	Heterogeneity of Risk Estimates	131

6.5.12	03 Metrics	131

6.5.13	03 Season	133

6.5.14	Shape of the Concentration-Response Relationship	133

6.5.15	Short-Term Lag Structure	136

6.5.16	Statistical Technique/Model Used to Quantify Risks in Epidemiologic Study	136

6.5.17	Temperature and Weather	137

6.5.18	Unquantified Impacts	137

7 References	138

iv


-------
List of Tables

Table 1. Criteria for Identifying Studies and Risk Estimates for Application in Benefits Assessment	6

Table 2. Study and Risk Estimate Criteria Prioritization Order	8

Table 3. Causality Determinations for PM2.5-Related Health Effects	12

Table 4. Causality Determinations for 03-Related Health Effects	13

Table 5. Study Information Tables	22

Table 6. PM2.5 Study and Risk Estimate Identification Diagram*	27

Table 7. 03 Study and Risk Estimate Identification Diagram	48

Table 8. Set of Health Endpoints for Main PM2.5 Benefits Assessments	58

Table 9. Set of Health Endpoints for Main 03 Benefits Assessments	59

Table 10. Set of Risk Estimates for Main PM2.5 Benefits Assessments	59

Table 11. Set of Risk Estimates for Main 03 Benefits Assessments	61

Table 12. Baseline Incidence Rates for Use in Impact Functions	64

Table 13. National Mortality Rates (per 100 people per year) by Health Endpoint and Age Group, 2012-

2014	67

Table 14. All-Cause Mortality Rate (per 100 people per year), by Source, Year, and Age Group	68

Table 15. Ratio of Future Year All-Cause Mortality Rate to 2013 Estimated All-Cause Mortality Rate, by

Age Group	69

Table 16. Asthma Prevalence Rates	73

Table 17. Weighted Average Asthma Prevalence by Study	73

Table 18. Lung Cancer Incidence Rates	74

Table 19. School Loss Day Rates (per student per year)	75

Table 20. Cost of Illness Economic Study Identification Consideration Factors	79

Table 21. Unit Values for Economic Valuation of Health Endpoints (2015$)1	80

Table 22. Central Unit Value for VSL based on 26-value-of-life studies	82

Table 23. Hospitalization and Emergency Department Cost Elements by Endpoint	82

Table 24. Medical Costs and Hospital Stay Data from the HCUP Database	84

Table 25. Medical Costs for AMIs (2015$)	85

Table 26. Total Valuation Estimates for AMIs (2015$)	86

Table 27. Age-adjusted Belova et al., 2020 Estimates of Lifetime Asthma Costs	87

Table 28. Valuation Estimate for Cardiac Arrests (2015$)	88

Table 29. Latency Periods Used in Lung Cancer Risk Assessment Papers	89

Table 30. Percent Lung and Bronchus Cancer Incidence by Age and Distribution of Risk Reduction by Age

for an Exposure Change at 55	90

Table 31. Income Elasticity Estimates for Minor Health Effects, Severe Health Effects, and Mortality	93

Table 32. Income-Based WTP Adjustments by Health Effect and Year	95

Table 33. Low Concentration PM2.5 Exposures from the ACS CSP-II, Medicare, and CanCHEC Cohorts.... 99
Table 34. Di et al., 2017a PM2.5-Attributable Mortality Risk Estimates per 10 ng/m3 from Different

Exposure Estimation Techniques	100

Table 35. PM2.5-Attributable ACS CSP-II Mortality Risk Estimates per 10 ng/m3 from Different Exposure

Estimation Techniques	100

Table 36. Single- and Two-Pollutant (Including 03 as a Copollutant) PM2.5-Attributable Mortality Risk
Estimates per 10 ng/m3	100

v


-------
Table 37. Di et al., 2017a PIVh.s-Attributable Mortality Risk Estimates per 10 ng/m3 from Different

Statistical Techniques	

Table 38. Potential Sensitivity of Estimated Instances of Asthma Onset	

Table 39. Beta Coefficients and Standard Errors (SE) from Studies of Examining Long-term PM2.5

Exposure and New Onset Asthma in Children	

Table 40. Potential Sensitivity of Estimated Cardiovascular Hospital Admissions	

Table 41. PIVh.s-Attributable Cardiovascular Hospital Admissions Beta Estimates	

Table 42. Potential Respiratory Hospital Admission Sensitivity Insights	

Table 43. PIVh.s-Attributable Respiratory Hospital Admissions Beta Risk Estimates	

Table 44. Comparison of the PIVh.s-Attributable Respiratory Hospital Admissions Beta Risk Estimate to

the EHA Respiratory Estimate	106

Table 45. PM2.5 At-Risk Study Identification Criteria	

Table 46. Identified PM2.5 At-Risk Beta Coefficients and Standard Errors	

Table 47. Single- and Two-Pollutant (Including PM2.5 as a Copollutant) Long-Term 03-Attributable

Respiratory Mortality Risk Estimates per 10 ppb	

Table 48. Single- and Two-Pollutant (Including PMi0 as a Copollutant) Short-Term 03 Exposure 03-

Attributable Excess Premature Respiratory Mortality Risk Estimates per 10 ppb	

Table 49. Long-Term 03-Attributable Total Mortality Risk Estimates per 10 ppb	

Table 50. Potential Sensitivity of Estimated Instances of Asthma Onset	

Table 51. Long-Term 03-Attributable Asthma Beta Coefficients	

Table 52. 03 At-Risk Study Identification Criteria	

Table 53. Identified 03 At-Risk Beta Coefficients and Standard Errors	

Table 54. Scaling Factors for Various Lung Cancer Lag Cessation Distribution Models	

Table 55. Ranges of Elasticity Values Used to Account for Projected Real Income Growth3	

Table 56. Ranges of Adjustment Factors Used to Account for Projected Real Income Growth3	

Table 57. Sensitivity of Monetized Benefits to Alternative Income Elasticities3	

Table 58. Annual Alzheimer's Disease Valuation Estimate Calculation	

Table 59. Lifetime Alzheimer's Disease Valuation Estimate Calculation (2015$)	

Table 60. Additional Lifetime Alzheimer's Disease Valuation Estimate Calculation with a 3% Discount

Rate(2015$)	

Table 61. Annual Parkinson's Disease Valuation Estimate Calculation	

Table 62. Lifetime Parkinson's Disease Valuation Estimate Calculation	


-------
List of Figures

Figure 1. Illustrative Diagram of Potential Biological Pathways of Health Effects Following Pollutant

jre.

Exposu

Figure 2. Potential Biologica

.15

Pathways for Cardiovascular Effects Following Short-Term PM2.5 Exposure

.16

Figure 3. Potential Biologica
Figure 4. Potential Biologica
Figure 5. Potential Biologica
Figure 6. Potential Biologica
Figure 7. Potential Biologica

Pathways for Cardiovascular Effects Following Long-Term PM2.5 Exposure 17
Pathways for Respiratory Effects Following Short-Term PM2.5 Exposure.... 17

Pathways for Respiratory Effects Following Long-Term PM2.5 Exposure	18

Pathways for Cancer Effects Following Long-Term PM2.5 Exposure	18

Pathways for Nervous System Effects Following Long-Term PM2.5 Exposure
	19

Figure 8. Potential Biologica
Figure 9. Potential Biologica

Pathways for Respiratory Effects Following Short-Term 03 Exposure	20

Pathways for Respiratory Effects Following Long-Term 03 Exposure	21

Figure 10. Potential Biological Pathways for Metabolic Effects Following Short-Term 03 Exposure	21

Figure 11. Functional Form of the Identified ACS CSP-II Risk Estimate	30

Figure 12. Functional Form of the Identified Medicare Risk Estimate	31

Figure 13. Cumulative Percentile of PM2.5 Cohort Exposure from the ACS CSP-II, Medicare, and CanCHEC

Cohorts	98

Figure 14. Example County-Level and National-Level Emergency Department Visit and Hospital

Admission Baseline Incidence Data	117

Figure 15. Lung Cancer Cases Cessation Lag Distribution by Model	121

Figure 16. Lung Cancer Cases Reduction Distribution	122

Figure 17. Correlation of MDA8 and DA8 03 Exposures Between 2000-2019 (R=0.986)	132

VII


-------
1 Introduction

1.1 Benefits Analysis Background

The EPA's Guidelines for Preparing Economic Analyses describe the purpose of benefit-cost analysis
(BCA), related economic analyses, and the best-practices for conducting them (U.S. EPA, 2014). As
described in the Guidelines, the fundamental objective of a BCA is to determine whether those who
experience a net gain from a regulatory action can potentially compensate those who experience a net
loss and remain no worse off. These gains and losses are measured by an individual's willingness to pay
(WTP) for, or willingness to accept, changes attributable to the regulatory action. Consistent with
economic theory, the WTP for reductions in exposure to an environmental hazard, like PM2.5 or 03,
depends on the expected effect of those reductions on human health. BCA is the primary tool used for
regulatory analysis and is used to inform the decision of whether the benefits of an action are likely to
justify the costs (EO 12886, 1993, OMB, 2003).

Estimating the health benefits of reductions in PM2.5 and 03 exposure in a BCA begins with estimating
the change in exposure for each individual and then estimating the change in each individual's risks for
those health outcomes affected by exposure. The benefit of the reduction in each health risk is based on
the exposed individual's WTP for the risk change.1 The greater the magnitude of the risk reduction from
a given change in concentration, the greater the individual's WTP, all else equal. The social benefit of the
change in health risks equals the sum of the individual WTP estimates across all of the affected
individuals.2 There are various sources of uncertainty inherent in each of these steps, many of which are
discussed in section 6.

There are three key information collection and assessment steps for implementing this framework for
evaluating the health benefits of changes in exposure:

(1)	Identifying health endpoints affected by exposure by assessing the strength of evidence,

(2)	Identifying suitable empirical estimates of the magnitude of the relationship between

exposure and these health endpoints, and

(3)	Estimating the WTP for reductions in the risk of these health endpoints.

1	As described in section 0, cost-of-illness (COI) estimates are used as a proxy for WTP estimates due to data
limitations.

2	BCA also often report the change in the sum of the risk, or the change in the total incidence, of a health outcome
across the population. If WTP per unit of risk is invariant across individuals, the total expected change in the
incidence of the health outcome across the population can be multiplied by the WTP per unit of risk to estimate
the social benefit of the total expected change in the incidence of the health outcome. Also, if suitable WTP
estimates for a health effect are unavailable, this effect will still, when possible, be quantified to provide a full
picture of the potential benefits of a regulation.

1


-------
This document describes all three steps for the purposes of estimating health benefits from changes in
ambient PM2.5 and 03 exposure.3

1.2 The Relationship Between Identifying Health Endpoints for Valuation and WTP

The first step requires collecting and integrating scientific evidence from different types of studies and
scientific fields (e.g., epidemiologic, controlled human exposure, and animal toxicological studies), as
well as evaluating the quality of evidence and the consistency in the pattern of effects. Determining the
strengths, limitations, and uncertainties in the overall evidence are key components, all of which could
affect WTP, as this information is the basis of the desire to avoid or reduce PM2.5 and 03 exposure.

While the first and third step are presented as independent, they are related for an individual. All else
equal, WTP is expected to be higher when there is stronger evidence of a causal relationship between
exposure to the contaminant and changes in a health outcome (McGartland et al., 2017).4-5 For example,
in the case where there is no evidence of a potential relationship the WTP would be expected to be zero
and the effect should be excluded from the analysis. Alternatively, when there is some evidence of a
relationship between exposure and the health outcome, but that evidence is insufficient to definitively
conclude that there is a causal relationship, individuals may have a positive WTP for a reduction in
exposure to that hazard (Honeycutt, 2020, Kivi and Shogren, 2010). Lastly, the WTP for reductions in
exposure to pollutants with strong evidence of a relationship between exposure and effect are likely
positive and larger than for endpoints where evidence is weak, all else equal. Unfortunately, the
economic literature currently lacks a settled approach for accounting for how WTP may vary with
uncertainty about causal relationships.

Given these challenges, for step 1 the EPA draws its assessment of the strength of evidence on the
relationship between exposure to PM2.5 or 03 and potential health endpoints from the Integrated
Science Assessments (ISAs) that are developed for the NAAQS process. Specifically, in the PM2.5 and 03
benefits analysis for the final Revised Cross-State Air Pollution Rule (CSAPR) Update RIA, the EPA
quantifies and monetizes all health effects that the ISA determines are "causal" or "likely to be causal,"
using scientific assessment methods described in the ISAs. The focus on categories identified as having a
"causal" or "likely to be causal" relationship with the pollutant of interest is to estimate the pollutant-

3	In addition to EPA's Guidelines for Preparing Economic Analyses, these methods and choices adhere to other
relevant EPA and OMB guidance documents, EPA regulations, previous scientific advisory reviews, and available
scientific information (U.S. EPA, 2014).

4	It is also case that the third step depends on sources of uncertainty in the second step. That is, even if a causal
relationship between exposure and a particular health risk were established with certainty, the precise empirical
relationship between exposure and effect may not be known, and of the resulting uncertainty may influence the
WTP to avoid this risk. For example, there may be parameter or model uncertainty in the empirical relationship
between exposure and a health effect that would influence the WTP to avoid exposure (Bleichrodt et al., 2019,
Freeman III et al., 2014). Section 5 describes how WTP estimates may be influenced by these sources of
uncertainty.

5	Here we are referring to causality as a general notion of how well established the relationship between a cause
and possible effect is for the purposes of estimating WTP, and not to the specific approach for evaluating and
determining causality between health effects and PM2.5 and O3 exposure used in the ISAs.

2


-------
attributable human health benefits in which we are most confident.6 All else equal, this approach may
underestimate the benefits of PM2.5 and 03 exposure reductions as individuals may be WTP to avoid
specific risks where the evidence is insufficient to conclude they are "likely to be caus[ed]" by exposure
to these pollutants.7 At the same time, WTP may be lower for those health outcomes for which causality
has not been definitively established. This approach treats relationships with ISA causality
determinations of "likely to be causal" as if they are known to be causal, and therefore benefits could be
overestimated (section 6.5.2). This approach may be revisited in the future with scientific advancements
and development of a theoretically consistent framework that jointly accounts for causal uncertainty
and individuals' WTP for reducing uncertain health impacts.

1.3 Document Purpose and Overview

This is a technical support document (TSD) to the Final Revised CSAPR Update for the 2008 Ozone (03)
Season NAAQS rulemaking. Sections relate to the three key information collection and assessment steps
presented in section 1.1 and detail the methodological approaches used for identifying new benefits
assessment data inputs:

1.	Establish criteria for identifying studies and risk estimates most appropriate to inform a PM? s
and O3 benefit analysis for an RIA (section 2.1). Study criteria, such as study design, location,
population characteristics, and other attributes, were used to identify the most suitable
estimates.8 This step precedes health endpoint identification to ensure impartial health
endpoint identification and prevent identification of non-quantifiable endpoints.

2.	Identify pollutant-attributable health effects for which the ISA reports strong evidence and that
may be quantified in a benefits assessment (section 2.2). EPA considered new evidence reported
in the recent ISAs (U.S. EPA, 2019c, U.S. EPA, 2020a) and clinically significant outcomes (e.g.
premature mortality and hospital admissions) for which endpoint-specific baseline incidence
data is available.

3.	Collect baseline incidence and prevalence estimates (section 3) and demographic information
(section 4). EPA develops either daily or annual baseline incidence and prevalence rates at the
most geographically- and age-specific levels feasible for each health endpoint assessed. EPA
uses population projections based on economic forecasting models developed by Woods and
Poole, Inc. (Woods & Poole, 2015). The Woods and Poole (WP) database contains county-level
projections of population by age, sex, and race out to 2050, relative to a baseline using the 2010
Census data.

6	This decision criterion for selecting health effects to quantify and monetize PM2.5 and O3 is only applicable to
estimating the benefits of exposure of these two pollutants. This decision criterion may not be applicable or
suitable for quantifying and monetizing health and ecological effects of other pollutants.

7	EPA includes an example health endpoint with a causality determination of "suggestive, but not sufficient to
infer" and associated with a potentially substantial economic value in the quantitative uncertainty characterization
(section 6.2.3).

8	If recent ISAs identify more new epidemiologic studies that are better suited than the prior studies for estimating
risks for endpoints whose causality did not change between the prior ISA and the current ISA (e.g. respiratory
hospital admissions), we use this new epidemiologic evidence to estimate risks despite the causality conclusion not
changing between the prior and most recent ISAs.

3


-------
4.	Develop economic unit values (section 5). To directly compare benefits estimates associated
with a rulemaking to cost estimates, the number of instances of each air pollution-attributable
health impact must be converted to a monetary value. This requires a valuation estimate for
each unique health endpoint, and potentially also discounting if the benefits are expected to
accrue over more than a single year. EPA develops valuation estimates at the most age-refined
level feasible for each health endpoint assessed.

5.	Characterize uncertainty associated with quantified benefits estimates (section 6). Building on
EPA's current methods for characterizing uncertainty, these approaches include, among others,
reporting confidence intervals calculated from risk estimates, separate quantification using
multiple studies and risk estimates for particularly influential endpoints (e.g., mortality risk), and
approaches for aggregating and representing the results of multiple studies evaluating a
particular health endpoint.9

9 Study quality, inter-study heterogeneity, and redundancy issues will be taken into consideration if epidemiologic
risk estimates are aggregated.

4


-------
2 Approach to Identifying Studies and Risk Estimates10

This section describes the criteria EPA applies to available fine particulate matter (PM2.5) and ozone (03)
epidemiologic studies and risk estimates to quantify air pollution-attributable health impacts for
regulatory purposes, such as Regulatory Impact Analyses (RIAs). We specify the criteria used to identify
the available body of epidemiologic literature potentially suitable for supporting a benefits analysis
(section 2.2); apply the identification criteria to the body of available literature (section 2.2.4.3); and,
finally, present the identified health endpoints and risk estimates (section 2.3) that best characterize risk
to the U.S. population for health impact benefits assessments. The identification criteria precede the
health endpoint identification because epidemiologic studies must meet certain minimum criteria
(section 2.1.1).

2.1 Study and Risk Estimate Identification Criteria

We follow a systematic approach to identify the studies and risk estimates most appropriate to inform a
PM2.5 and 03 benefit analysis for an RIA.11 Epidemiologic studies report estimated risks of population
exposure to one or more pollutants across a variety of geographic locations, age groups, population
attributes, methods for estimating exposure, PM2.5 and 03 concentrations, time periods, study sizes,
follow-up durations, and other attributes. Identification criteria, specified below, provide transparency
into the scientific judgements used for identifying benefit assessment input parameters. These criteria
are similar to those applied in previous EPA RIAs (Table 1) with the primary goal of identifying risk
estimates that best characterize risk from PM2.5 and 03 exposure among the total population located
throughout the U.S.12

2.1.1 Minimum Criteria

All studies must meet the following minimum required criteria to be considered for use in PM2.5 and 03
benefits assessments. These minimum criteria ensure that the subset of studies evaluated include the
information necessary to justifiably quantify health effects when estimating benefits across the U.S.

1. The study must be referenced in the latest externally reviewed ISA or equivalent assessment
(e.g., provisional assessment or supplement) to ensure the literature search and screening
process were performed in a transparent and systematic manner and only included peer-
reviewed research.

10	What we call risk estimates in this document are results from epidemiologic studies characterizing the
magnitude of exposure-related risk. This term is synonymous with several others, including concentration-
response functions, effect estimates, health impact functions, risk models, and beta (P) coefficients.

11	Epidemiological studies estimate the association between exposure to air pollution concentrations and adverse
health outcomes and generally provide a relative comparison about the strength of the relationship between
exposure to air pollution and the health outcome, rather than estimating the absolute health impact of an
exposure (i.e. the number of avoided cases). For example, a 10 ng/m3 decrease in daily PM2.5 levels might be
associated with a decrease in hospital admissions of 5% or a 5 ppb decrease in 8-hour maximum daily ozone
concentration might be associated with a decrease in hospital admissions of 3%. A benefits analysis reports
absolute values with respect to the public health impact of an exposure.

12	See: https://www3.epa.gov/ttn/ecas/docs/ria/naaqs-pm_ria_final_2012-12.pdf

5


-------
2.	The study must have been conducted in either the U.S. or Canada and represent air quality
conditions, affected populations, and other underlying characteristics of the U.S.13

3.	The study must have been epidemiologic in nature, assess either PM2.5 or 03, and report
numerical risks/hazards expressed as per a unit change in pollutant concentration to provide
necessary information for health effect quantification.

2.1.2 Preferred Criteria Categories

Studies meeting the minimum criteria are then evaluated based on various factors, which we call
preferred criteria, in order to identify risk estimates that best characterize risk across the U.S. These
preferred criteria define other important study design features or attributes and are considered
collectively (Table 1). Most criteria described below can be applied to both the studies and risk
estimates, though criteria applicable only to risk estimates are noted.

Importantly, preferred criteria are established prior to study and risk estimate evaluation and these
choices are based on study quality and suitability. Conversely, factors such as the magnitude of the risk
estimates, are not considered when identifying studies and risk estimates.14 Considering these factors
might inadvertently bias our choice of studies or risk estimates.

Preferred criteria (Table 1) are considered simultaneously when identifying studies and risk estimates
best for use in benefits assessment. Table 2 identifies specific attributes within each preferred criterion
that make a particular study more (or less) suitable for identification. In practice, an identified study or
risk estimate identified may not have the ideal attributes for all criteria, thus there needs to be a
simultaneous assessment of the collective merits of any study or risk estimate. This means that the risk
estimates ultimately identified for application in benefits assessment may not be the highest ranked in
each individual preferred study criterion category, but that they rise to the top when all criteria are
considered simultaneously.

Table 1. Criteria for Identifying Studies and Risk Estimates for Application in Benefits Assessment

Criteria1

Description

Study Period

Studies examining a relatively longer period of time (and therefore having more temporal
coverage) are preferred because they have greater statistical power to detect effects (e.g.,
all else being equal, a study over a five year duration would be preferred over a study
duration of one year). Studies that are more recent are also preferred because of possible
changes in pollution mixes, medical care, and lifestyle overtime. When identifying risk
estimates, models with the broadest time coverage and overlapping air quality and health
data are preferred.

Exposure
Estimate

Studies estimating air quality/exposure using a combination of approaches (e.g., remote
sensing techniques ground-truthed by monitoring data) are preferred over those that use a
single method (e.g., monitor data), because multiple measurement methods can reduce

13	While there are differences between the U.S. and Canada, notably with regards to the health care systems, there
is considerable pollutant transport between Canada and the US, ~90% of Canadians live within ~100 miles of the
US border, and ambient PM2.5 concentrations are similar in Canada and the US (Canada, 2016, CBC, 2016, U.S. EPA,
2019b).

14	Forest plots of the magnitudes of central risk estimates and associated confidence intervals from epidemiologic
studies evaluated by the ISAs by health endpoint can be found in the respective ISAs (U.S. EPA, 2019c, U.S. EPA,
2020a). These figures illustrate the heterogeneity in the size of the reported effect among this subset of studies
and risk estimates.

6


-------


exposure estimate bias and generate higher-resolution of estimates than exposure data
from a single source. When available, studies of long-term/chronic exposure are preferred
over short-term exposure (i.e., hours up to 1 month), considering the limitations of each
exposure duration, as risk estimates based on long-term exposures may include some short-
term exposure effects and provide a more comprehensive estimates of health impacts.

Study Type

Among epidemiologic studies that consider long-term exposure (e.g., one month to years),
cohort studies are preferred over case-control15 studies when estimating benefits across the
U.S., as they are more representative of the overall population, and both are preferred over
cross-sectional16 or ecological17 studies because they control for important individual-level
confounding. An exception to the preference for cohort studies is for rare disease, when
case-control studies may have more power and less selection bias. For short-term exposure
studies, case-crossover and time series studies are preferred over cross-sectional or
prevalence studies also because they are better able to control for potential confounders.

Population
Attributes

Study populations representative (in terms of age, sex, race/ethnicity, etc.) of the
population in which health effects are supported are preferred. The most technically
appropriate measures of benefits would be based on health impact functions that cover the
entire sensitive population but allow for effect modification by age, sex, race/ethnicity, or
other relevant demographic factors (e.g., educational status). In the absence of effect
estimates specific to age, sex, preexisting condition status, or other relevant factors, it may
be appropriate to identify effect estimates that cover the broadest population to match
with the desired outcome of the analysis, which for most EPA benefit-cost analyses is total
national-level health impacts. Where both are available, both age-stratified and overall risk
estimates should be considered for inclusion.

Study Location

U.S. or Canadian studies are used exclusively because studies conducted elsewhere may
exhibit influences of potential differences in pollution characteristics, exposure patterns,
medical care system, population behavior, and lifestyle. National estimates are most
appropriate when benefits are nationally distributed; the impact of regional differences may
be important when benefits only accrue to a single area. City-specific risk estimates from
multi-city studies of hospital admissions or emergency department visits for non-fatal
morbidities may be evaluated for site-specific application to the corresponding city.
Canadian studies are considered when U.S study options are limited or less informative. Risk
estimates with the broadest geographic coverage are preferred (e.g., multi-city studies
preferred to single-city studies) because they provide a more generalizable representation
of the health impacts.

Health
Endpoint

To comprehensively capture the suite of attributable public health impacts and increase the
power to detect effects, when estimating hospital admissions and emergency department
visits, broad health endpoints are preferred over narrower, more specific endpoints. For
example, more-inclusive respiratory hospital admissions endpoint would be selected over
combining hospital admissions for various individual respiratory endpoints, such as asthma,
long-term obstructive pulmonary disease, and respiratory infection. Please note, broad
endpoint categories do not overlap (e.g., nervous system effects and respiratory effects), so
there is no potential for double counting impacts.

Study Size

Studies examining a relatively large sample are preferred because they generally have more
power to detect small magnitude effects. A large sample can be obtained in several ways,
including through a large study population, through repeated observations on a smaller
population (e.g., through a symptom diary recorded for a panel of asthmatic children) or

15	Retrospective study in which two groups, differing in a health outcome, are identified and compared based on
some hypothesized causal characteristic or exposure.

16	Analysis of a cross-section of a population at a single point in time.

17	Comparison of groups, rather than individuals.

7


-------


through a case crossover study design. In general, studies of larger numbers of participants
and/or events are preferred.

Pollutant
Concentrations

Studies evaluating air pollutant exposures closer to or below current conditions are
preferred, as the risk associated with exposure may change at different pollutant
concentrations and air pollution concentrations may decrease in the future,

Hazard/Risk
Estimate

Studies evaluating multiple well-established statistical models adjusted for the most
relevant covariates are preferred.

Inclusion of

Other

Pollutants

When estimating the effects of O3 and PM (or other pollutant combinations) jointly, it is
preferable to use properly specified health impact functions that include both pollutants.
Using single-pollutant estimates in cases where both pollutants are expected to affect a
health outcome can lead to double-counting of benefits when pollutants are correlated.

Lag Period

Lag durations were identified according to the hierarchy described in Table A-l of the PM
and O3 ISAs. Briefly, the strongest multi-day/distributed lag periods that are more
biologically plausible are preferred.

O3 Season

Studies and risk estimates of O3 exposure for the full year are preferred over those
estimating O3 exposures in the summer or warm season only, as O3 concentrations can
remain relatively high outside of the standard warm season in many parts of the country. As
such, year-round time coverage can provide a more complete estimate of O3 exposure
health impacts.

O3 Metric

Risk estimates based on changes in the maximum daily 8-hour average (MDA8) O3
concentration are preferred. As discussed in the 2020 PM Policy Assessment (PA), there is
considerable support from human chamber and epidemiologic studies, as well as advice
from EPA's Clean Air Scientific Advisory Committee (CASAC) to support relationships
between an 8-hour exposure period and short- and long-term health impacts of O3 (U.S.
EPA, 2020c).

'Although preferred criteria categories are not hierarchical, not all criteria are weighted equally, and expert judgement is involved.

2.1.2.1.1 Prioritizing Preferred Identification Criteria

Where Table 1 provides general information on how we determine which studies and risk estimates best
characterize U.S. risk, Table 2 describes how the attributes for each of the 13 criteria are prioritized
within each criteria category. Again, we use the overall study information, and studies ultimately
identified generally performed better across all categories. Importantly, the order of prioritization
presented in Table 2 are relative. For example, the most preferred option may be considered only
slightly more preferable than the other alternative.

Table 2. Study and Risk Estimate Criteria Prioritization Order

Criteria

Prioritization Detail (In order of most to least preferred)

Study Period

1.	Most recent years with overlapping air quality and health data

2.	Less recent years with partially overlapping air quality and health data

3.	Studies with air quality monitoring conducted prior to 2000

Exposure
Estimate

1.	Studies estimating exposure using a combination of approaches (e.g., chemical transport
modeling, monitoring data, land use regression techniques, and satellite data)

2.	Studies estimating exposure using some, but not all, of the above approaches (prioritized if
using monitoring data and/or chemical transport modeling)

3.	Studies estimating exposure using monitoring data only (prefer data from federal
reference [FRM] monitors

Study Type

Long-Term Exposure (i.e., Short-Term Exposure (i.e., hours up to one month)
one month to years) Studies Studies

1.	Cohort studies2 1. Case-crossover (each subject serves as own

2.	Case-control studies control)/Time series studies1

8


-------
Criteria

Prioritization Detail (In order of most to least preferred)



2. Cross-sectional/prevalence (population-level) studies

Population
Attributes

Prefer studies that include broad population attributes with diverse race/ethnicities, both
sexes, and broader age groups (e.g., 0-99 as opposed to only ages 0-17 or only ages 65-99)

Study Location

1.	Nationwide coverage (most or all states represented), including rural areas

2.	Nationwide coverage, including only urban areas

3.	Multi-city and multi-state coverage

4.	Multi-city or multi-state coverage

5.	Single-city or -state coverage

Health
Endpoints

1.	Broad hospital admissions and emergency department visit endpoint categories (e.g.,
hospital admissions and emergency department visits for cardiovascular and respiratory
effects as opposed to admissions or emergency department visits by individual ICD codes)
and broad age groups (e.g., 0-99 as opposed to only 0-17 or only 65-99)

2.	Broad hospital admissions and emergency department visit health endpoint categories and
specific age groups

3.Specific hospital admissions and emergency department visit health endpoint categories
and broad age groups

4. Specific hospital admissions and emergency department visit health endpoints and specific
age groups

Note: The first two options are highly preferred over the second two options

Study Size

Larger study size preferred

Pollutant
Concentrations

Pollutant exposures concentrations closest to current conditions preferred.

Hazard/Risk
Estimate

1.	Risk estimates including the most relevant covariates (e.g., age, sex, race, education,
smoking status, etc.)

2.	Risk estimates including some relevant covariates

3.	Risk estimates that do not include relevant covariates

Inclusion of

Other

Pollutants

1.	Multipollutant risk estimates including other pollutants and not likely to be affected by
collinearity among pollutant covariates.

2.	Copollutant risk estimates including either PM2.5 or O3.

3.	Single-pollutant risk estimates.

Lag Period

1.	Distributed lag models

2.	Average of multiple days (e.g., 0-2)

3.	A priori lag days

4.	Individual lag days, using expert judgment to identify the appropriate result to focus on
considering the time course for physiologic changes for the health effect or outcome being
evaluated.

O3 Season

Annual/full-year exposures are preferred over summer/warm season-only O3 exposures for
long-term exposure-related health endpoints. Summer/warm season-only exposures are
preferred over annual/full-year exposures for short-term O3 exposure-related health
endpoints.

O3 Metric

1.	8-hour maximum O3

2.	1-hour maximum O3

3.	24-hour average O3

4.	Other metrics

ICD- International Statistical Classification of Diseases and Related Health Problems

1lf a study presents both case crossover and time series results, case crossover will be identified

2An exception to the preference for cohort studies is for rare disease, when case control studies may have more power and less
selection bias.

9


-------
2.2 Available Epidemiologic Literature

We follow a structured and transparent process, documented below, to identifying epidemiologic
literature from the body of available epidemiologic literature (section 2.1). This involves the
identification of health endpoints that are both attributable to exposure (section 2.2.1) and for which
we can quantify counts of cases (section 2.2.2). This literature is then reviewed using the criteria
identified in Table 1 and Table 2.

2.2.1 Identification of Exposure-Attributable Health Outcomes

Our process for identifying exposure-attributable health endpoints is informed by the findings of the
Integrated Scientific Assessment (ISA), which identifies broad endpoint categories causally related to
pollutant exposure (section 2.2.1.1); these findings are in turn supported by plausible biological
pathways of disease (section 2.2.1.2).

Each potential health endpoint must satisfy the below conditions prior to inclusion in the main benefits
assessment:

•	The broad endpoint category is sufficiently causally related to exposure (section 2.2.1.1)

•	The specific health endpoint is a biologically plausible health effect of exposure (section 2.2.1.2)

The air quality criteria used to support the review of the National Ambient Air Quality Standards
(NAAQS) undergo a structured and transparent review process for evaluating scientific information and
reaching conclusions about causal determinations that are supported by the scientific information for air
pollution exposures and health effects, as presented in the ISAs. To inform the NAAQS, ISAs draw upon
the existing body of evidence to comprehensively evaluate and synthesize policy-relevant air pollution
science. ISAs transparently identify, critically evaluate, and synthesize the current scientific literature,
including epidemiology studies, making them a suitable source of 1) the causal relationships between
exposure and health outcomes18 (section 2.2.1.1), and 2) available epidemiologic literature from which
to identify studies and risk estimates for consideration in benefits assessments (section 2.2.3).

A 2002 National Academy of Science review supported the use of ISAs as the basis for determining
which health endpoints to include in benefits assessment, stating "the goal of health benefits analysis is
to consider all relevant health outcomes" and "a comprehensive discussion of causality is not necessary
for a benefits analysis" if the information is "provided in the scientific documentation for the rule-
making, such as the criteria document and other related reports, and in guidance provided by EPA's
Science Advisory Board" (NRC, 2002). For background, we provide a "brief review of the evidence for
causality" from the most recent ISAs to "provide justification for inclusion and exclusion of specific
health outcomes considered" and to acknowledge "uncertainties] associated with this assumption"
(section 2.2.1). This section of the TSD also provides background information with regard to potential

18 While ISAs form causal determinations for broad endpoint categories (e.g., respiratory effects), which are
generally preferred over specific health endpoints (e.g., hay fever symptoms) for comprehensive benefits
assessments, they do not make causal determinations for each specific health endpoint. Instead, the ISAs provide
information on the strength and consistency of the evidence supporting more specific endpoints within each broad
category. The strength and consistency of evidence supporting relationships with specific health endpoints,
together with the broad category causality determinations, are used when identifying specific health endpoints for
inclusion in benefits assessments.

10


-------
biological plausibility pathways presented in the ISAs (section 2.2.1.2) that support the causality
determinations.

In addition to the causality determinations, ISAs can also serve as a curated source of pollutant-,
exposure-, and endpoint-specific available epidemiologic literature. Each ISA begins with a broad,
thoroughly documented literature search, the results of which undergo several screening stages to
ensure included studies are within the clearly defined scope of each ISA, in order to identify the most
policy-relevant science (U.S. EPA, 2015b, U.S. EPA, 2019c).19 For example, with regard to the PM2.5-
related health effects, the 2019 PM ISA focused on epidemiologic exposures reflecting current PM2.5
levels for health effect categories where the 2009 PM ISA concluded a "causal" or "likely to be causal"
relationship (U.S. EPA, 2009).20 As such, EPA relies on the systematic and externally reviewed ISAs for
criteria pollutant health endpoints and began the process of identifying epidemiologic risk estimates for
PM2.5- and 03-attributable benefits assessment with the literature sets identified in the 2019 PM and
2020 03 ISAs (U.S. EPA, 2019c, U.S. EPA, 2020a). All epidemiologic studies newly considered for use in
benefits estimation are available in a separate Study Information Table, described in section 2.2.3.

2.2.1.1 ISA Causality Determinations

ISAs characterize the strength and consistency of underlying human clinical, animal toxicological, and
epidemiologic evidence in making causality determinations. Generally, to estimate the pollutant-
attributable human health benefits in which we are most confident, we estimate benefits resulting from
health effects that we have high confidence are attributable to pollutant exposure, so we focus on
categories identified as having a "causal" or "likely to be causal" relationship with the pollutant of
interest in the most recently published ISA.21 These causality determinations are applied to broad health
endpoint categories (e.g., mortality, cardiovascular effects, respiratory effects, nervous system effects,
metabolic effects, etc.) using a weight-of evidence approach (U.S. EPA, 2015b, U.S. EPA, 2019c),
according to the rationale described below:22

o Causal relationship- Evidence is sufficient to conclude that there is a causal relationship with
relevant pollutant exposures (e.g., doses or exposures generally within one to two orders of

19	Studies identified for the 2019 PM ISA were based on the review's opening "call for information" (79 FR 71764,
December 3, 2014), as well as literature searches conducted routinely to identify and evaluate "studies and reports
that have undergone scientific peer review and were published or accepted for publication between January 1,
2009 and March 31, 2017. A limited literature update identified some additional studies that were published
before December 31, 2017" (U.S. EPA, 2009, U.S. EPA, 2019c, Appendix, p. A-3). For the 2020 O3 ISA that date was
March 30, 2018. Relevant studies published after these dates were provisionally considered by the EPA for the
final PM and O3 NAAQS 2020 decisions but were not found to materially change any of the broad scientific
conclusions regarding the health effects of PM and (^exposure made in the 2019 PM ISA and 2020 03 ISAs. This
process ensures a thorough and transparent strategy for literature identification.

20	The 2019 PM ISA focuses on studies conducted in areas where mean PM2.5 concentrations are <20 ng/m3 or, in
the case of a multicity study, where more than half of the cities have concentrations <20 ng/m3. However, studies
with mean PM2.5 concentrations exceeding 20 ng/m3 are included if they address specific areas of uncertainty or
where limitations remain in the evidence base, as identified in the 2009 PM ISA, such as copollutant confounding.

21	This is not to imply that there may not be benefits associated with endpoints having a "suggestive of, but not
sufficient to infer, a causal relationship" but rather that there is greater uncertainty associated with estimating
these potential benefits (section 1.2). While these benefits are not included in the main assessment, they may be
included in sensitivity analyses.

22	See Preamble to Integrated Science Assessments, EPA/600/R-15/067,
https://cfpub.epa.gov/ncea/isa/recordisplay.cfm?deid=347534

11


-------
magnitude of recent concentrations). That is, the pollutant has been shown to result in health
effects in studies in which chance, confounding, and other biases could be ruled out with reasonable
confidence. For example: (1) controlled human exposure studies that demonstrate consistent
effects, or (2) observational studies that cannot be explained by plausible alternatives or that are
supported by other lines of evidence (e.g., animal studies or mode-of-action information). Generally,
the determination is based on multiple high-quality studies conducted by multiple research groups.

o Likely to be causal relationship- Evidence is sufficient to conclude that a causal relationship is likely
to exist with relevant pollutant exposures. That is, the pollutant has been shown to result in health
effects in studies where results are not explained by chance, confounding, and other biases, but
uncertainties remain in the evidence overall. For example: (1) observational studies show an
association, but copollutant exposures are difficult to address and/or other lines of evidence
(controlled human exposure, animal, or mode of action information) are limited or inconsistent or
(2) animal toxicological evidence from multiple studies from different laboratories demonstrate
effects but limited or no human data are available. Generally, the determination is based on
multiple high-quality studies.

Conclusions made in the 2019 PM and 2020 03 ISAs regarding the relationships between exposure and
various broad health endpoints, as well as previous determinations from the 2009 PM and 2013 03 ISAs,
are provided below, with "causal" and "likely to be causal" judgements highlighted (Table 3 and Table 4)
(U.S. EPA, 2009, U.S. EPA, 2013, U.S. EPA, 2019c, U.S. EPA, 2020a).23 There were no "causal" or "likely to
be causal" relationships for PM10-2.5 or ultrafine particles in the 2019 PM ISA, so Table 3 focuses on PM2.5
determinations.24 Table 3 also highlights how the causal determinations in the 2019 PM ISA are similar
to, or different from, the determinations from the 2009 PM ISA. Table 4 highlights how the new causal
determinations in the 2020 03 ISA are similar to, or different from, the determinations from the 2013 03
ISA. Sections of the 2019 PM and 2020 03 ISAs related to "causal" and "likely to be causal"
determinations were used as the basis for identifying the set of available epidemiologic literature best
suited for consideration in benefit estimation (U.S. EPA, 2009, U.S. EPA, 2013, U.S. EPA, 2019c, U.S. EPA,
2020a), as discussed in more detail on section 2.2.4.3.

Table 3. Causality Determinations for PM2.s-Related Health Effects

Exposure

Health Outcome

2009 ISA Conclusion

2019 ISA Conclusion



Mortality1

Causal

Causal



Cardiovascular Effects

Causal

Causal



Respiratory Effects

Likely to be causal

Likely to be causal



Nervous System Effects

None

Likely to be causal

Long-term

Cancer

Suggestive of, but not
sufficient to infer

Likely to be causal



Metabolic Effects

None

Suggestive
of, but not sufficient to infer



Male and Female
Reproduction and Fertility

Suggestive of, but not
sufficient to infer

Suggestive of, but not
sufficient to infer

23	Full summaries of causality determinations by exposure duration and health outcome are available in Table ES-
of both the 2019 PM and 2020 O3 ISAs.

24	Ultrafine particles are generally considered to have an aerodynamic diameter less than or equal to 0.1 nm.

12


-------


Pregnancy and Birth
Outcomes

Suggestive of, but not
sufficient to infer

Suggestive
of, but not sufficient to infer

Short-
term

Mortality1

Causal

Causal

Cardiovascular Effects

Causal

Causal

Respiratory Effects

Likely to be causal

Likely to be causal

Metabolic Effects

None

Suggestive
of, but not sufficient to infer

Nervous System Effects

Inadequate to infer

Suggestive
of, but not sufficient to infer

1Total mortality includes all nonaccidental causes of mortality and is informed by findings for the spectrum of morbidity effects
(e.g., respiratory, cardiovascular) that can lead to mortality. Many studies contributing to the total mortality determination
assess all causes of mortality. The proportion of cause-specific deaths differs by analysis.

Table 4. Causality Determinations for 03-Related Health Effects

Exposure

Health Outcome

2013 ISA Conclusion

2020 ISA Conclusion

Long-term

Respiratory Effects

Likely to be causal

Likely to be causal

Cardiovascular
Effects

Suggestive of a causal
relationship

Suggestive of, but not sufficient to infer,
a causal relationship

Metabolic Effects

None

Suggestive of, but not sufficient to infer,
a causal relationship

Total Mortality1

Suggestive of a causal
relationship

Suggestive of, but not sufficient to infer,
a causal relationship

Reproductive
Effects

Suggestive of a causal
relationship

Effects on fertility and reproduction:
suggestive of, but not sufficient to infer,
a causal relationship

Effects on pregnancy and birth
outcomes: suggestive of, but not
sufficient to infer, a causal relationship

Central Nervous
System Effects

Suggestive of a causal
relationship

Suggestive of, but not sufficient to infer,
a causal relationship

Short-
term

Respiratory Effects

Causal

Causal

Total Mortality1

Likely to be causal

Suggestive of, but not sufficient to infer,
a causal relationship

Cardiovascular
Effects

Likely to be causal

Suggestive of, but not sufficient to infer,
a causal relationship

Metabolic Effects

None

Likely to be causal

Central Nervous
System Effects

Suggestive of a causal
relationship

Suggestive of, but not sufficient to infer,
a causal relationship

1Total mortality includes all nonaccidental causes of mortality and is informed by findings for the spectrum of morbidity effects
(e.g., respiratory, cardiovascular) that can lead to mortality. Many studies contributing to the total mortality determination
assess all causes of mortality. The proportion of cause-specific deaths differs by analysis.

2.2.1.2 Biological Plausibility

ISAs establish causality determinations for broad health effect categories (e.g., cardiovascular effects)
and provide information on the strength and consistency of the evidence supporting more specific
endpoints (e.g., heart failure) within each section. Both types of information are utilized in benefits
assessments. Broad causality determinations can support the use of more comprehensive endpoints

13


-------
found in epidemiologic studies of hospital admissions and emergency department visits, whereas the
support of specific endpoints resulting from pollutant exposure may be more relevant to incidence
endpoints such as cardiac arrest.

In forming the key science judgments for each of the health effects categories evaluated, the recent ISAs
draw conclusions about relationships between PM and 03 exposure and health effects by integrating
information across scientific disciplines and related health outcomes and synthesizing evidence from
previous and recent studies. An advancement in these most recent ISAs is the inclusion of biological
plausibility sections that are specific for each exposure duration and broad health outcome category for
which causality determinations are formed. These discussions outline potential pathways along the
exposure-to-outcome continuum and provide plausible links between pollutant inhalation and health
outcomes at the population level. We include unedited diagrams from the biological plausibility sections
of the 2019 PM and 2020 03 ISAs here to provide information regarding the plausibility of individual
health endpoints resulting from PM and/or 03 exposures.

Biological plausibility can strengthen the basis for causal inference (U.S. EPA, 2015b). In the recent ISAs,
biological plausibility is part of the weight-of-evidence analysis that considers the totality of the health
effects evidence, including consistency and coherence of effects described in experimental and
observational studies. Although there is some overlap in the potential pathways between the ISA health
effects chapters, each biological plausibility section is tailored to the health outcome category,
pollutant, and exposure duration being evaluated within the respective section of each ISA health
effects chapter. Diagrams illustrate possible pathways relating exposure to evidence evaluated in
current and previous assessments, considering physiology and pathophysiology (Figure l).25 These
diagrams portray the available evidence that supports the biological plausibility of exposures leading to
specific health outcomes, but does not provide information on the weight of evidence supporting each
biological pathway (section 2.2.1.2). Gaps and limitations in the evidence base, shown by the absence of
a connecting arrow, correspond to gaps in the figure.

Each box represents evidence that has been demonstrated in a study or group of studies for a particular
effect related to exposure. While most of the studies used to develop the figures are experimental
studies (i.e., animal toxicological and controlled human exposure studies), some observational
epidemiologic studies also contribute to the pathways. These epidemiologic studies are generally: 1)
panel studies that measure the same or similar effects as the experimental studies (and thus provide
supportive evidence) or 2) emergency department and hospital admission studies or studies of
mortality, which are effects observed at the population level. The boxes are arranged horizontally, with
boxes on the left side representing initial effects that reflect early biological responses and boxes to the
right representing potential intermediate (i.e., subclinical or clinical) effects and potential effects at the
population level. The boxes are color coded according to their position in the exposure-to-outcome
continuum.

The arrows that connect the boxes indicate a potential progression of effects resulting from exposure. In
most cases, arrows are dotted (Figure 1, Arrow 1), denoting a possible relationship between the effects.
While most arrows point from left to right, some arrows point from right to left, reflecting progression

25 Information in the biological plausibility diagrams includes studies identified in previous ISAs and Air Quality
Criteria Documents (AQCDs).

14


-------
of effects in the opposite direction or a feedback loop (Figure 1, Arrow 2). In a few cases, the arrows are
solid (Figure 1, Arrow 2), indicating that progression from the upstream to downstream effect occurs as
a direct result of exposure. This relationship between the boxes, where the upstream effect is necessary
for progression to the downstream effect, is termed "essentiality" (OECD, 2016). Evidence supporting
essentiality is generally provided by experimental studies using pharmacologic agents (i.e., inhibitors) or
animal models in which the molecular pathway is obstructed. The use of solid lines, as opposed to
dotted lines, reflects the availability of specific experimental evidence that exposure results in an
upstream effect which is necessary for progression to a downstream effect, for example, by a genetically
deficient model or a chemical inhibitor used in an experimental study involving pollutant exposure.

In the diagrams, upstream effects are sometimes linked to multiple potential downstream effects. Boxes
represent the effects for which there is experimental or epidemiologic evidence related to air pollutant
exposure, and the arrows indicate a proposed relationship between those effects. To illustrate the
proposed relationship using a minimum number of arrows, downstream boxes are grouped together
within a larger shaded box and a single arrow (Figure 1, Arrow 3) connects the upstream single box to
the outside of the downstream shaded box containing multiple green boxes. Multiple upstream effects
may similarly be linked to a single downstream effect using an arrow (Figure 1, Arrow 4) that connects
the outside of a shaded box which contains multiple boxes, to an individual box. In addition, arrows
sometimes connect one individual box to another individual box that is contained within a larger shaded
box (Figure 1, Arrow 2) or two individual boxes both contained within larger shaded boxes (Figure 1,
Arrow 5). Thus, arrows may connect individual boxes, groupings of boxes, and individual boxes within
groupings of boxes depending on the proposed relationships between effects represented by the boxes.
Population level effects generally reflect results of epidemiologic studies. When there are gaps in the
evidence base, there are complementary gaps in the figure and the accompanying text below.

Note: For additional information, please refer to the original biological plausibility diagrams in the ISAs (U.S. EPA, 2019c, U.S.
EPA, 2020a).

Figure 1. Illustrative Diagram of Potential Biological Pathways of Health Effects Following Pollutant
Exposure.

15


-------
2.2.1.2,1 PM;.y Attributable Endpoints and Biological Plausibility

Below are the ISA biological plausibility diagrams for PM2.5 and 03 endpoints judged to have either a
"causal" and "likely to be causal" relationship with pollutant exposure in the 2019 PM and 2020 03 ISAs,
as well as information on which of the endpoints identified in the diagrams have or have not been
previously included in benefits assessments. These diagrams have been reproduced verbatim from the
ISAs, for the convenience of the reader, and no new independent judgements are rendered regarding
biological plausibility in this TSD. Although it is not possible to develop a biological plausibility diagram
for total mortality, taken together, the individual endpoint-specific biological plausibility diagrams each
provide potential pathways by which PM2.5 exposures could result in mortality.

2.2.1.2.1.1 Cardiovascular Effects

The 2019 PM ISA diagram of biological pathways for cardiovascular effects following short-term PM2.s
exposure includes emergency department visits and hospital admissions as population level effects, for
which EPA has historically presented benefits impact estimates (Figure 2). The diagram also includes
mortality as a key endpoint, which EPA has not included in benefits estimates due to the possibility of
overlap with all-cause mortality impacts from long-term exposure resulting in double counting.

Thrombosis

Figure 2. Potential Biological Pathways for Cardiovascular Effects Following Short-Term PM2.5 Exposure

The 2019 PM ISA diagram of biological pathways for cardiovascular effects following long-term PM2.5
exposure includes acute myocardial infarctions (AMI; heart attacks) and mortality, which EPA has
historically presented benefits impact estimates for, and conductance abnormalities/arrhythmia, heart
failure, stroke, and thromboembolic disease, which have not been included in previous benefits
estimates (Figure 3).

16


-------
Figure 3. Potential Biological Pathways for Cardiovascular Effects Following Long-Term PM2.5 Exposure

2.2,1.2.1.2 Respiratory Effects

The 2019 PM ISA diagram of biological pathways for respiratory effects following short-term PM2.5
exposure includes emergency department visits and hospital admissions for asthma
exacerbation/symptoms, chronic obstructive pulmonary disease (COPD), and respiratory infections as
key population level health endpoints, for which EPA has historically presented benefits impact
estimates (Figure 4).

Short-
Term
Exposure
to PM2.5

V	/





\



Emergency



Department Visits/





Hospital





Admissions





Asthma





Exacerbation







Emergency





Department Visits/





Hospital





Admissions



v-

COPD

J





Emergency

\



Department Visits/





Hospital





Admissions





Respiratory



V

Infections

J

Figure 4. Potential Biological Pathways for Respiratory Effects Following Short-Term PM2.5 Exposure

The 2019 PM ISA diagram of biological pathways for respiratory effects following long-term PM2.5
exposure includes asthma development/onset and impaired lung function, for which we have not
previously presented benefits impact estimates (Figure 5).

17


-------
Figure 5. Potential Biological Pathways for Respiratory Effects Following Long-Term PM2.5 Exposure

2.2.1.2.1.3 Cancer

The diagram of biological pathways of cancer following long-term PM2.5 exposure is provided (Figure 6).,
This relationship was "suggestive" in the 2009 PM ISA and "likely to be causal" in the 2019 ISA (U.S. EPA,
2009, U.S. EPA, 2019c).

As cancer is a long-term disease, the 2019 PM ISA did not provide a diagram of biological pathways for
cancer following short-term PM2.5 exposure.

Exposure
to PM2.5

x	/

Figure 6. Potential Biological Pathways for Cancer Effects Following Long-Term PM2.5 Exposure

2.2.1.2.1.4 Nervous System Effects

The 2019 PM ISA diagram of biological pathways for nervous system effects following long-term PM2.s
exposure includes neurodevelopmental disorders, Parkinson's and Alzheimer's disease hospital
admissions and emergency department visits, cognitive decrements/behavioral effects, and cognitive

18


-------
issues (Figure 7). Please note, the weight of evidence supporting each potential is not equivalent and
additional information can be found in the 2019 PM ISA (U.S. EPA, 2019c). As the previous nervous
system effect ISA determination did not rise to the "causal" or "likely to be causal" level, EPA has not
previously included any nervous system endpoints in benefits impact estimates.

Long-Term

PMa.s
Exposure

Systemic
Inflammation

Neuroinflammation
In Adults:
Whole Brain

Neurodegeneration:
I Cortical White Matter
Reductions in Brain
Volume
| Dopaminergic Neurons

in Substantia Nigra
| Hippocampal Neurons

Cognitive
Decrements and
Behavioral Effects

Cognitive Issues,
Some Apolipoprotein E
Allele- Dependent

Impaired
Neurodevelopment:
i Hippocampal Area
Ventriculomegaly
t Corpus Callosum

Area and
Hypermethylation

Parkinson and
Alzheimer's Di:
Hospital Admissions
and Emergency
Department Visits

Neuro-
developmental
Disorders

Figure 7. Potential Biological Pathways for Nervous System Effects Following Long-Term PM2.s Exposure
2.2.1.2.2 03-Attributable Endpoints and Biological Plausibility
2.2.1.2.2.1 Respiratory Effects

The 2020 03 ISA diagram of biological pathways for respiratory effects following short-term 03 exposure
includes emergency department visits and hospital admissions for asthma exacerbation/symptoms and
respiratory infections, for which we have historically presented benefits impact estimates, and lung
function decrements, which EPA has not previously estimates associated benefits (Figure 8). Although
respiratory mortality is supported as a key clinical effect of short-term ozone exposure in the ISA text
and should be included in this diagram, it was mistakenly left out due to the expedited timeline of the
2020 03 ISA.26

26 This information was obtained through conversations with the authors of the 2020 03 ISA.

19


-------
Extrapulmonary
Responses

Decrements in Lung
Function (FVC,
FEVO and
Increased
Respiratory
Symptoms



Emergency
Department Visits/
Hospital
Admissions

Asthma
Exacerbation

Emergency
Department Visits/
Hospital
Admissions
Respiratory
Infections

Figure 8. Potential Biological Pathways for Respiratory Effects Following Short-Term 03 Exposure

The 2020 03 ISA diagram of biological pathways for respiratory effects following long-term 03 exposure
includes mortality, which EPA has included in prior benefits assessments, and asthma
development/onset, fibrotic- or emphysema-like disease/COPD, and altered lung development, which
EPA has not previously included in benefits impact estimates (Figure 9).

Long-
Term
Ozone
Exposure

^			/



M

Altered
Morphology in
Lower Airways and
Alveolar Region

Serotonin
Upregulation/
Altered Neural
innervation

Fibrotic- or
Emphysema-
Like
Disease/COPD

20


-------
Figure 9. Potential Biological Pathways for Respiratory Effects Following Long-Term 03 Exposure

2.2.1.2.2.2 Metabolic Effects

The 2020 03 ISA concluded that short-term exposure was likely to cause metabolic effects and long-term
exposure was suggestive of a causal relationship. Neither short- nor long-term causal determinations
were made for metabolic effects in the 2013 03 ISA, The ISA diagram of biological pathways for
metabolic effects indicates that long-term 03 exposure leads to complications related to diabetes and
changes or contributors to metabolic syndrome, which EPA has not previously included in benefits
assessments.









—



M —





Complications

|



Related to





Diabetes















Changes in





Contributors to





Metabolic





Syndrome

J

v



Figure 10. Potential Biological Pathways for Metabolic Effects Following Short-Term 03 Exposure
2.2.2 Identification of Quantifiable Health Outcomes2

Health endpoints referenced in the latest externally reviewed ISA or equivalent assessment include both
subclinical and clinically relevant endpoints. However, health impacts assessments tend to focus on
quantifying the number of instances of clinically relevant endpoints (e.g., mortality, hospital admissions,
and disease onset/development) and not subclinical endpoints (e.g., inflammation, oxidative stress,
changes in circulation biomarkers, or changes in heart or lung function) for several reasons.

1. Specific baseline event incidence, or the amount of a particular health endpoint present within
the population, is required when using epidemiologic risk estimates to project health impacts of
changes in air quality. Baseline incidence data is more likely to be available for clinically relevant
health endpoints (e.g., hospital admissions for cardiovascular ICD codes 390-459 or the

27 This approach is consistent with the "effect by effect" approach described in the benefits chapter of the
Guidelines for Preparing Economic Analyses (U.S. EPA, 2014). Quantification is treated as separable from
monetization given resource and data limitations (section 1.2).

21


-------
prevalence of asthma in children) then for subclinical endpoints (e.g., forced expiratory volume
or hypertension).

2. Quantifying subclinical endpoints involve additional uncertainties when relating upstream
subclinical effects with clinically relevant downstream health impacts. This does not mean that
subclinical health impacts of PM2.5 and 03 do not exist. In fact, considerably more instances of
pollutant-attributable subclinical effects than clinically relevant effects would be expected.

Although causal determinations are made for broad health endpoint categories (e.g., cardiovascular
effects), the ISAs do review support for specific endpoints (e.g., acute myocardial infarctions) (section
2.2.1.2). Evidence associating specific health endpoints falling under broad health endpoints with "likely
to be causal" or "causal" relationships with air pollution exposures are used to identify comprehensive,
but not overlapping, health endpoints, when suitable studies for quantification based on the criteria
identified above are available.

2,2.3 Study Information Table

Extensive and comprehensive study information is provided for transparency regarding study
comparisons and identification for benefits assessment. Specific study information, corresponding to the
preferred criteria in section 2.1.2, is available for all endpoint-specific, ISA-derived epidemiologic studies
newly considered for use in the main and sensitivity benefits estimation in a separate Excel file titled
Study Information Table for the Estimating PM2.5- and Ozone-Attributable Health Benefits TSD.28,29
Specific studies are listed once per pollutant health endpoint.

Descriptions of the specific types of information extracted from the studies and included in the Study
Information Table are available in Table 5. Studies differed in the type and level of detail of information
provided. Additionally, sometimes information was not reported (NR) by the study and is therefore not
available in the Study Information Table. For example, not all studies provided information on the race,
sex, age range of the study population. Please note, individual studies may be listed multiple times in
the Study Information Table if they report results for multiple endpoints, but they are listed only once
per pollutant endpoint.

Table 5. Study Information Tables

Column Name

Description

Endpoint Group

"Causal" or "likely to be causal" health endpoint included in benefits
assessment.

Endpoint

Specific health endpoint included in benefits assessment.

HERO ID

Identifier used by the Health and Environmental Research Online (HERO)
database. This database is a repository for studies and other references
and is used for various peer-reviewed documents, such as ISAs and
research projects.

First Author

First study author listed.

Publication Year

Year the article was published, according to PubMed.

Pollutant

PM2.5 or 03.

28	Study information is kept in a separate Excel file to support ease of use.

29	Risk estimate information tables are not provided as the identification of suitable risk estimates followed the
hierarchy described in Table A-l of the 2019 PM ISA appendix.

22


-------
Study Type

Epidemiologic study design (e.g., cohort, case control, case crossover, etc.).

Meta-Analysis

Identifies meta-analyses for use when considering pooling city- or regional-
specific estimates to generate an overall risk estimate.

Exposure Duration

Long-term (one month to years) or short-term (hours to less than one
month) exposures.

Study Population

Cohort name or description if unnamed.

Study Size (number of
participants or events)

This can take various forms, such as the number of participants, person-
years, number of hospitalizations/admissions/discharges, or number of
cases and controls. All information provided by the study is included.

Demographics

Demographics of the study participants, such as race/ethnicity, education
level, income level, and socioeconomic status.

Ages

Ages of study participants, with maximum age of 99 reported when
maximum participant age 99 and older.

Exposure Method

Summary of the type of exposure estimate technique. Monitor studies
denote monitor-based studies, and often include land use regression (LUR)
techniques. Hybrid studies include photochemical model and/or satellite
data.

Country

Location of study population (U.S. or Canada).

Study Location

Brief description of the locations included in the study.

Health Years

Years of health data included.

Air Quality Years

Years of air quality data included. Many studies used a specific common
time frame for entire sample, but some used other criteria, such as
exposure over the first year of life.

Pollutant

Concentrations (author-
reported)

Typically, the overall mean and/or median concentrations across study
areas, but sometimes provided information was at a different geographic
and temporal scales (e.g., by state and over multiple years).

LRL/Minimum Exposure
Concentration

The lowest reported pollutant level/concentration.

Pollutant Concentration
Notes

Author-reported exposure estimation method. If multiple types of
exposure estimation techniques were used for an individual pollutant, all
are included.

Outcome Measure

Specific health outcome. Examples include the ICD codes used for hospital
admission and emergency department visits or the criteria used to identify
disease onset.

Lag Periods

For short-term studies, the time period of exposure prior to health effect.

Copollutants

Adjustments for copollutants in the risk estimates.

Covariates/Confounders

Author-reported covariates/confounders included in the risk estimate.

Statistical Technique

Analytical methods used to generate the risk estimate.

Qualitative Limitations

Summary comparison of each study to others investigating the same
pollutant and specific health endpoint.

Relative Determination

Denotes which studies were identified as best characterizing risk as
compared to other available studies for each pollutant endpoint.

23


-------
2.2.4 Methods for Presenting Health Benefits Estimates Using Multiple Risk Estimates for a
Single Endpoint

2.2.4.1 Pooling

If more than one study or risk estimate is suitable for characterizing risk across the U.S. for an individual
health endpoint, we prefer to use multiple risk coefficients to the extent technically feasible. In such
instances, we combine the risk estimates using pooling30 methods in order to avoid a loss of information
from multiple suitable studies; this approach is consistent with advice received from the National
Academies of Sciences (NRC, 2002). Pooling yields a summary mean value estimate and confidence
intervals reflecting variability across the pooled risk estimates. These pooled estimates take into account
both the within-study variances and the between-study variance when weighting.

Unfortunately, the heterogeneous nature31 of epidemiological studies often make them difficult to pool
or otherwise aggregate. For example, it would be inadvisable to combine results from a long-term
exposure study with a short-term exposure study. Other types of study heterogeneity that would
prevent one from aggregating across studies include exposure duration (i.e., short- and long-term),
some population attributes (e.g., age or race/ethnicity), health endpoint outcome measure (e.g., specific
international classification of disease [ICD] codes), and study type (e.g., cohort vs case control).

Combining studies that differ in other aspects can be less straightforward, as there can be both
advantages and disadvantages. For example, recent advancements in exposure estimation methods
allow newer hybrid techniques to estimate pollutant concentrations at more detailed temporal and
spatial scales. As the uncertainties associated with hybrid- and monitor-based exposure estimates vary
and we consider the quality of the exposure estimate during study and risk estimate identification, we
expect pooling risk estimates that vary by exposure technique will increase the confidence in the overall
benefits estimate. We also consider pooling studies that differ in the study period, North American
country, geographic area, pollutant concentrations, included covariates, or regression technique.

Conversely, while consistency between studies is generally desirable when pooling, there are some
instances when it could introduce uncertainty and/or bias. For example, we would not pool multiple
studies of the same cohort over different time periods, as these are not independent results—but rather
different results from the same cohort. Hence, we would instead identify the most recent analysis or the
analysis considering the longest time series of air quality data, so that the study population is not over-
represented when estimating health impacts.32

30	Pooling estimates would be accomplished by performing a meta-analysis, a statistical method of aggregating
independent risk estimates to provide an overall single estimate. EPA has in the past sought to characterize the
magnitude of uncertainty across risk estimates by either applying a fixed effect or random effects pooling
technique to combine two or more risk estimates

31	This is considered a strength when determining whether an outcome is causally linked to a pollutant.

32	If multiple studies of the same cohort suitably characterize risk, multi-step pooling to avoid over-weighting is an
option. This would involve first combining analyses of the same cohort and then combining with estimates from
other cohorts.

24


-------
2.2.4.2	Individual Alternate Risk Estimates

In situations where multiple scientifically robust risk estimates should not or cannot be pooled33, we
instead estimate incidence using each risk estimate independently. Where pooling synthesizes the
results of multiple risk estimates into a single value, presenting multiple estimates from various key
epidemiological studies identified by the latest ISA or equivalent could provide readers with insight to
the plausible range of air pollution-attributable impacts. Therefore, if pooling is infeasible due to the
issues mentioned above, we report individual results from each risk estimate suitable for characterizing
risk across the U.S. for an individual health endpoint separately. Reporting a large number of individual
estimates may characterize the heterogeneity associated with risk but may also make the resulting risk
estimates more difficult to interpret. To keep results manageable, we may report additional estimates as
a quantitative sensitivity analysis.

2.2.4.3	Systematic Identification of Epidemiologic Studies and Risk Estimates for Benefits Assessment
This section describes the systematic application of the identification criteria (section 2.1) to the body of
available epidemiologic studies and risk estimates (section 2.2). Summary information on the number of
available and included studies and risk estimates is presented in Table 6 and Table 7. Descriptions of
endpoint-specific ISA support and available epidemiologic literature are available for each pollutant-
attributable and quantifiable health endpoint.

2.2.5 PM2.s

The following sections of the PM ISA correspond to health endpoints judged as having a "causal" or
"likely to be causal" relationship with PM2.5 exposure:

•	5.1 Short-Term PM2.5 Exposure and Respiratory Effects,

•	5.2 Long-Term PM2.5 Exposure and Respiratory Effects,

•	6.1 Short-Term PM2.5 Exposure and Cardiovascular Effects,

•	6.2 Long-Term PM2.5 Exposure and Cardiovascular Effects,

•	8.2 Long-Term PM2.5 Exposure and Nervous System Effects,

•	10.2 [Long-Term] PM2.5 Exposure and Cancer,

•	11.2 Long-Term PM2.5 Exposure and Total Mortality

Following the approach to identifying available epidemiologic literature (section 2.2), we began with the
2,656 studies cited by the 2019 PM ISA. Of these, 491 studies evaluated mortality or morbidity health
endpoints that the 2019 PM ISA determined as having a "causal" or "likely causal" relationship with
PM2.5 exposure and are clinically relevant (sections 2.2.1 and 2.2.2).34 Of these, 82 studies met the
minimum required criteria (section 2.1.1).35

As studies that evaluated broad and more inclusive hospital admissions and emergency department visit
health endpoints (e.g., hospital admissions including a variety of respiratory endpoints) were preferred
over studies that focused on hospital admissions or emergency department visits for specific health

33	As an example, we often cannot pool across hazard ratios reported in long-term exposure cohort studies. The
challenges associated with synthesizing the results of long-term cohort studies have been described elsewhere
(Burnett et al., 2018).

34	Mortality studies were treated slightly differently. More information is available in section 2.2.5.1.1.

35	This number may not equal the sum of available studies in Table 6 as individual studies may present risk
estimates for multiple health endpoints.

25


-------
endpoints (e.g., hospital admissions for asthma only), we began by focusing on epidemiologic studies
including broad hospital admissions and emergency department visits ICD-9 codes. This reflects strong
support for these broad endpoints in the ISA, the desire to avoid double-counting of health benefits
across categories, and recommendations from advisory board recommendations.36 The remaining
studies were sorted by health endpoint and PM2.5 exposure relationship (i.e., short-term or long-term
exposure to PIVh.s)- These studies included 16 unique health endpoints (Table 6). In Table 6, the number
of available studies refers to the number of North American studies meeting the minimum required
criteria within each health endpoint, and a single study may be relevant to multiple endpoints. The risk
estimates from the different studies for each endpoint can either be pooled (section 2.2.4.1) or kept as
separate estimates (section 2.2.4.2), the latter of which is more common for mortality endpoints. The
two columns to the far left provide the number of available risk estimates from each included study, as
well as the number of risk estimates to be pooled or kept separate for each endpoint.

Once the studies were grouped by health endpoint, we applied the preferred criteria to obtain the final
set of studies to inform each health endpoint. For each of the 16 endpoints, we performed a study
ranking process based on these criteria that emphasized characteristics in Table 2. When no new
epidemiologic studies of health endpoints previously supported by the ISAs were available, the risk
estimates used previously were brought forward (e.g., work loss days).

36 The Health Effects Subcommittee (HES) of the Advisory Council on Clean Air Compliance Analysis (Council)
provided recommendations on the distinction between specific diagnostic codes and broad health outcome
categories in 2004 (Ostro, 2004). The HES recommended "health outcome estimates that can be more closely
linked to the results of epidemiologic studies. However, if in the efforts to achieve a match, the outcome
specification is too narrow (e.g., "acute bronchitis" instead of "all respiratory conditions"), small numbers will
seriously reduce the reliability of the analysis. Therefore, careful consideration of the diagnostic codes to use (with
the related tradeoffs in uncertainty) will be an important step in constructing the baseline data sets."

26


-------
Table 6. PM2.5 Study and Risk Estimate Identification Diagram*

PM Endpoint and
Exposure Duration

Studies
Available

Studies
Included

Ages

Risk
Estimates
Available

Risk
Estimates
Included

Mortality (LT)

2

1

Infants

1

1

19

2

Adults and older adults

19

1

Older adults

64

1

Cardiovascular
Hospital Admissions
(ST)

10

1

Children, adults, and older
adults

28

7

Cardiovascular
Emergency

Department Visits (ST)

1

1

Children, adults, and older
adults

3

1

AMI (ST)

NA

1

Adults and older adults

NA

1

Stroke (LT)

3

1

Older adults

1

1

Cardiac Arrest (ST)

3

3

Adults and older adults

12

1

12

1

7

1

Respiratory Hospital
Admissions (ST)

13

2

Children

3

1

Older adults

4

1

Respiratory
Emergency

Department Visits (ST)

10

1

Children

16

4

Asthma Onset (LT)

5

1

Children

7

1

Asthma Symptoms
(ST)

8

1

Children

8

1

Allergic Rhinitis

1

1

Children

5

1

Minor Restricted
Activity Days

NA

1

Adults and older adults

NA

1

Work Loss Days

NA

1

Adults and older adults

NA

1

Lung Cancer(LT)

4

1

Adults and older adults

24

1

Alzheimer's Disease
(LT)

1

1

Older adults

53

1

Parkinson's Disease
(LT)

3

1

Older adults

53

1

ST- short-term exposure; LT- long-term exposure; NA
studies in the ISA; Risk estimates identified in the 20

not applicable due to the absence of recent available epidemiologic
2 PM NAAQS RIA will continue to be utilized.

*See associated Study Information Table for specific study details.

2.2.5.1 All-Cause Mortality

The 2019 PM ISA concluded that a "causal" relationship exists between both long- and short-term PM2.5
exposure and all-cause mortality. Specifically, the 2019 ISA states that:

Recent U.S. and Canadian cohort studies demonstrate consistent, positive associations
between long-term PM2.5 exposure and mortality across various spatial extents,
exposure assessment metrics, and statistical techniques, and locations, where mean

27


-------
annual average concentrations are <12 /ig/m3. Additionally, the evidence from recent
studies reduce uncertainties related to potential copollutant confounding and continues
to provide strong support for a linear, no-threshold concentration-response relationship.
The body of evidence for total mortality is supported by generally consistent positive
associations with cardiovascular and respiratory mortality. There is coherence of effects
across the scientific disciplines (i.e., animal toxicologicalcontrolled human exposure,
and epidemiologic studies) and biological plausibility for PM2.5-related cardiovascular,
respiratory, and metabolic disease, which supports the PM2.5-mortality relationship. (U.S.
EPA, 2019c, section 11.2.7)

As the biological pathways by which short- and long term PM2.5 exposures are understood to lead to
health effects are quite similar (section 2.2.1.2.1), we assume that effects found in studies of long-term
exposures may include some effects of short-term exposures. Therefore, only mortality impacts from
long-term PM2.5 exposure will be quantified, so as not to overestimate impacts. This may potentially bias
long-term, all-cause PM2.5-attributable mortality impact estimates toward the null in the main benefit
estimate.

Additional support for including estimates of all-cause PM2.5 mortality, as opposed to cause-specific,
comes from the recommendations of advisory boards and review panels. For example, in response to
suggestions made in a 2002 National Resources Council (NRC) report, expert judgement studies of
mortality impacts were conducted (NRC, 2002). The report recommended that "the main quantitative
question should focus on all-cause mortality as the outcome, rather than eliciting separate
concentration-response functions for specific causes of death" (lEc, 2006). Similarly, a more recent
Health Effects Subcommittee (HES) of the Advisory Council on Clean Air Compliance Analysis (Council)
affirmed the inclusion of all-cause long-term PM2.5 estimate with no threshold was "sound" (Hammitt
and Bailar, 2010).

2.2.5.1.1	Available Epidemiologic Literature

Whereas for all other health endpoints we began by identifying North American epidemiologic studies
from the relevant ISA (U.S. EPA, 2019c, Figures 11-17 and 11-18), available literature for this health
endpoint had been further reviewed by EPA in the 2020 PM Policy Assessment (PA) (U.S. EPA, 2020c
section 3.2 and Figure 3-3). As part of this review, the PA identified multi-city studies and more recent
reanalysis or extensions of some of the commonly used cohorts. As such, for this health endpoint we
began with the 19 epidemiologic multi-city cohort studies identified in the 2020 PM PA, which all met
the minimum criteria (section 2.1.1).

We separately evaluated the more limited literature available regarding PIVh.s-attibutable infant
mortality (ages 0-12 months) cited in the 2009 ISA, as no more recent studies of PIVh.s-attributable all-
cause infant mortality were available in the 2019 ISA. Full study information can be found in the Study
Information Table.

2.2.5.1.2	Identifying Suitable Studies for Use in Benefits Assessments

The systematic identification criteria (section 1) was applied to the 19 studies of PIVh.s-attributable long-
term all-cause mortality in adults, which prioritized particularly germane attributes including geographic
coverage, population representativeness, and method of exposure estimation. As it is not relevant to
study identification, we did not consider the risk effect magnitude as a criterion for study identification.

28


-------
The 19 studies varied considerably with regards to all criteria considered. For example, study sizes
ranged from the thousands to the tens of millions. Ultimately, all preferred criteria relevant to PM2.5
factored into the identification of the studies best characterizing risk across the country, although
geographic diversity, exposure estimation technique, population attributes, PM2.5 concentrations, and
inclusion of the copollutant 03 were particularly germane. Specific information can be found in the
corresponding Study Information Table.

2.2.5.1.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
As mortality impacts constitute such a large portion of health impacts and every study has inherent
strengths and limitations, we expect to present mortality estimates using multiple cohort risk estimates.
This approach is also consistent with previous RIAs (e.g., U.S. EPA, 2011b, U.S. EPA, 2011c, U.S. EPA,
2011d, U.S. EPA, 2012a, U.S. EPA, 2012b, U.S. EPA, 2015a, U.S. EPA, 2019a).

The systematic approach led to the initial identification of three studies best characterizing risk across
the U.S. (Di et al., 2017b, Pope et al., 2015, Turner et al., 20 1 6).37-38 These three studies used data from
two cohorts, a retrospective analysis of Medicare beneficiaries (Medicare) and the American Cancer
Society Cancer Prevention II study (ACS CPS-II). The identification of these studies is consistent with the
2019 PM ISA, which concluded that the ACS CSP-II and Medicare cohorts provide strong evidence of the
association between long-term PM2.5 exposure and mortality with support from several additional
cohort studies (U.S. EPA, 2019c). We discuss uncertainty and sensitivity considerations related to the
identified mortality risk estimates in sections 6.1.2 and 6.5.39

2.2.5.1.3.1 Adult Mortality

2.2.5.1.3.1.1 ACS CSP-II

Two independent studies evaluated the same years of data from the large, nationwide ACS CSP-II cohort
of those > 29 years old (Pope et al., 2015, Turner et al., 2016). These studies extended the follow-up
period of the ACS CSP-II to 22 years (1982-2004), evaluating 669,046 participants over 12,662,562
person-years of follow up and 237,201 observed deaths. These two studies applied a more advanced
exposure estimation approach than had previously been used when analyzing the ACS cohort,
combining the geostatistical Bayesian Maximum Entropy framework with national-level land use
regression models.

37	The PM risk assessment, performed as part of the 2020 PM PA, included Di et al., 2017b, Turner et al., 2016, and
Pope et al., 2015 as sources of key PM2.5-attributable mortality risk estimates, further supporting their
identification for benefits assessment. The 2020 PM PA cited a number of relative advantages of these studies
related to the extended period of observation, the rigorous examination of model forms and effect estimates, the
coverage for ecological variables, and the large dataset with regards to both population and area (U.S. EPA, 2019c).

38	The Harvard Six Cities Study, which had been identified for use in estimating mortality impacts in the 2012 PM
NAAQS RIA, was not identified using this approach due to geographic limitations (U.S. EPA, 2012b).

39	There are several other assumptions implicit in the calculation of PM2.5-related mortality impacts. Briefly, these
include 1) an assumption of "cessation" lag in time between the reduction in PM exposure and the full reduction in
mortality risk that affects the timing (and thus discounted monetary valuation) of the resulting deaths, 2) following
conclusions of U.S. EPA, 2019c, we assume that all fine particles are equally potent in causing mortality, and 3)
following conclusions of the U.S. EPA, 2019c, we assume that the health impact function for fine particles is linear
within the range of ambient concentrations affected by these standards.

29


-------
In addition to adjusting for individual-level and ecological covariates, Turner et al., 2016 also controlled
for occupational PM2.5 exposure and adjusted for the potential copollutants 03 and nitrogen dioxide.
Although the copollutant adjustment did not significantly change the hazard ratio, similar to the risk
assessment performed as part of the PM PA (U.S. EPA, 2020c), we identified it as the most suitable
hazard ratio when estimating health benefit impacts.40 Thus, the total mortality risk estimate is based on
the random-effects Cox proportional hazard model that incorporates multiple individual and ecological
covariates41 (relative risk =1.06, 95% confidence intervals 1.04-1.08 per lOpig/m3 increase in PM2.5). The
relative risk estimate is identical to a risk estimate drawn from earlier ACS analysis of all-cause long-term
exposure PIVh.s-attributable mortality (Krewski et al., 2009). Of note, the ACS cohort participants were
recruited by approximately 77,000 ACS volunteers and may not precisely represent the overall U.S.
population demographics.

A depiction of the slope and standard error of the hazard ratio associated with the identified risk
estimate from the minimum to maximum PM2.5 concentrations evaluated is provided (Figure 11). The
static standard error is reflected in the proportionally constant 95% confidence intervals shown with red
dashed lines, depicted as relative to the lowest reported level.

PM2.5 (lug/m3)

Figure 11. Functional Form of the Identified ACS CSP-II Risk Estimate

2.2.5.1.3.1.2 Medicare

The recent Di et al., 2017b analysis evaluated nearly 61 million U.S. Medicare enrollees through 460
million person-years of follow-up and roughly 22 million observed deaths. This cohort comprised
approximately 15% of the total U.S. population, included people living in rural areas, and is one of the
largest cohort studies published to date. The authors modeled PM2.5 exposure across the contiguous

40	Hazard ratios are a subtype of risk estimates.

41	Covariates include: education; marital status; body mass index (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).

30


-------
U.S. using a sophisticated hybrid methodology that included land use regression, satellite data, and
monitor data, and resolved estimations to 1 x 1-kilometer areas. Adjustment for potential confounding
by the copollutant 03 was performed, which slightly attenuated the relationship between PM2.5 and
mortality. The authors also performed statistical testing for the potential of non-linear effects and
concluded that the data supported a nearly linear concentration-response relationship with no signal of
a threshold down to at least 5 ng/m3. This study is restricted to adults over the age of 64, and thus will
only be applied to that age group in benefits assessments.

In addition to the main hazard ratio, Di et al., 2017b presented three additional hazard ratios: one that
excluded the copollutant 03, one that estimated exposure using only monitor data, and one that
evaluated a subset of the population experiencing lower exposures. Of these hazard ratios, only the low-
exposure analysis differed substantially from the others and was considerably higher (HR = 1.136 [1.131-
1.141] per 10 ng/m3 PIVh.s)- However, this analysis was restricted to person-years with both PM2.5
exposures lower than 12 ng/m3 and 03 exposures lower than 50 parts per billion (ppb). Restricting the
analysis in this way reduces the sample size and restricts the air quality concentrations, making
estimates of risk less applicable to the entire U.S. Similarly, we prefer methods for assigning exposure
that leverage both monitor and modeling techniques and models that account for potential copollutant
confounding. Hence, we identified a hazard ratio from the main analysis to be the most appropriate for
use (HR=1.073 [1.071-1.075] per 10 ng/m3 PM25).

A depiction of the slope and 95% confidence intervals of the hazard ratio associated with the identified
risk estimate from the minimum to maximum PM2.5 concentrations evaluated is provided (Figure 12).

PM2.5 (jig/m3)

Figure 12. Functional Form of the Identified Medicare Risk Estimate

2.2.5.1.3.1.3 Summary

EPA previously used the two best estimates of mortality available, one from the ACS cohort and one
from the Harvard Six Cities study. While two estimates were again identified as best characterizing risk
across the U.S., their relative magnitude will depend on the populations included in the analysis (e.g.,
analyses of older or younger populations experiencing higher concentrations will lead to the Medicare
estimate or the ACS CSP-II generating the "higher" estimate, respectively). Qualitatively, the two risk
estimates identified here are both very similar to the previously used Krewski et al., 2009 estimate of
mortality derived from the ACS cohort.

31


-------
2.2.5.1.3.2 Infant Mortality

In addition to the adult mortality studies described above, several studies show an association between
PM exposure and mortality in children under 5 years of age (U.S. EPA, 2009).42The 2019 PM ISA
concluded that evidence exists for a stronger effect at the post neonatal period and for respiratory-
related mortality, although this trend is not consistent across all studies. In addition, compared to
avoided deaths estimated for adult mortality, avoided deaths for infants are significantly smaller
because the number of infants in the population is much smaller than the number of adults and the
epidemiology studies on infant mortality provide smaller risk estimates associated with exposure to PM.
EPA has included estimates of post neonatal infant mortality from Woodruff et al., 1997 (U.S. EPA,
2012a, U.S. EPA, 2019a).

In a more recent study including a larger and more nationally-representative study size by the same
group, authors examined the relationship between long-term exposure to fine PM2.5 air pollution and
post neonatal infant mortality in 3,583,495 births from 96 counties containing >249,999 residents across
the U.S. between 1999-2002 using data from the National Center for Health Statistics (Woodruff et al.,
2008). They linked average PM2.5 monitoring data over the first two months of life with 6,639 post
neonatal deaths, using logistic regression that incorporated generalized estimating equations (GEE) to
estimate the odds ratios for all-cause and cause-specific post neonatal mortality by exposure to air
pollution.43 The study population experienced a median PM2.5 concentration of 14.8 ng/m3, with 25% of
the population experiencing concentrations below 12 ng/m3 and above 18.8 ng/m3. The study included
an evaluation of the appropriateness of a linear form from analysis based on quartiles of exposure and
determined the linear form as a reasonable assumption. Study results included a single risk estimate of
PM2.5-attributable all-cause mortality, 1.04 (0.98-1.11) per 7 ng/m3 (interquartile range) increase in
PM2.5.

2.2.5.2 Cardiovascular Hospital Admissions

The ISA found "generally consistent, positive associations observed in numerous epidemiologic studies
of emergency department visits and hospital admissions for ischemic heart disease, heart failure, and
combined cardiovascular-related endpoints" (U.S. EPA, 2019c section 6.1.16). Also, the ISA calls out
cardiovascular hospital admissions as a population level health endpoint related to short-term PM2.5
exposure (section 2.2.1.2.1.1).

2.2.5.2.1 Available Epidemiologic Literature

Ten North American epidemiologic studies of cardiovascular hospital admissions44 were identified in
section 6.1 of the PM ISA (U.S. EPA, 2019c). Relevant information related to the identification criteria,
including study location, population attributes, and study period, were extracted from the studies and is
available in the associated Study Information Table. The hospital admissions endpoint reports the
number of events, as opposed to the number of individuals who experienced the event.

42	For the purposes of this analysis, we only calculate benefits for infants age 0-12 months, not all children under 5
years old.

43	Odds ratios are a subtype of risk estimates.

44	Of the ~35 million annual hospital discharges, ~20% are related to cardiovascular effects, ~10% to respiratory
effects, and ~2% to nervous system effects (https://www.cdc.gov/nchs/data/nhsr/nhsr029.pdf).

32


-------
2.2.5.2.2	Identifying Suitable Studies for Use in Benefits Assessments

Relevant study information was used to identify the most nationally representative study or studies
available. Study details can be found in the associated Study Information Table. The available
cardiovascular hospital admissions studies predominantly included locations across the contiguous U.S.
and evaluated the Medicare cohort, although two studies evaluated all ages. Few studies included
health or air quality data post-2006, used hybrid exposure estimation techniques, or included 03 as a
copollutant in the risk estimates. Importantly, while all studies assessed a broad range of cardiovascular
effects, the specific ICD codes included varied widely. Of the available studies, Bell et al., 2015 evaluated
the most recent study period and included the most nationally representative study locations.

2.2.5.2.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Bell et al., 2015 investigated the effects of short-term fine particulate matter (PM2.5) exposure on
cardiovascular health (ICD-9 410, omitting 410.x2; 410-414; 426-427; 428; 429; 430-438; and 440-448).
Authors acquired data for 213 U.S. counties (1999-2010) from the Medicare Claims Inpatient Files for
U.S. residents >65 years of age. Authors chose variables including sex, age, county of residence, and
cause of hospital admission, as determined by ICD-9 codes. Authors collected PM2.5 exposure data from
county population-based ambient monitors from the US EPA Air Quality System and averaged for county
and day. Data were present for 56.5% of study days. Bell et al. (2015) utilized Bayesian hierarchal
modeling to examine the links between PM2.5 and hospital admissions, running separate models to
generate risk models for time lags (0-2 days) and season for any estimated variation in health effects.
The percent increase in risk of 0.65% (95% CI: 0.48-0.83) for an increase of 10 ng/m3 in same-day daily
mean PM2.5 concentrations came from a single pollutant model.

2.2.5.3 Cardiovascular Emergency Department Visits

The ISA found that "generally consistent, positive associations observed in numerous epidemiologic
studies of emergency department visits and hospital admissions for ischemic heart disease, heart failure,
and combined cardiovascular-related endpoints" (U.S. EPA, 2019c section 6.1.16). The ISA also calls out
cardiovascular emergency department visits as a population level health endpoint related to short-term
PM2.5 exposure (section 2.2.1.2.1.1).

2.2.5.3.1	Available Epidemiologic Literature

Although there were several studies of both emergency department visits and hospital admissions,
there was only one short-term exposure epidemiologic study specific to cardiovascular emergency
department visits available in the 2019 PM ISA (U.S. EPA, 2019c).

2.2.5.3.2	Study and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments

Ostro et al., 2016 investigated the association between short-term, source-specific (vehicular emissions,
biomass burning, soil, and secondary NCr3 and S04 sources) PM2.5 concentrations and emergency
department visits for respiratory and cardiovascular diseases in eight cities in California from 2005 to
2008. Authors obtained medical and demographic data from the Office of Statewide Health Planning
and Development in California, and diagnosis was defined with ICD-9 codes: all cardiovascular (390-459),
ischemic heart disease (410-414), AMI (410), cardiac dysrhythmia (427), and heart failure (428). Ostro et
al., 2016 conducted a case cross-over analysis, stratified by year and month, controlling for weather and
day of the week covariates. Authors used a county-level logistic regression and random-effects meta-
analysis to examine the association between source-specific PM2.5 and emergency department visits for
respiratory and cardiovascular diseases. Results indicate a positive association between vehicle PM2.5

33


-------
emissions and emergency department visits for all cardiovascular diseases. The identified excess risk
estimate of 0.7% (95% CI: -0.2-1.7) per 11.4 ng/m3 (interquartile range) daily mean PM2.5 concentration
increase came from a single pollutant model lagged by 2 days.

2.2.5.4 Cardiac Arrest (Out-of-Hospital)

The 2019 PM ISA stated that "in contrast to the studies from the previous review, recent studies have
reported generally positive associations between short-term PM2.5 exposure and out-of-hospital cardiac
arrest" (U.S. EPA, 2019c, section 6.1). The ISA also called out conductance abnormalities as a key clinical
effects associated with both short-and long-term PM2.5 exposures (section 2.2.1.2.1.1).

This endpoint, like several others (e.g., lung cancer incidence, section 2.2.5.14) has a very high rate of
fatality. As mortality due to any cause is captured separately (section 2.2.5.1), we focus on impacts
following cardiac arrest, in the population that survive the initial event when considering this health
endpoint.45

2.2.5.4.1	Available Epidemiologic Literature

The 2019 PM ISA included three epidemiologic studies of out-of-hospital cardiac arrest that met our
minimum identification criteria (section 2.1.1).

2.2.5.4.2	Identifying Suitable Studies for Use in Benefits Assessments

All three studies each evaluated separate locations and were similar with regards to study aspects such
as age range (Ensor et al., 2013, Rosenthal et al., 2008, Silverman et al., 2010). Due to differences only in
the study period and locations, the three studies are pooled using the random or fixed effects pooling
method for benefits assessment purposes.46

2.2.5.4.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Ensor et al., 2013 studied the association between short-term ambient air pollution (PM2.5 and 03)
exposure and out-of-hospital cardiac arrest. Ensor et al., 2013 gathered medical and demographic data
from an Emergency Medical Services database in Houston, Texas between 2004 and 2011. Authors
assessed the medical data and defined out-of-hospital cardiac arrest as emergency medical services
performing chest compressions. Authors collected ambient air pollution and weather data from Texas
Commission of Environmental Quality monitors and calculated hourly and daily averages for PM2.5 and
03. The authors used a time-stratified case crossover analysis and conditional logistic regression to
interpret the data and found that with a daily increase of 6 ng/m3 in PM2.5, averaged from a 0- and 1-day
lag, there was an increased risk of out-of-hospital cardiac arrest of 3.9% (95% CI: 0.5-7.4).

45	Similarly, as any emergency department visits or hospital admissions resulting from cardiac arrest would be
included in other endpoints (sections 2.2.5.2 and 2.2.5.3), monetized benefits of this health endpoint would not
include and emergency department visits or hospital admissions costs.

46	Random or fixed effects pooling is a method to combine two or more distributions into a single new distribution,
allowing for the possibilities that either 1) a single true underlying relationship exists between the component
distributions, and that differences among estimated parameters are the result of sampling error, or 2) the
estimated parameter from different studies may in fact be estimates of different parameters, rather than just
different estimates of a single underlying parameter, and weights for the pooling are generated via inverse
variance weighting, thus giving more weight to the studies that exhibit lower variance and less weight to the input
distributions with higher variance.

34


-------
Silverman et al., 2010 investigated the link between short-term ambient air pollution exposure (PM2.5,
NO2, S02, 03, and CO) and out-of-hospital cardiac arrest in New York City between 2002 and 2006.
Authors obtained medical data from the Emergency Medical Services of the New York City Fire
Department for 8,216 subjects aged 0 to 99, average age 65.6 with slightly more men than women.
Authors collected air pollution and weather data from the US EPA's Air Quality System monitors within a
20-mile radius of New York City and averaged over 24-hour periods. Authors conducted time series and
case crossover analyses with 0- and 1-day lagged air pollution levels and by season. Silverman et al.,
2010 found that in a single-pollutant case crossover model, each 10 ng/m3 increase in ambient PM2.5
resulted in a relative risk of 1.04 (95% CI: 0.99-1.08) in out-of-hospital cardiac arrest incidences 0- and 1-
day prior to onset.

Rosenthal et al., 2008 examined the effects of short-term PM2.5 exposure on out-of-hospital cardiac
arrest incidence and whether these effects were connected to demographic data or presence of heart
rhythm. Additionally, Rosenthal et al., 2008 compared exposure time and measurement method on the
effects of short-term PM2.5 exposure and out-of-hospital cardiac arrest incidence. Authors obtained
medical data from the Wishard Ambulance Service, a local emergency medical service in Indianapolis,
Indiana, from July 2, 2002 to July 7, 2006. The study defined out-of-hospital cardiac arrest using the
same criteria as Ensor et al., 2013 and Silverman et al., 2010. Authors collected daily and hourly PM2.5
concentrations from two City of Indianapolis monitoring sites and using two separate methods: the
Federal Reference Method (FRM) for 24-hour filter samples, and a Federal Equivalence Method (FEM).
The authors used a case crossover analysis with conditional logistic regressions in order to study the
effects of short-term PM2.5 exposure on out-of-hospital cardiac arrest incidence. Rosenthal et al., 2008
found a positive but statistically insignificant association between non-dead on arrival (non-DOA) out-of-
hospital cardiac arrest cases and ambient PM2.5 concentrations. Although they also noted a statistically
significant positive association when restricted to witnessed, non-DOA out-of-hospital cardiac arrest
cases, that subgroup is less applicable to the available baseline incidence rate of non-DOA out-of-
hospital cardiac arrest cases. The identified risk estimate of 1.02 (95% CI: 0.92-1.12) for each 10 ng/m3
increase in daily mean PM2.5 concentrations lagged by 0-1 days, came from a single-pollutant model of
all non-DOA out-of-hospital cardiac arrest cases.

2.2.5.5 Acute Myocardial Infarctions (AMI)

The 2019 PM ISA found that "evidence from the current review strengthens the epidemiologic results
reported in the 2009 PM ISA" with respect to AMI. Specifically:

Several new epidemiologic studies conducted in the U.S. and Europe provide additional evidence
of positive associations between short-term PM2.5 exposure and [ischemic heart disease
emergency department] visits and hospital admissions. Uncertainties noted in the last review
with respect to exposure measurement error for those not living near a PM2.5 monitor were
reduced in the current review by considering recent studies that applied hybrid exposure
assessment techniques that combine land use regression data with satellite measurements and
PM2.5 concentrations measured at fixed-site monitors to estimate PM2.5 concentrations. In
addition to these [emergency department] visit and hospital admissions studies, there is also
evidence for ST segment depression from epidemiologic panel studies. (U.S. EPA, 2019c).

The 2019 PM ISA also stated that "associations between long-term exposure to PM2.5 and cardiovascular
morbidity outcomes (i.e., ischemic heart disease, stroke) were observed in some studies with the most

35


-------
consistent results in people with pre-existing diseases" (U.S. EPA, 2019c, section 6.2). The ISA also found
that "although the results are not entirely consistent across studies or stroke subtype, some recent well-
conducted studies also support a positive association between long-term exposure to PM2.5 and stroke."
Additionally, conductance abnormalities, which can lead to cardiac arrest, were called out as a key
clinically relevant outcome associated with both short-and long-term PM2.5 exposures (section
2.2.1.2.1.1).

2.2.5.5.1	Available Epidemiologic Literature

While the 2019 PM ISA did identify epidemiological studies associating AMIs with PM2.5 exposures, the
studies that passed the initial screening stage were not more suitable than those currently used for
benefits estimation. One (Zhang et al., 2009) involved only postmenopausal women and the other
(Delfino et al., 2011) studied a population with a history of coronary artery disease. Hence, we retained
all five studies47 used in the 2012 PM NAAQS RIA. The 2019 PM ISA did not identify any newer studies of
this type so current risk estimates will continue to be used in benefit assessments (U.S. EPA, 2019c).

2.2.5.5.2	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments

As no studies new to the 2019 PM ISA provided better estimates of PM2.5 -attributable AMI risk, EPA
continues to rely upon a study by Peters et al., 2001 as the basis for the impact function estimating the
relationship between PM2.5 and nonfatal heart attacks. Peters et al., 2001 exhibits a number of
strengths. In particular, it includes a robust characterization of populations experiencing AMIs. The
researchers interviewed patients within four days of their AMI events and, for inclusion in the study,
patients were required to meet a series of criteria including minimum kinase levels, an identifiable onset
of pain or other symptoms and the ability to indicate the time, place and other characteristics of their
AMI pain in an interview.

Since the publication of Peters et al., 2001, a number of other single and multi-city studies have
appeared in the literature. These studies include Sullivan et al., 2005, which considered the risk of PM2.5-
related hospitalization for AMIs in King County, WA; Pope III et al., 2006, based in Wasatch Range, UT;
Zanobetti and Schwartz, 2006, based in Boston; and, Zanobetti et al., 2009, a multi-city study of 26 U.S.
communities. Each of these single and multi-city studies, except for Pope III et al., 2006, measure AMIs
using hospital discharge rates. Conversely, the Pope III et al., 2006 study is based on a large registry with
angiographically characterized patients—arguably a more precise indicator of AMI. Because the Pope III
et al., 2006 study reflected both myocardial infarctions and unstable angina, this produces a more
comprehensive estimate of acute ischemic heart disease events than the other studies. However, unlike
the Peters et al., 2001, Pope III et al., 2006 did not measure the time of symptom onset, and PM2.5 data
were not measured on an hourly basis.

As a means of recognizing the strengths of Peters et al., 2001 while also incorporating the more recent
evidence found in the four single and multi-city studies, we present a range of AMI estimates. The upper
end of the range is calculated using Peters et al., 2001 while the lower end of the range is the result of
an equal-weights pooling of the four newer studies (Pope III et al., 2006, Sullivan et al., 2005, Zanobetti
et al., 2009, Zanobetti and Schwartz, 2006).

47 Specific study details available in the associated Study Information Table.

36


-------
Peters et al., 2001 studied the relationship between increased particulate air pollution and onset of
heart attacks in the Boston area from 1995 to 1996. The authors used air quality data for PMio, PM10-2.5,
PM2.5, "black carbon", 03, CO, N02, and S02 in a case crossover analysis. For each subject, the case
period was matched to three control periods, each 24 hours apart. In univariate analyses, the authors
observed a positive association between heart attack occurrence and PM2.5 levels hours before and days
before onset. The authors estimated multivariate conditional logistic models including 2-hour and 24-
hour pollutant concentrations for each pollutant. They found significant and independent associations
between heart attack occurrence and 24-hour PM2.5 concentrations before onset. Significant
associations were observed for PM10 as well. None of the other particle measures or gaseous pollutants
were significantly associated with AMI for the 2-hour or 24-hour period before onset. The mean age of
participants was 62 years old, with 21% of the study population under the age of 50. In order to capture
the full magnitude of heart attack occurrence potentially associated with air pollution and because age
was not listed as an inclusion criterion for sample selection, we apply an age range of 18 and over in the
risk estimate. According to the National Hospital Discharge Survey, there were no hospitalizations for
heart attacks among children <15 years of age in 1999 and only 5.5% of all hospitalizations occurred in
those aged 15-44 years. The odds ratio is 1.62 (95% CI: 1.13-2.34) for a 20 ng/m3 increase in 24-hour
average PM2.5-

Pope III et al., 2006 evaluated the association between short-term exposure to PM2.5 and acute ischemic
heart disease events, including nonfatal AMI, all acute coronary events, and subsequent myocardial
infarctions in individuals living in greater Salt Lake City, Utah. In a case crossover study, these ischemic
events were assessed in relation to a 10 ng/m3 increase in PM2.5. The researchers determined that a 10
Hg/m3 increase in PM2.5 resulted in a 4.5% increase (95% CI: 1.1-8.0) in unstable angina and myocardial
infarction.

Sullivan et al., 2005 studied the relationship between onset time of acute myocardial infarction and the
preceding hourly PM2.5 concentrations in 5,793 confirmed cased of myocardial infarction through King
County, Washington. In this case crossover study from 1988-1994, air pollution exposure levels averaged
before onset of myocardial infarction were compared to a set of time-stratified referent exposures from
the same day of the week in the month of the case event. The authors estimated that an associated risk
of 1.01 (95% CI: 0.98-1.05) for myocardial infarction onset could be attributed to a 10 ng/m3 increase in
PM 2.5 the hour before the Ml onset. No increased risk was found in all cases with preexisting cardiac
diseases with an odds ratio of 1.05 (95% CI: 0.95-1.16). Furthermore, stratification for hypertension,
diabetes, and smoking status did not modify the association between PM2.5 and onset of myocardial
infarction.

Zanobetti et al., 2009 examined the relationship between daily PM2.5 levels and emergency hospital
admissions for cardiovascular causes, myocardial infarction, congestive heart failure, respiratory
disease, and diabetes among 26 U.S. communities from 2000-2003. The authors used meta-regression
to examine how this association was modified by season- and community-specific PM2.5 composition
while controlling for seasonal temperature as a substitute for ventilation. For a 10 ng/m3 increase in 2-
day averaged PM2.5, a 1.89% (95% CI: 1.34-2.45) increase in cardiovascular disease admissions, a 2.25%
(95% CI: 1.10-3.42) increase in myocardial infarction admissions, a 1.85% (95% CI: 1.19-2.51) increase in
congestive heart failure admissions, a 2.74% (95% CI: 1.30-4.20) increase in diabetes admissions, and a
2.07% (95% CI: 1.20-2.95) increase in respiratory admissions were observed. The relationship between
PM2.5 and cardiovascular admissions was significantly modified when the mass of PM2.5 was high in Br,
Cr, Ni, and sodium ions, while mass high in As, Cr, Mn, organic carbon, Ni and sodium ions modified the

37


-------
myocardial infarction relationship and mass high in As, organic carbon, and sulfate ions modified the
diabetes admission rates.

Zanobetti and Schwartz, 2006 analyzed hospital admissions through emergency department for Ml (ICD-
9 code 410) and pneumonia (ICD-9 codes 480-487) for associations with fine particulate air pollution in
the greater Boston area from 1995-1999. The authors used a case-crossover analysis with control days
matched on temperature. Significant associations were detected for PM2.5 with an 8.6% increase (95%
CI: 1.2-15.4) in emergency myocardial infarction hospitalizations. The study looked at hospital
admissions of AMI through the ER. Under the assumption that all heart attacks will end in
hospitalization, we consider the endpoint as heart attack events to be consistent with other studies. In a
single-pollutant model, the coefficient and standard error are estimated from the percent change in risk
(8.65%) and 95% confidence interval (95% CI: 1.22-15.38%) for a 16.32 ng/m3 increase in daily 24-hour
mean PM2.5 for an average of the 0- and 1-day lag (Zanobetti and Schwartz, 2006, Table 4).

2.2.5.6	Stroke

2.2.5.6.1	Available Epidemiologic Literature

The 2019 PM ISA included three epidemiologic studies of stroke that met the minimum identification
criteria (section 2.1.1).

2.2.5.6.2	Identifying Suitable Studies for Use in Benefits Assessments

One of the available epidemiologic studies was more recent, included a larger population, evaluated
long-term exposure effects, and was more representative of the U.S. with regards to both geography
and population attributes than the other two studies (Kloog et al., 2012).

2.2.5.6.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Kloog et al., 2012 analyzed the effects of long- and short-term PM2.5 exposure on hospital admissions
due to strokes with a new PM2.5 exposure model in New England (Connecticut, Maine, Massachusetts,
New Hampshire, Rhode Island, and Vermont) from 2000 to 2006. We use this endpoint as a surrogate
for PM2.5-attributable stroke incidence. Authors collected medical data from 67,678 adults aged 65 to 99
in the U.S. Medicare program database from 2000 to 2006. They defined all respiratory, cardiovascular
disease, stroke, and diabetes based on emergency department visits and primary discharge diagnosis
records. Authors used a hybrid exposure technique comprised of daily PM2.5 concentration data from
aerosol optical depth (AOD) measurements and ambient air monitors from the U.S. EPA and Interagency
Monitoring of Protected Visual Improvements (IMPROVE). Authors also obtained land use regressions,
meteorological data (National Climatic Data Center), and socioeconomic data (U.S. Census Bureau)
matched to zip codes. Utilizing land use Poisson regression single-pollutant models, the authors found
an 3.49% (95% CI: 0.09-5.18) increase in stroke incidence for a 10 ng/m3 increase in the 7-year mean
PM2.5 concentrations.

2.2.5.7	Respiratory Hospital Admissions

After considering the relationships between specific and broad respiratory hospital admissions
endpoints, the 2019 PM ISA stated that "recent studies further expand analyses with older adults, with
multicity studies conducted in the U.S. providing evidence of consistent, positive associations between
short-term PM2.5 exposure and respiratory-related diseases" ((U.S. EPA, 2019c, section 5.1.6).

38


-------
The ISA noted that "it is often difficult to determine whether the associations observed indicate that
PM2.5 may affect the spectrum of respiratory diseases or reflects the evidence supporting associations
with specific respiratory diseases, such as asthma." Taking this into consideration, hospital admissions
for asthma exacerbation/symptoms, COPD, and respiratory infections were specifically called out in the
short-term PM2.5 exposure biological plausibility diagram (section 2.2.1.2.1.2).

2.2.5.7.1	Available Epidemiologic Literature

Like the cardiovascular hospital admission/emergency department visit endpoints, several respiratory
studies identified by the ISA combined the hospital admissions and emergency department endpoints.
There was also a subset of studies that only considered emergency hospital admission, defined as
hospital admissions that originated in the emergency department. As using either studies of emergency
hospital admissions or combined emergency department and hospital admission studies would result in
increased uncertainty around the economic estimate and/or with the baseline incidence data, we
limited our pool of available studies to those specifically evaluating unplanned respiratory hospital
admissions, of which there were 12 available studies.

2.2.5.7.2	Identifying Suitable Studies for Use in Benefits Assessments

Studies for this endpoint tended to focus on specific age groups, with approximately half focusing on
older adults (>64) and none specifying ages 19-64. Importantly, studies varied widely by ICD codes,
making pooling of two or more studies difficult. Considering the preferred criteria, two studies were
identified, one of children and one of older adults, primarily due to the inclusion of diverse and large
study locations. The single study of older adults is more informative than pooling it with other studies of
the same population as it is more recent and includes exposure to lower PM2.5 concentration levels.

2.2.5.7.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Bell et al., 2015 investigated the effects of short-term fine particulate matter (PM2.5) exposure on
respiratory health (ICD-9 464-466, 480-487, 490-492, 493) in older adults (>64 years). Authors acquired
data for 213 U.S. counties (1999-2010) from the Medicare Claims Inpatient Files for U.S. residents >65
years of age. Authors chose variables including sex, age, county of residence, and cause of hospital
admission, as determined by ICD-9 codes. Authors collected PM2.5 exposure data from county
population-based ambient monitors from the US EPA Air Quality System and averaged for county and
day. Data were present for 56.5% of study days due to the sampling schedule of the monitors. Bell et al.,
2015 utilized Bayesian hierarchal modeling to examine the links between PM2.5 and hospital admissions.
They ran separate models for time lags (0-2 days) and season to determine if there were any estimated
variation in health effects. The identified percent increase in risk of 0.25% (95% CI: 0.01-0.48) for an
increase of 10 ng/m3 in same-day daily mean PM2.5 concentrations came from a single-pollutant model.

Ostro et al., 2009 estimated the association between ambient PM2.5, EC, organic carbon (OC), N03, and
S04 on hospital admissions for respiratory diseases in children ages 5 to 19. The study used the
California Office of Statewide Health Planning and Development, Healthcare Quality and Analysis
Division hospitalization data from six California counties for the 2000 to 2003 study period. Ostro et al.,
2009 classified hospital admissions into: all respiratory disease (ICD-9 codes 460-519), asthma (ICD-9
code 493), acute bronchitis (ICD-9 code 466), and pneumonia (ICD-9 codes 480-486). They aggregated
the hospital admission data to the county level to create a daily time series of admissions for each
county. Authors took air quality measurements from the California Air Resources Board, which captured
speciated 24-hour average pollutant measurements using a filter-based Met One Speciation Air

39


-------
Sampling System. Meteorological measurements for average daily temperature and relative humidity
came from the California Air Resources Board or the California Irrigation Management Information
System. Authors analyzed data using a Poisson regression with time, day of the week, temperature,
relative humidity, and pollutant as explanatory variables. Ostro et al., 2009 controlled for seasonality
and time dependent effects by including a natural spline smoother for the daily time trend and
meteorology. The identified percent increase in risk of excess risk of 4.1% (95% CI: 1.8-6.4) for a 14.6
Hg/m3 increase in daily mean PM2.5 concentrations, lagged by 3 days, came from a single-pollutant
model.

2.2.5.8 Respiratory Emergency Department Visits

After considering the relationships between specific and broad respiratory emergency department visit
endpoints, the 2019 PM ISA stated that "recent studies further expand analyses with older adults, with
multicity studies conducted in the U.S. providing evidence of consistent, positive associations between
short-term PM2.5 exposure and respiratory-related diseases" (U.S. EPA, 2019c, section 5.1.6).

The ISA noted that "it is often difficult to determine whether the associations observed indicate that
PM2.5 may affect the spectrum of respiratory diseases or reflects the evidence supporting associations
with specific respiratory diseases, such as asthma." Emergency department visits for asthma
exacerbation/symptoms, COPD, and respiratory infections were specifically called out in the short-term
PM2.5 exposure biological plausibility diagram (section 2.2.1.2.1.2).

2.2.5.8.1	Available Epidemiologic Literature

Like the cardiovascular hospital admission/emergency department visit endpoints, several respiratory
studies identified by the ISA combined the hospital admissions and emergency department endpoints.
As using the combined study endpoint would result in increased uncertainty around the economic
estimate, we limited our pool of available studies to those specifically looking at respiratory emergency
department visits for the main benefits assessment, of which there were 10 studies.

2.2.5.8.2	Identifying Suitable Studies for Use in Benefits Assessments

One study out of the 10 was more recent, provided greater geographic representation, and included all
ages.

2.2.5.8.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Krall et al., 2013 investigated the associations between short-term, source-specific (traffic and coal
combustion) ambient PM2.5 exposure and emergency department visits for respiratory diseases in U.S.
cities (Atlanta, GA, Birmingham, AL, St. Louis, MO, and Dallas, TX). Authors obtained medical data from
hospital electronic billings for emergency department visits due to respiratory disease, identified using
ICD-9 codes (460-465, 466, 477, 480-486, 491, 492, 493, 496, 786.07). Authors collected PM2.5
concentrations from one ambient air monitor in each of the four cities and gathered meteorological
data from the National Climactic Data Center. Krall et al., 2013 estimated source-specific PM2.5 using
apportionment models, which separate PM2.5 sources based on chemical composition. This model also
included data on gaseous pollutant concentrations from the Community Multiscale Air Quality (CMAQ)
with Tracers model. Krall et al., 2013 used Poisson time series regression models to analyze associations
between short-term PM2.5 exposure and emergency department visits for respiratory diseases. They
then compared source-specific PM2.5 exposures across cities to estimate associations with the
emergency department visit data. To limit confounders, the authors adjusted models for indicator

40


-------
variables, meteorological variables, and long-term trends in emergency department visits. The identified
relative risk estimates of 1.005 (95% CI: 1.000-1.010) for Atlanta, GA; 1.009 (95% CI: 1.003-1.015) for
Birmingham, AL; 1.008 (95% CI: 1.002-1.014) for St. Louis, MO; and 1.012 (95% CI: 1.002-1.023) for
Dallas, TX were calculated from a single-pollutant model for a 9.16 ng/m3 increase in daily mean PM2.5
concentrations, lagged by 0 days.

2.2.5.9 Asthma Onset

The 2019 PM ISA stated that "longitudinal studies provide evidence of associations with asthma
incidence in children" and found "evidence for a relationship between PM2.5 exposure and asthma
prevalence in children and childhood wheeze" (U.S. EPA, 2019c). Additionally, asthma onset was called
out as a key clinically relevant health endpoint in the biological plausibility pathways included in the ISA
(section 2.2.1.2.1.2) (U.S. EPA, 2019c).

2.2.5.9.1	Available Epidemiologic Literature

The final 2019 PM ISA found that "recent studies of asthma in children supplement the limited number
of studies reviewed in the 2009 PM ISA and provide evidence of an association between long-term PM2.5
exposure and asthma development in children" (U.S. EPA, 2019c). There was also evidence of PM2.5-
attributable asthma onset in adults, but results were inconsistent across studies. As a result, asthma
onset in adults is not included in our main benefits assessment.

Five North American epidemiologic studies of asthma onset in children were identified in section 5.2 of
the 2019 PM ISA (U.S. EPA, 2019c). Relevant information related to the identification criteria, including
study location, population attributes, and study period, were extracted from the studies and is available
in the corresponding Excel file.

2.2.5.9.2	Identifying Suitable Studies for Use in Benefits Assessments

Although the available asthma onset studies vary widely in all criteria considered, relevant study
information was again used to identify the most nationally representative study or studies (see Excel
file). Interestingly, three of the five studies were conducted in Canada.48 One study conducted in Canada
evaluated a recent and extensive time series of air quality and health data; assigned exposures to
populations using a combination of monitor and remote sensing data; validated the outcome measure;
observed effects with relatively low (~10 ng/m3) PM2.5 concentrations, and included over 30-fold the
number of participants as any other study. Other available literature was also more limited with regards
to population demographic and geographic diversity.

2.2.5.9.3	Study Identified as Most Suitable for Use in Benefits Assessments

Tetreault et al., 2016 investigated the relationship between childhood asthma onset and long-term
pollution exposure (PM2.5, N02, 03) in Quebec, Canada. The authors obtained data from four medical-

48 Although several key studies identified in the 2019 PM ISA come from Europe, we excluded studies outside of
North America. Studies taking place in Canada were retained, as there is considerable PM2.5 transport between
Canada and the US (https://www.epa.gov/airmarkets/canada-united-states-transboundary-particulate-matter-
science-assessment-2013), ~90% of Canadians live within ~100 miles of the US border

(https://www.cbc.ca/news/canada/by-the-numbers-l.801937), and ambient PM2.5 concentrations are similar in
Canada and the US. Additionally, this endpoint is more related to health physiology then healthcare systems and
Canada and the US have similar prevalence rates of asthma (https://ourworldindata.org/grapher/asthma-
prevalence).

41


-------
administrative databases collectively known as Quebec Integrated Chronic Disease Surveillance System
(QICDSS) between April 1, 1996 and March 31, 2011. The study defined the onset of asthma as a
hospital discharged diagnosis of asthma or two reports of asthma from two separate physicians within a
two-year period. The authors used Cox proportional hazard models to estimate the association between
asthma onset and pollution exposure, controlling for demographics and socioeconomic status. Time-
varying exposure models assessed time-varying exposures to the three pollutants in question. Tetreault
et al., 2016 showed that childhood asthma onset may be associated with exposure to PM2.5, N02, and
03.

The identified study presented 24 hazard ratios using various adjustment methods and included multiple
sensitivity analyses, evaluating the effects by sex, urbanicity, and those who moved during the study
period. More-adjusted risk estimates using a time-varying estimate of PM2.5 exposure and including the
full cohort were identified over less-adjusted or striated estimates using exposure estimates at birth.
The study identified as best characterizing risk across the U.S. was Tetreault et al., 2016, although the
two older and demographic-limited U.S. studies were also identified as potentially informative.

The risk estimate identified from Tetreault et al., 2016 for use in the main benefits estimates was a
single-pollutant time-varying hazard ratio model of 1.33 (95% CI: 1.31-1.34) for a 6.53 ng/m3
(interquartile range) increase in annual PM2.5 concentration at the residential address.

As the physiology and etiology of lung development in children is similar in children 6-17, we apply the
4-12 year age-striated effect estimate from Tetreault et al., 2016 to children ages 4-17 (Baena-Cagnani
et al., 2007, Guerra et al., 2004, Ochs et al., 2004, Sparrow et al., 1991, Trivedi and Denton, 2019).

2.2.5.10 Asthma Symptoms/Exacerbation

The 2019 PM ISA stated that "evidence for effects on asthma exacerbation are generally more
consistent than associations for other respiratory outcomes." The ISA went on to note that "recent
studies strengthen the relationship between asthma exacerbation in children and short-term PM2.5
exposure, while, in adults, the relationship continues to be inconsistent."

2.2.5.10.1	Available Epidemiologic Literature

Based on evidence provided by the 2019 PM ISA, available studies of asthma symptoms were limited to
children, of which there were eight.

2.2.5.10.2	Identifying Suitable Studies for Use in Benefits Assessments

Due to the specificity required when evaluating this health endpoint, Individual studies of asthma
exacerbation/symptoms tended to focus on relatively small cohorts of children of discrete ages in
distinct locations, making pooling difficult. One study evaluated a directly monetizable outcome of
albuterol inhaler use. Albuterol inhalers are separated from long-term asthma control medications and
is considered a "rescue medication" by the Mayo Clinic, making it an informative endpoint when
considering asthma symptoms.49

2.2.5.10.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Rabinovitch et al., 2006 analyzed the relationship between short-term PM2.5 exposure and asthma
exacerbation in children. The study followed children, ages 6 to 13 attending the Kunsberg School at the

49 https://www.mayoclinic.org/diseases-conditions/asthma/in-depth/asthma-medications/art-20045557

42


-------
National Jewish Medical Research Center with diagnosed asthma for two consecutive winters from
2001-2003. Authors gave an electronic bronchodilator (albuterol) to the children to capture the
frequency of use within a 24-hour period. The children also responded to three questions to determine
if they may have an upper respiratory infection (URI), and urine samples were taken to measure urinary
leukotriene E4 levels on select days. The authors collected hourly ambient PM2.5 levels from the Colorado
Department of Health Air Pollution Control Division's Tapered Element Oscillating Microbalance (TEOM)
monitor, located 2.7 miles west of the school. Additionally, a Federal Reference Monitor (FRM) located
next to the TEOM measured 24-hour PM2.5 levels. The authors obtained meteorological data from the
Colorado Department of Health Air Pollution Control Division and the National Climatic Data Center. A
Poisson regression modeled albuterol use as a function of the morning (12:00am to 11:00 am) maximum
hourly PM2.5 level or the morning mean hourly PM2.5 level. The model used both the TEOM and FRM
data, individually, incorporated four lag periods (0 to 2 days and 0- to 2-day average), and included
several covariates: temperature, pressure, humidity, time trend, Friday indicator, and URI indicator.
Rabinovitch et al., 2006 found that, although the PM2.5 pollution levels were well below the National
Ambient Air Quality Standards, there is a consistent association between peak ambient PM2.5 levels and
increased albuterol use in asthmatic children. The identified percent of use increase estimate of 1.2%
(95% CI: -0.6-2.9) for a 6 ng/m3 increase in averaged daily mean PM2.5 concentration lagged by 0-, 1-,
and 2-days came from a single-pollutant model.

2.2.5.11 Allergic Rhinitis (Hay Fever/Respiratory Allergies))

The 2019 PM2.5 ISA stated that "recent studies evaluated associations between long-term exposure to
PM2.5 and various allergic outcomes in a mix of large representative cohort and cross-sectional survey
studies" finding "generally consistent evidence of an association between long-term PM2.5 exposure and
allergic sensitization in single pollutant models" ((U.S. EPA, 2020a, section 5.2.4). Additionally, the ISA
called out "allergic responses" in the biological plausibility diagram for long-term PM2.5-attributable
respiratory effects (U.S. EPA, 2020a, section 2.2.1.2.1.2). Although cross sectional analyses do not
establish a temporal sequence, they can be used to estimate benefits associated with changes in air
quality.

2.2.5.11.1	Available Epidemiologic Literature

The 2019 PM ISA identified one epidemiologic study of long-term 2019 PM2.5 exposure and allergic
rhinitis.

2.2.5.11.2	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Parker et al., 2009 investigated the associations between long-term PM2.5 exposure and respiratory
allergies in an unrestricted population of children (aged 3-17 years) sampled from the United States
National Health Interview Survey. Authors obtained symptom data from participant parents, who
reported respiratory allergies on annual surveys. Parker et al., 2009 placed all study participants
reporting symptoms of respiratory allergies or hay fever into a combined rhinitis group. Parker et al.,
2009 then linked annual averages of S02, N02, PM2.5, and PM2.5-10 and warm season (May to September)
03 averages to participant's addresses through ambient air pollution and meteorological data (03, S02,
N02, PM2.5, and PM10-2.5) collected from US EPA Air Quality System monitors. The authors adjusted
models for survey year, poverty-level, race/ethnicity, age, family structure, insurance coverage, usual
source of care, education of adult, urban-rural status, region, and median county-level income. Through

43


-------
multi-pollutant, logistic regression models, an odds ratio of 1.29 (95% CI: 1.07-1.56) for a 10 ng/m3
increase in PM2.5 concentrations and respiratory allergies was identified.

2.2.5.12	Minor Restricted Activity Days

No new epidemiologic studies of minor restricted activity days (MRADs) were identified in the 2019 PM
ISA (U.S. EPA, 2019c).

2.2.5.12.1 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Ostro and Rothschild, 1989 estimated the impact of PM2.5 and 03 on the incidence of minor restricted
activity days (MRADs) and respiratory-related restricted activity days (RRADs) in a national sample of the
adult working population, ages 18 to 65, living in metropolitan areas. The study population is based on
the Health Interview Survey (HIS), conducted by the National Center for Health Statistics. In publications
from this ongoing survey, non-elderly adult populations are generally reported as ages 18-64. From the
study, it is not clear if the age range stops at or includes those aged 65. We apply the risk estimate
function to individuals ages 18-64 for consistency with other studies estimating impacts to non-elderly
adult populations. The annual national survey results used in this analysis were conducted in the period
1976-1981, controlling for PM2.5, two-week average 03.

2.2.5.13	Work Loss Days

No new studies of work loss days (WLDs) were identified in the 2019 PM ISA (U.S. EPA, 2019c).

2.2.5.13.1 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Ostro, 1987 estimated the impact of PM2.5 on the incidence of work-loss days (WLDs), restricted activity
days (RADs), and respiratory-related RADs (RRADs) in a national sample of the adult working population,
ages 18 to 65, living in metropolitan areas. The study population is based on the Health Interview Survey
(HIS), conducted by the National Center for Health Statistics. The annual national survey results used in
this analysis were conducted in 1976-1981. Ostro, 1987 reported that two-week average PM2.5 levels
were significantly linked to work-loss days, RADs, and RRADs, however there was some year-to-year
variability in the results. Separate coefficients were developed for each year in the analysis (1976-1981);
these coefficients were pooled. The coefficient used in the concentration-response function presented
here is a weighted average of the coefficients in Ostro, 1987, Table III, using the inverse of the variance
as the weight.

2.2.5.14	Lung Cancer

The 2019 PM ISA determined that a "likely to be causal" relationship exists between long-term PM2.5
exposure and cancer (U.S. EPA, 2019c), a change in the causality determination from the 2009 ISA (U.S.
EPA, 2009). Specifically, the ISA found evidence of generally consistent positive associations between
long-term PM2.5 exposure and lung cancer incidence.50 Additional details regarding potential pathways
of disease development are summarized in the biological plausibility diagram provided by the ISA
(section 2.2.1.2.1.3).

For an outcome such as lung cancer, there is an expected time lag between changes in pollutant
exposure in a given year and the reduction in lung cancer incidence, known as the latency period. The

50 The ISA also found generally consistent positive associations between long-term PM2.5 exposure and lung cancer
mortality, but as mortality impacts are included elsewhere (section 2.2.5.1), this endpoint focuses on non-fatal
lung cancer incidence.

44


-------
time between exposure and diagnosis can be quite long, on the order of years to decades. We discuss
methods used to account for the latency period and other economic considerations relevant to this
health endpoint in section 5.3.6. Importantly, we include this health endpoint to assess impacts of living
with a lung cancer diagnosis, prior to disease resolution or death.

2.2.5.14.1	Available Epidemiologic Literature

We limited our pool of available studies to those of lung cancer incidence, excluding those assessing
lung cancer mortality as that endpoint is included in the long-term exposure-attributable all-cause
mortality endpoint (section 2.2.5.1.3.1). This resulted in four study options.

2.2.5.14.2	Identifying Suitable Studies for Use in Benefits Assessments

The four available studies varied in terms of population demographics included and country. Two of the
four studies took place entirely in Canada. Of the two U.S.-based studies, one included all ages, sexes,
and demographics and was restricted to non- and never-smokers, although it included some participants
living in Canada. The identified study was most suitable as it took place in the U.S., included both males
and females, and was restricted to non- and never-smokers (Gharibvand et al., 2017).

2.2.5.14.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Gharibvand et al., 2017 evaluated whether positive associations exist between PM2.5 exposure and
incidence of lung cancer in non-smokers among the Adventist Health and Smog Study-2 (AHSMOG-2), a
group of health-conscious individuals of which 81% are never smokers. Authors collected ambient air
pollution data (PM2.5 and 03) from the US EPA Air Quality system over two years (January 2000-
December 2001). Three a priori factors were added to the models as covariates: time spent outdoors,
residence length, and moving distance during follow-up. Authors modeled the association between
PM2.5 exposure and incidence of lung cancer using a Cox proportional hazards regression, with attained
age as the time variable. The authors conducted both a single and a two-pollutant (PM2.5 and 03)
analyses. The study concluded that each 10 ng/m3 increase in ambient PM2.5 concentrations was
positively associated with increased lung cancer risks within the single-pollutant and two-pollutant
multivariable models with 03. The identified hazard ratio of 1.46 (95% CI: 1.13-1.89) for each 10 ng/m3
increase in mean monthly ambient PM2.5 concentrations came from a two-pollutant multivariable model
with 03 (including a priori covariates).

2.2.5.15 Alzheimer's Disease

Evidence connecting long-term PM2.5 exposure to nervous system effects led to the 2019 ISA concluding
a "likely to be causal" relationship exists (U.S. EPA, 2019c) and various clinically relevant nervous system
endpoints were called out in the biological plausibility section, including Alzheimer's disease, Parkinson's
disease, autism spectral disorder, cognitive decline, and dementia (section 2.2.1.2.1.3). Regarding
biological plausibility, the ISA stated that "neuroinflammation and neurodegeneration provide biological
plausibility for epidemiologic results of increased hospital admissions or emergency department visits
for Alzheimer's and Parkinson's disease."

There were over a dozen epidemiologic studies in the 2019 PM ISA evaluating cognitive-related
outcomes (U.S. EPA, 2019c, sections 8.2.5-8.2.7). However, due to the nature of the endpoint, many of
the outcomes were defined using scales and scores from cognitive tests. As we are currently unable to
transfer that type of result into a clinically relevant population level outcome, we focused on the more
clearly defined outcomes of Alzheimer's disease and Parkinson's Disease. These endpoints were also

45


-------
specifically called out in the ISA, as "epidemiologic studies also provide evidence of cognitive
impairment and Alzheimer's and Parkinson's disease in association with exposure to PM2.5" ((U.S. EPA,
2019c, section 8.2.6).

2.2.5.15.1	Available Epidemiologic Literature

One epidemiologic study of Alzheimer's disease met our minimum required identification criteria
(section 2.1.1).

2.2.5.15.2	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Kioumourtzoglou et al., 2016 evaluated the potential impact of long-term PM2.5 exposure on first
hospital admission for dementia, Alzheimer's, or Parkinson's diseases among Medicare beneficiaries (>=
65 years old) in 50 cities in the northeastern U.S. (Connecticut, Delaware, Maine, Maryland,
Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island, Vermont, and
Washington, D.C.). Authors retrieved medical data from the Center for Medicaid and Medicare from the
years 1999-2010. The study followed enrollees as a cohort, which included annual follow-up records
identifying the first hospital admissions for dementia (ICD-9 290), Alzheimer's (ICD-9 331.0), Parkinson's
(ICD-9 332), and other cardiovascular comorbidities. With respect to Alzheimer's disease, the study
evaluated 9,817,806 Medicare enrollees and included 266,725 cause-specific hospital admissions
indicating disease onset. Annual average PM2.5 concentrations were estimated for each city using data
from the U.S. EPA Air Quality System database. Kioumourtzoglou et al., 2016 fit a time-varying Cox
proportional hazards model for each city, using the city-wide annual PM2.5 concentrations as the time-
varying exposure of interest and a linear term for the calendar year. This eliminated the impact of PM2.5
variation by city and any PM2.5 trends within cities. The model adjusted for cardiovascular comorbidities,
and incorporated a counting process extension which created an observation for each year of follow-up
per person. The results were then pooled across individuals and cities. A single-pollutant model was
used to develop the identified hazard ratio of 1.15 (95% CI: 1.11-1.19) for a 1 ng/m3 increase in the
average annual PM2.5 concentrations.

2.2.5.16 Parkinson's Disease

Evidence connecting long-term PM2.5 exposure to nervous system effects led to the 2019 ISA concluding
a "likely to be causal" relation exists (U.S. EPA, 2019c) and various clinically relevant nervous system
endpoints were called out in the biological plausibility section, including Alzheimer's disease, Parkinson's
disease, autism spectral disorder, cognitive decline, and dementia (section 2.2.1.2.1.3). Regarding
biological plausibility, the ISA stated that "neuroinflammation and neurodegeneration provide biological
plausibility for epidemiologic results of increased hospital admissions or emergency department visits
for Alzheimer's and Parkinson's disease."

There were over a dozen epidemiologic studies in the 2019 PM ISA evaluating cognitive-related
outcomes (U.S. EPA, 2019c, sections 8.2.5-8.2.7). However, due to the nature of the endpoint, many of
the outcomes were defined using scales and scores from cognitive tests. As we are currently unable to
transfer that type of result into a clinically relevant disease incidence, we focused on the more clearly
defined outcomes of Alzheimer's disease and Parkinson's Disease. These endpoints were also specifically
called out in the ISA, as "epidemiologic studies also provide evidence of cognitive impairment and
Alzheimer's and Parkinson's disease in association with exposure to PM2.5" ((U.S. EPA, 2019c, section
8.2.6).

46


-------
2.2.5.16.1	Available Epidemiologic Literature

Three epidemiologic studies of Parkinson's disease were identified in the 2019 PM ISA. All evaluated
relatively low PM2.5 concentrations and included participants from multiple states, however there were
differences with respect to the ages and sexes evaluated, number of overall participants, and exposure
estimation technique.

2.2.5.16.2	Identifying Suitable Studies for Use in Benefits Assessments

The prospective study with the lowest mean PM2.5 concentrations and most recent timespan included
over 14 times the number of participants as the other two studies combined. It was also the only study
to include participants over the age of 71, which is relevant as Parkinson's disease prevalence rises with
age (Pringsheim et al., 2014).

2.2.5.16.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Kioumourtzoglou et al., 2016 evaluated the potential impact of long-term PM2.5 exposure on first
hospital admission for dementia, Alzheimer's, or Parkinson's diseases among Medicare beneficiaries (>=
65 years old) in 50 cities in the northeastern U.S. (Connecticut, Delaware, Maine, Maryland,
Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island, Vermont, and
Washington, D.C.). Authors retrieved medical data from the Center for Medicaid and Medicare from the
years 1999-2010. The study followed enrollees as a cohort, which included annual follow-up records
identifying the first hospital admissions for dementia (ICD-9 290), Alzheimer's (ICD-9 331.0), Parkinson's
(ICD-9 332), and other cardiovascular comorbidities. With respect to Parkinson's disease, the study
evaluated 9,817,806 Medicare enrollees and included 119,425 cause-specific hospital admissions
indicating disease onset. Annual average PM2.5 concentrations were estimated for each city using data
from the US EPA Air Quality System database. Kioumourtzoglou et al., 2016 fit a time-varying Cox
proportional hazards model for each city, using the city-wide annual PM2.5 concentrations as the time-
varying exposure of interest and a linear term for the calendar year. This eliminated the impact of PM2.5
variation by city and any PM2.5 trends within cities. The model adjusted for cardiovascular comorbidities,
and incorporated a counting process extension which created an observation for each year of follow-up
per person. The results were then pooled across individuals and cities. A single-pollutant model was
used to develop the identified hazard ratio of 1.08 (1.04 - 1.12) for a 1 ng/m3 increase in the average
annual PM2.5 concentrations.

2.2.6 O3

The following sections of the 2020 03 ISA correspond to health endpoints judged as having a "causal" or
"likely causal" relationship with 03 exposure:

•	Appendix 3: Health Effects - Respiratory, 3.1 Short-Term Ozone Exposure;

•	Appendix 3: Health Effects - Respiratory, 3.2 Long-Term Ozone Exposure;

•	Appendix 5: Health Effects - Metabolic Effects, 5.1 Short-Term Ozone Exposure;

•	Appendix 6: Health Effects - Mortality, 6.1 Short-Term Ozone Exposure and Mortality; and

•	Appendix 6: Health Effects - Mortality, 6.2 Long-Term Ozone Exposure and Mortality.

Following the approach to identifying available epidemiologic literature (section 2.2), we began with the
1,678 studies cited by the 2020 03 ISA. Of these, 130 morbidity studies evaluate health endpoints the
2020 03 ISA determined as having a "causal" or "likely causal" relationship with 03 exposure (sections
2.2.1 and 2.2.2). 77 studies remained after the required minimum criteria were applied, and that

47


-------
number decreased to 27 when broad hospital admissions and emergency department endpoints were
identified.51 No studies of short-term 03 exposure metabolic effects meeting the minimum required
criteria (section 2.1.1) were identified in the ISA.

Table 7. 03 Study and Risk Estimate Identification Diagram

03 Endpoint and
Exposure Duration

Studies
Available

Studies
Included

Ages

Risk
Estimates
Available

Risk
Estimates
Included

Respiratory Mortality
(LT)

4

1

Adults and older adults

2

1

Respiratory Mortality
(ST)

6

2

Children, adults, and
older adults

16

1

120

1

Respiratory Hospital
Admissions (ST)

3

1

Older adults

7

1

Respiratory Emergency
Department Visits (ST)

7

1

Children, adults, and
older adults

45

5

Asthma Onset (LT)

4

1

Children

8

1

Asthma Symptoms (ST)

4

1

Children

160

1

Minor Restricted
Activity Days

NA

1

Adults

NA

1

Allergic Rhinitis

1

1

Children

5

1

School Loss Days

NA

2

Children

NA

1

NA

1

NA- not applicable due to the absence of additional ISA evidence. Risk estimates identified in the 2015 Ozone NAAQS RIA will
continue to be utilized.

2.2.6.1 Respiratory Mortality

We separate respiratory mortality impacts resulting from short- and long-term exposures for several
reasons. Firstly, the biological pathways of short- and long-term 03-attributable health effects may differ
in ways that affect the manner in which this endpoint is quantified (section 2.2.1.2.2.1). For example,
some impacts of long-term exposure to 03 may be incremental to impacts attributable to short-term
exposure. Conversely, some impacts associated with long-term exposure to 03 may include impacts
attributable to short-term exposure. However, we lack the evidence to determine the extent to which
these risks are mutually exclusive or overlapping. Secondly, the level of support for respiratory mortality
effects of short- and long-term 03 exposures may differ. Therefore, we continue to include risk
estimates of respiratory mortality from both short- and long-term exposures to present a range of
health impact estimates.

2.2.6.1.1 Respiratory Mortality Attributable to Short-Term Exposures

The 2020 03 ISA determined that there exists a "causal" relationship between short-term 03 exposure
and respiratory outcomes (U.S. EPA, 2020a). The short-term exposure causality determination "was
made on the basis of a strong body of evidence integrated across controlled human exposure, animal
toxicological, and epidemiologic studies, in addition to established findings from previous [Air Quality

51 This number may not be equal to the sum of available studies in Table 7 as individual studies may present risk
estimates for multiple health endpoints.

48


-------
Criteria Documents], demonstrating respiratory effects due to short-term exposure to ozone." While the
ISA found that "recent epidemiological evidence for respiratory mortality is limited, but there remains
evidence of consistent, positive associations, specifically in the summer months" and "when recent
evidence is considered in the context of the larger number of studies evaluated in the 2013 Ozone ISA,
there remains consistent evidence of an association between short-term ozone exposure and
respiratory mortality." Due to the strength of the ISA evidence relating short-term exposures to
respiratory mortality, estimates of respiratory mortality impacts are included in the main benefits
assessment of 03-attributable health impacts.

Separately, 2020 ISA determined that the relationship between short-term 03 exposure and total
mortality is "suggestive of, but not sufficient to infer, a causal relationship." By comparison, the 2013 ISA
identified this endpoint as "likely to be causal." Evidence supporting a relationship between short-term
03 exposure and total mortality included consistent epidemiologic evidence from multiple high-quality
studies at relevant ozone concentrations, some support for an independent 03 association, and
biological plausibility from studies of respiratory morbidity. In contrast, uncertainties remain regarding
geographic heterogeneity in 03 mortality associations and there is limited biological plausibility from
studies of cardiovascular morbidity. Regarding the biological plausibility of cardiovascular effects, while
animal toxicological studies provide evidence of cardiovascular effects, recent controlled human
exposure studies do not provide evidence to support potential biological pathways. Additionally, there is
a lack of coherence with epidemiologic studies of cardiovascular morbidity, specifically, cardiovascular-
related emergency department visits and hospital admissions, to support cardiovascular mortality. Due
to limitations in ISA evidence relating short-term exposures to total mortality, estimates of all-cause
mortality impacts will not be calculated when estimating benefits attributable to changes in 03
exposure.

2.2.6.1.1.1	Available Epidemiologic Literature

There were six North American studies of short-term 03-attributable respiratory mortality identified in
the 2020 03 ISA, one of which was new to this review but took place entirely in Canada. Of the U.S.-
based studies, two were single-city. The other three studies were fairly equally geographically and
demographically representative, although one was a meta-analysis.

2.2.6.1.1.2	Identifying Suitable Studies for Use in Benefits Assessments

Of the six studies of short-term 03-attributable respiratory mortality, all but one study period ended
during or before the year 2000 and the individual study that extended into the 2000s was geographically
limited to a single city. This list also included a meta-analysis and a study that took place entirely in
Canada. Two U.S.-based, nationally representative studies were identified as best characterizing risk
across the U.S. for this endpoint.

2.2.6.1.1.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments

As Zanobetti and Schwartz, 2008 investigated the effects of short-term 03 exposure on mortality (all-
cause, cardiovascular, stroke, and respiratory) in an unrestricted population of children, adults, and
older adults (aged 0-99 years), it remained the superior analysis of short-term 03-attributable
respiratory mortality. Between 1998 and 2000, the authors collected mortality data from the National
Center for Health Statistic in 48 cities across the United States. Along with eight-hour ozone
concentrations and meteorological data obtained from US EPA's Air Quality System Technology Transfer
Network, the authors utilized a generalized linear model with quasi Poisson link functions to estimate

49


-------
the effects of short-term ozone on respiratory mortality. The model adjusted for season, day of the
week, and temperature. Since ozone concentrations vary between seasons, the authors decided to
restrict their analysis to ozone warm season (June - August). The identified single-pollutant, warm
season excess risk estimate of 0.83% (95% CI: 0.38-1.28%) for an increase of 10 ppb in DA8 03
concentrations over a summed lag structure of zero to three days.

Katsouyanni et al., 2009 also used time series methods to examine the relationship between short-term
03 exposures and mortality across the U.S for all ages. The study utilized mortality data from the
National Center for Health Statistics (www.cdc.gov/nchs) for years 1987 through 1996, excluding
accidental deaths (i.e., International Classification of Diseases (ICD]-9 800). 90 U.S. cities with population
sizes varying from about 250,000 to above 9 million with the largest populations were included. Daily
number of deaths ranged from 5 to 198. All 90 cities had daily summer 03 measurements. Investigators
conducted extensive simulation studies to test 1) the choice of the smoothing method and basic
functions used to estimate the smooth function of time in the city-specific models, and 2) the number of
degrees of freedom to be used in the smooth function of time. The investigators also evaluated whether
each city should be assigned the same model specification or whether each city-specific model should
depend on city-specific characteristics. For the former, the same degrees of freedom (ranging from 1 to
20 df/year of data) were assigned to the smooth function of time for every city. The range was
determined by choosing the minimum possible degrees of freedom per year up to a maximum degrees
of freedom per year that essentially removed all variation in the data beyond time scales of one week.
Also, the collective experience of the investigators indicated that using more than 20 df/year does not
substantially affect the risk estimates. For the latter approach, the degrees of freedom for the smooth
function of time were chosen separately for each city using a fit criterion, such as the Akaike Information
Criterion (AIC), or by minimizing the partial autocorrelation function (PACF) of the residuals.
Nonparametric methods underestimated the standard error of the air pollution regression coefficient,
penalized splines gave relatively small bias, and PACF in combination with penalized splines performed
relatively well in terms of bias. Therefore, the identified risk estimate was a summer-only penalized
spline estimate of respiratory mortality of 0.73 (-0.39, 1.85) per 10 ng/m3 increase in 03 from distributed
lag days was identified.

The two risk estimates identified are not directly comparable to previous estimates of short-term
exposure-related mortality as previous estimates were of nonaccidental mortality and current estimates
are of respiratory mortality.

2.2.6.1.2 Respiratory Mortality Attributable to Long-Term Exposures

The 2020 03 ISA determined that there exists a "likely to be causal" relationship between long-term 03
exposure and respiratory outcomes ({U.S. EPA, 2020 #343. The overall "likely to be causal"
determination for long-term exposures "was based on epidemiologic evidence of associations between
long-term ozone exposure and asthma development, respiratory symptoms in children with asthma, and
respiratory mortality." More specifically, the ISA found that "there is strong coherence between animal
toxicological studies of changes in lung morphology and epidemiologic studies reporting positive
associations between long-term ozone exposure and new onset asthma, respiratory symptoms in
children with asthma, and respiratory mortality" and the "several multicity studies and a multi-continent
study reported associations between short-term increases in ozone concentrations and increases in
respiratory mortality." Overall, the 2020 03 ISA concluded there was "some evidence that long-term

50


-------
ozone exposure is associated with respiratory mortality, but the evidence is not consistent across
studies." Due to the strength of the ISA evidence relating long-term exposures to respiratory mortality,
estimates of respiratory mortality impacts are included when estimating benefits attributable to
changes in 03 exposure.

2.2.6.1.2.1	Available Epidemiologic Literature

There were four North American studies of long-term 03-attributable respiratory mortality identified in
the 2020 03 ISA, three of which were new to this review.52 All four studies evaluated either the ACS CSP-
II or the Canadian Census Health and Environment Cohort (CanCHEC) prospective cohorts, differing in
study size, timespan, exposure estimation technique, and specific risk models analyzed.

2.2.6.1.2.2	Identifying Suitable Studies for Use in Benefits Assessments

Three of the four studies evaluating long-term 03-attributable respiratory mortality assessed the ACS
CSP-II prospective cohort and the fourth evaluated the Canadian Census Health and Environment Cohort
(CanCHEC) prospective cohort. Two of the three ACS CSP-II analyses were nationwide, with the third
focusing on California. One of the two nationwide ACS CSP-II analyses included a longer and more recent
study period, utilized hybrid exposure estimates, and included a larger number of participants.

2.2.6.1.2.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
{Turner, 2016 #544} examined the relationship between long-term 03 exposure (1982-2004) and
mortality (all-cause, cause-specific) in American Cancer Society Cancer Prevention Study-ll participants
(aged 30-99 years). A hierarchal Bayesian space-time model based on National Air Monitoring Stations,
State and Local Air Monitoring Stations, and Community Multi-Scale Air Quality model data estimated
daily eight-hour maximum ozone concentrations at the participant's address. The models considered
meteorological data and levels of other ambient pollutants (PM2.5, both regional and near-source, and
N02). Turner et al., 2016 utilized Cox proportional hazard models adjusted a priori for individual, socio-
demographic, and ecological variables. The hazard ratio of 1.12 (1.08 - 1.16) from a multi-pollutant, all-
year model of respiratory mortality for a 10 ppb increase in the annual average of daily 8-hour maximum
ozone concentrations was likely the most comprehensive risk estimate. This study also provided a warm-
season specific hazard ratio of 1.08 (1.06-1.11) per 10 ppb increase in seasonal average of daily 8-hour
maximum 03 concentrations, which will be used when air quality surfaces are only available for the
summer season. Notably, the study compared annual mortality with warm-season 03 exposures, so full-
year baseline incidence rates will be used with risk estimates from this study.

The identified risk estimate of long-term exposure associated mortality is larger than the risk estimate
used in previous benefits assessments (Jerrett et al., 2009), but differs in many aspects including study
size, included study locations, and exposure estimation technique.

2.2.6.2 Respiratory Hospital Admissions

After considering the relationships between specific and broad respiratory hospital admission endpoints,
the 2020 03 ISA stated that "studies conducted in diverse locations with a variety of exposure
assignment techniques continue to provide evidence of an association between ozone and both hospital

52The 2020 O3 ISA identified five North American studies of long-term 03-attributable respiratory mortality, but as
Weichenthal et al., 2016 examined the combined oxidant capacity of O3 and NO2, and not direct effects of O3
alone, it did not meet required minimum criteria for consideration for inclusion in benefits assessments (section
2.1.1).

51


-------
admissions and emergency department visits for combined respiratory diseases" (U.S. EPA, 2020a,
section 3.1.8).

2.2.6.2.1	Available Epidemiologic Literature

Three epidemiologic studies of respiratory hospital admission were identified in the 2020 03 ISA, which
varied considerably with respect to the timespans evaluated and study population locations (U.S. EPA,
2020a).

2.2.6.2.2	Identifying Suitable Studies for Use in Benefits Assessments

The two older studies either included only Canadian participants or included U.S., Canadian, and some
European participants. Therefore, we identified the most recent and only entirely U.S.-based study as
best characterizing risk across the U.S.

2.2.6.2.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Katsouyanni et al., 2009 used time series methods to examine the relationship between daily 03
concentrations and hospital admissions in North America. For U.S. benefits estimation purposes, we
focus on analyses performed using the U.S hospital admission datasets. These datasets included 14 cities
with populations between 291,000 and 5,377,000 between 1987-1996 with city-wide MDA1 03
concentrations ranging from ~34-60 ng/m3. The authors used a first stage analysis protocol that used
generalized linear models with either penalized or natural splines to adjust for seasonality, with varying
degrees of freedom. The number of degrees of freedom were also chosen by minimizing the partial
autocorrelation function of the model's residuals. Model specification approach accounted for seasonal
patterns, weekend and vacation effects, and epistemics of respiratory disease. Data were also analyzed
to detect potential thresholds in the concentration-response relationships. The second stage analysis
used pooling approaches and assessed potential effect modification by sociodemographic characteristic
and indicators of the pollution mixture across study regions. The identified percent change in respiratory
disease admission for those aged >64 was from a copollutant model including PMio is 0.28 (-0.07, 0.62)
per 10 ng/m3 increase in 03.

2.2.6.3 Respiratory Emergency Department Visits

After considering the relationships between specific and broad respiratory emergency department visit
endpoints, the 2020 03 ISA stated that "studies conducted in diverse locations with a variety of exposure
assignment techniques continue to provide evidence of an association between ozone and both hospital
admissions and emergency department visits for combined respiratory diseases" and "there is some
evidence, previously characterized in the 2013 03 ISA, that daily 8 hour max, 1 hour max, and daytime
average 03 concentrations may be most strongly associated with respiratory emergency department
visits" (U.S. EPA, 2020a, section 3.1.8).

2.2.6.3.1 Available Epidemiologic Literature

Seven U.S.-based studies of respiratory emergency department visits were identified in the 2020 03 ISA
(U.S. EPA, 2020a). As is common with hospital admission and emergency department health endpoints,
the specific ICD codes varied across all studies, making pooling difficult. Most studies evaluated only a
single city or state and took place in a similar time period, including the early 2000s.

52


-------
2.2.6.3.2	Identifying Suitable Studies for Use in Benefits Assessments

Available studies varied most widely by geographic area, exposure estimation method, population age
range, and 03 season. While most studies focused on a specific city, state or region, one study included
five different multi-county areas. In addition, it included a recent time period, all ages, current 03
concentrations, and was one of only two studies based on hybrid exposure techniques.

2.2.6.3.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Barry et al. (2018) investigated the effects of short-term ozone exposure on emergency department
visits for respiratory disease (ICD-9 493, 786.07, 460-466, 477, 491, 492, 496, 480-486, 466.1, 466.11,
466.19) in an unrestricted population of children, adults, and older adults (aged zero-99 years) within
five cities (Atlanta, GA, Birmingham, AL, Dallas, TX, Pittsburgh, PA, and St. Louis, MO-IL) across the
United States. Authors obtained individual-level health data from hospitals and hospital associations in
each of the five cities. Models fusing air quality monitor data with Community Multi-Scale Air Quality
modeled data at 12 x 12-km grids were used to estimate ozone exposure. Barry et al. (2018) assessed
associations with short-term ozone exposure with daily eight-hour maximum ozone concentrations. The
authors implemented Poisson log-linear models to estimate risk values with three day moving averages.
They identified single-pollutant rate ratios of 1.03 (95% CI: 1.01-1.05) in Atlanta, GA, 1.03 (95% CI: 1.00-
1.06) in Birmingham, AL, 1.05 (95% CI: 1.02-1.07) in Dallas TX, 1.03 (95% CI: 1.01-1.05) in Pittsburgh, PA,
and 1.02 (95% CI: 1.01-1.04) in St. Louis, MO-IL for an increase of 25 ppb in full-year MDA8 03
concentrations (three day moving average). Results from individual cities are pooled.

2.2.6.4 Asthma Onset

The 2020 03 ISA concluded that "recent epidemiologic studies provide generally consistent evidence for
associations of long-term ozone exposure with the development of asthma in children" (U.S. EPA,
2020a, section IS.4.3). The ISA also found that "recent animal toxicological studies demonstrate effects
on airway development in rodents...and build on and expand the evidence for long-term ozone
exposure-induced effects that may lead to asthma development" and asthma onset was called out as a
key population level clinically relevant health endpoint in the biological plausibility pathways (section
2.2.1.2.2.1, Figure 9). More specifically, the 03 ISA stated that a "limited number of recent epidemiologic
studies provide generally consistent evidence that long-term ozone exposure is associated with the
development of asthma in children" (U.S. EPA, 2020a, section 3.2.6).

2.2.6.4.1	Available Epidemiologic Literature

The 2020 03 ISA identified children as the population in which this health effect was observed, so we
began with the four ISA-identified studies of people <21 (U.S. EPA, 2020a).

2.2.6.4.2	Identifying Suitable Studies for Use in Benefits Assessments

Three studies evaluated prospective cohorts, two of which included more recent timespans and likely
lower 03 concentrations. One of those studies took place entirely in Canada but included a substantially
larger study size (>200 times larger) than the other. As the asthma onset endpoint is consistent between
studies, pooling may be appropriate.

2.2.6.4.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Tetreault et al., 2016 investigated the effects of long-term 03 exposure on asthma onset in children
(aged zero-12 years) from Quebec, Canada. The study followed participants from the Quebec Integrated
Chronic Disease Surveillance System open birth cohort between 1999 and 2011. The authors defined

53


-------
new cases of asthma based on hospital discharge reports and physician diagnoses (two diagnoses within
a two-year span). Monitor data (Canadian National Air Pollution Surveillance network) and land-use
mixed effect models estimated warm season (June to August) 03 exposures. Authors assessed
associations with asthma onset with both time of birth and time-varying exposure models and adjusted
for year of birth, sex, and indices of social and material deprivation. Tetreault et al., 2016 used Cox
proportional hazard models to observe associations between long-term 03 exposure and asthma onset
in children. The identified single-pollutant, warm-season hazard ratio was 1.07 (95% CI: 1.06-1.08) for a
3.26 ppb (interquartile range) increase in annual 03 concentrations.

As the physiology and etiology of lung development in children is similar in children 6-17 (Baena-Cagnani
et al., 2007, Guerra et al., 2004, Ochs et al., 2004, Sparrow et al., 1991, Trivedi and Denton, 2019), we
apply the 4-12 year age-stratified effect estimate from Tetreault et al., 2016 to children ages 4-17.

2.2.6.5 Asthma Symptoms

Asthma symptoms/exacerbation is identified as a health effect of short-term 03 exposure (section
2.2.1.2.2.1). Overall, the ISA found that "evidence from recent epidemiologic and experimental studies
continues to support an association between ozone and asthma exacerbation" with
"associations...observed across a range of ozone concentrations, and...consistent in models with
measured or modeled concentrations" (U.S. EPA, 2020a, section 3.1.5.7).

2.2.6.5.1	Available Epidemiologic Literature

Four epidemiologic studies of asthma symptoms meeting our minimum criteria (section 2.1.1) were
identified in the 2020 03 ISA (U.S. EPA, 2020a). Most studies took place in the late nineties and very
early 2000s, and although no study included >1000 participants, there was appreciable geographic
representation. There were also differences regarding the ozone season. Only the oldest study
specifically evaluated a warm season, although a more recent study did skew slightly toward warmer
seasons, evaluating eight seasons over a four-year timespan (two Summers, three Springs, two Falls, and
one Winter).

2.2.6.5.2	Identifying Suitable Studies for Use in Benefits Assessments

Two of the studies evaluated much higher 03 concentrations (~50 ppb vs 30 ppb). Of the two studies
evaluating lower pollutant concentrations, one employed a prospective study design and clearly defined
the specific asthma symptoms evaluated (e.g., wheeze).

2.2.6.5.3	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Lewis et al., 2013 studied the effects of short-term 03 exposure on frequency of asthma symptoms in an
asthmatic population of primarily lower-income, African American and Latino children (aged five-12
years) in East and Southwest Detroit, Ml. Authors obtained health and demographic data through
questionnaires filled out by parents or guardians for 14 consecutive days in each studied season.
Questionnaires highlighted participant's asthma symptoms (cough, wheeze, shortness of breath, chest
tightness), demographic information, medication use, and presence of second-hand smoke. The authors
acquired maximum one-hour and maximum 8-hour 03 concentrations and meteorological data from two
community-level monitors placed on East and Southwest Detroit, Ml school rooftops. Lewis et al., 2013
implemented a combination of generalized estimating equations and alternative logistic regression
models to estimate the associations between short-term 03 exposure and rate of asthma symptoms.
Models adjusted for age, sex, location (Eastside or Southwest), race, household income, smoker in the

54


-------
home, season, and variables for companion home intervention study (control or intervention), time
(pre- or post-intervention), and the interaction between intervention group status and time. Lewis et al.,
2013 observed positive associations between short-term 03 exposure and asthma symptoms, including
the identified single-pollutant, all year odds ratios of 1.12 (95% CI: 0.99-1.25) for cough, 1.13 (95% CI:
0.99-1.28) for wheeze, 1.20 (95% CI: 1.02-1.40) for chest tightness, and 1.07 (95% CI: 0.95-1.21) for
shortness of breath, all for a 16 ppb (interquartile range) increase in 8-hour maximum 03 concentrations
(five-day average lag).53

2.2.6.6	Minor Restricted Activity Days

No new epidemiologic studies of minor restricted activity days (MRADs) were identified in the 2020 03
ISA (U.S. EPA, 2020a).

2.2.6.6.1 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Ostro and Rothschild, 1989 estimated the impact of PM2.5 and 03 on the incidence of minor restricted
activity days (MRADs) and respiratory-related restricted activity days (RRADs) in a national sample of the
adult working population, ages 18 to 65, living in metropolitan areas. The study population is based on
the Health Interview Survey (HIS), conducted by the National Center for Health Statistics. In publications
from this ongoing survey, non-elderly adult populations are generally reported as ages 18-64. From the
study, it is not clear if the age range stops at or includes those aged 65. We apply the risk estimate
function to individuals ages 18-64 for consistency with other studies estimating impacts to non-elderly
adult populations. The annual national survey results used in this analysis were conducted in the period
1976-1981, controlling for PM2.5, two-week average 03.

2.2.6.7	Allergic Rhinitis (Hay Fever/Respiratory Allergies))

The 2020 03 ISA stated that "cross-sectional epidemiologic studies provide generally consistent evidence
that ozone concentrations are associated with hay fever/rhinitis" and included "allergic responses" in
the biological plausibility diagram for long-term 03-attributal respiratory effects (U.S. EPA, 2020a,
section 2.2.1.2.2.1). Although cross sectional analyses do not establish a temporal sequence, they can be
used to estimate benefits associated with changes in air quality.

2.2.6.7.1	Available Epidemiologic Literature

The 2020 03 ISA identified one epidemiologic study of long-term 03 exposure and allergic rhinitis.

2.2.6.7.2	Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Parker et al., 2009 investigated the associations between long-term 03 exposure and respiratory
allergies in an unrestricted population of children (aged 3-17 years) sampled from the United States
National Health Interview Survey. Authors obtained symptom data from participant parents, who
reported respiratory allergies on annual surveys. Parker et al., 2009 placed all study participants
reporting symptoms of respiratory allergies or hay fever into a combined rhinitis group. Parker et al.,
2009 linked annual averages of S02, N02, PM2.5, and PM2.5-10 and warm season (May to September) 03
averages to participant's addresses through ambient air pollution and meteorological data collected
from US EPA Air Quality System monitors. The authors adjusted models for survey year, poverty-level,
race/ethnicity, age, family structure, insurance coverage, usual source of care, education of adult,
urban-rural status, region, and median county-level income. Through multi-pollutant, logistic regression

53 Estimates were obtained from figures. Authors did not respond to requests for exact results.

55


-------
models, the odds ratio of 1.18 (95% CI: 1.09-1.27) for a 10 ppb increase in 24-hour mean, warm season
03 and respiratory allergies was identified.

2.2.6.8 School Loss Days

No new studies of work loss days (WLDs) were identified in the 2020 03 ISA (U.S. EPA, 2020a).

2.2.6.8.1 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
Gilliland et al., 2001 examined the association between air pollution and school absenteeism among
fourth grade school children (aged nine to 10) in 12 southern Californian communities. The study was
conducted from January through June 1996. The authors used school records to collect daily absence
data and parental telephone interviews to identify causes. They defined illness- related absences as
respiratory or non-respiratory. A respiratory illness was defined as an illness that included at least one of
the following: runny nose/sneezing, sore throat, cough, earache, wheezing, or asthma attack. The
authors used 15- and 30-day distributed lag models to quantify the association between 03 and incident
school absences. 03 levels were positively associated with all school absence measures and significantly
associated with all illness-related school absences (non-respiratory illness, respiratory illness, URI and
LRI). The health impact function for ozone is based on the results of the single pollutant model.

Gilliland et al., 2001 defines an incident absence as an absence that followed attendance on the
previous day and the incidence rate as the number of incident absences on a given day over the
population at risk for an absence on a given day (i.e. those children who were not absent on the
previous day). Since school absences due to air pollution may last longer than one day, an estimate of
the average duration of school absences could be used to calculate the total avoided school loss days
from an estimate of avoided new absences. A simple ratio of the total absence rate divided by the new
absence rate would provide an estimate of the average duration of school absences, which could be
applied to the estimate of avoided new absences as follows:

TotalAbsences

Duration =	-	

NewAbsences

ATotalAbsences = —[incidences x (e~Px03 — l)] x duration x pop

Since the function is log-linear, the baseline incidence rate (in this case, the rate of new absences) is
multiplied by duration, which reduces to the total school absence rate. Therefore, the same result would
be obtained by using a single estimate of the total school absence rate in the risk estimate. Using this
approach, we assume that the same relationship observed between pollutant and new school absences
in the study would be observed for total absences on a given day. As a result, the total school absence
rate is used in the function below. The derivation of this rate is described in the section on baseline
incidence rate estimation.

For all absences, the coefficient and standard error are based on a percent increase of 16.3 percent
(95% CI -2.6 percent, 38.9 percent) associated with a 20 ppb increase in eight-hour average ozone
concentration (2001, Table 6, p. 52).

56


-------
A scaling factor is used to adjust for the number of school days in the ozone season. In the modeling
program, the function is applied to every day in the ozone season (May 1 - September 30), however, in
reality, school absences will be avoided only on school days. We assume that children are in school
during weekdays for all of May, two weeks in June, one week in August, and all of September. This
corresponds to approximately 2.75 months out of the five-month season, resulting in an estimate of
39.3% of days (2.75/5*5/7).

In addition, not all children are at-risk for a new school absence, as defined by the study. On average,
5.5% of school children are absent from school on a given day (NCES, 1996, Table 42-1). Only those who
are in school on the previous day are at risk for a new absence (1-0.055 = 94.5%). As a result, a factor of
94.5% is used in the function to estimate the population of school children at-risk for a new absence.

2.3 Identified Study and Risk Estimates for Benefits Assessments

While we begin with studies identified in ISAs, the goals of an ISA differ greatly from those of benefits
assessments. ISAs evaluate the overall state of the science and develop overarching conclusions relating
exposure to health effects. This includes analyses of specific subgroups, such as people with pre-existing
conditions, that may not be transferrable to the entire U.S. population.

In an effort to make our study and risk estimate identification process as transparent and reproducible
as possible, we have explicitly stated the criteria used in our approach (section 1) as well as the available
epidemiologic studies evaluated (section 2.2). However, even with such detailed information, expert
judgment can be required if multiple estimates meet the required criteria, satisfy a similar number of
preferred criteria, and are unable to be statistically aggregated into a single risk estimate (i.e., pooling).

The two tables in this section provide information on the health endpoints and risk estimates identified
for use in PM2.5 and 03 benefits estimation (Table 10 and Table 11) using the systematic approach
described above (sections 2.1 and 2.2).

2.3.1 Health Endpoints

These summary tables provided an overview of the PM2.5 and 03 health endpoints included in the main
benefits analysis. They are the outcome of the systematic approach described above, which involved

57


-------
consideration of recent ISA conclusions along with the availability of clinically relevant epidemiologic risk
estimates (Table 8 and Table 9).

2.3.1.1 PM25

Table 8. Set of Health Endpoints for Main PM2.5 Benefits Assessments

Endpoint Group

Endpoint Type

Specific Endpoint

Exposure

Ages

Mortality

Mortality

All cause

LT

Adults and older adults
(30-99 and 65-99 years)

ST

Infants (1-12 months)

Cardiovascular
Effects

Hospital
Admissions

Cardiovascular
Outcomes

ST

Older adults (65-99
years)

Emergency
Department Visits

Cardiovascular
Outcomes

ST

Children, adults, and
older adults (0-99 years)

Incidence

Acute Myocardial
Infarction

ST

Adults and older adults
(18-99 years)

Stroke3

LT

Older adults (65-99
years)

Cardiac Arrest3

ST

Adults and older adults
(0-99 years)

Respiratory
Effects

Hospital
Admissions

Respiratory
Outcomes

ST

Children and older
adults (65-99 years)

Emergency
Department Visits

Respiratory
Outcomes

ST

Children, adults, and
older adults (0-99 years)

Incidence

Asthma Onset3

LT

Children (0-17 years)

Asthma Symptoms

ST

Children (6-17 years)

Allergic Rhinitis3

LT

Children (3-17 years)

Minor Restricted
Activity Days

NA

Adults and older adults
(18-64 years)

Work Loss Days

NA

Adults and older adults
(18-64 years)

Cancer

Incidence

Lung Cancer3

LT

Adults and older adults
(30-99 years)

Nervous System
Effects

Hospital
Admissions

Alzheimer's Disease3

LT

Older adults (65-99
years)

Parkinson's Disease3

LT

Older adults (65-99
years)

aNew health endpoint.

58


-------
2.3.1.2 03

Table 9. Set of Health Endpoints for Main 03 Benefits Assessments

Endpoint
Group

Endpoint Type

Specific Endpoint

Exposure

Ages

Mortality

Mortality

Respiratory3

ST

Children, adults, and older
adults (0-99 years)

LT

Adults and older adults
(30-99 years)

Respiratory
Effects

Hospital Admissions

Respiratory
Outcomes

ST

Older adults (65-99 years)

Emergency
Department Visits

Respiratory
Outcomes

ST

Children, adults, and older
adults (0-99 years)

Incidence

Asthma Onset3

LT

Children (0-17 years)

Asthma Symptoms

ST

Children (5-17 years)

Allergic Rhinitis3

LT

Children (3-17 years)

Minor Restricted
Activity Days

ST

Adults and older adults
(18-64 years)

School Loss Days

ST

Children (5-12 years)

ST

Children (9-10 years)

aNew or updated health endpoint.

2.3.2. Risk Estimates

This section presents the risk estimates identified for the main PM2.5 (section 2.3.2.1) and 03 (section
2.3.2.2) benefits assessments. These lists reflect the application of the available epidemiologic literature
(section 2.2) to the identification criteria (section 2.1).

2.3.2.1 PM2.5

Table 10. Set of Risk Estimates for Main PM2.5 Benefits Assessments

Endpoint

Study Information

Ages

Exposure
Duration

Beta Coefficient
(SE)1

Mortality

Di etal., 2017b

Older adults
(65-99 years)

LT

|B = 0.0070 (0.0001)

Turner et al., 2016

Adults (30-99
years)

LT

|B = 0.0058 (0.00096)

Woodruff et al., 2008

Infants (1-12
months)

LT

|B = 0.0056 (0.00454)

Hospital

Admissions,

Cardiovascular

Bell etal., 2015 —ICD
410, omitting 410.x2;
410-414; 426-427;
428; 429; 430-438;
and 440-448

Older adults
(65-99 years)

ST

P = 0.00065 (0.00009)

Emergency

Department

Visits,

Cardiovascular

Ostro et al., 2016—
ICD 390-459

Children older
adults (0-99
years)

ST

P = 0.00061 (0.00042)

59


-------
Endpoint

Study Information

Ages

Exposure
Duration

Beta Coefficient
(SE)1

Acute

Myocardial

Infarction

Peters et al., 2001

Adults and
older adults
(18-99 years)

ST

13 = 0.02412(0.00928)

Pope III et al., 2006
Sullivan et al., 2005
Zanobetti et al., 2009
Zanobetti and
Schwartz, 2006

Adults and
older adults
(18-99 years)

ST

13 = 0.00481(0.00199)
13 = 0.00198(0.00224)
13 = 0.00225 (0.00059)
13 = 0.0053 (0.00221)

Cardiac Arrest

Ensor et al., 2013
Rosenthal et al., 2008
Silverman et al., 2010

Adults and
older adults
(0-99 years)

ST

13 = 0.00638(0.00282)
13 = 0.00198(0.00502)
13 = 0.00392(0.00222)

Stroke

Kloog et al., 2012—
ICD 430-436

Older adults
(65-99 years)

LT

13 = 0.00343 (0.00127)

Hospital

Admissions,

Respiratory

Bell et al., 2015—ICD
490-492, 464-466,
480-487, 493

Older adults
(65-99 years)

ST

13 = 0.00025 (0.00012)

Ostro et al., 2009—
ICD 460-519

Children (0-18
years)

ST

13 = 0.00275 (0.00077)

Emergency
Department
Visits,
Respiratory

Krall et al., 2013—
ICD 480-486, 491,
492, 496, 460-465,
466, 477, 493, 786.07

Children,
adults, and
older adults
(0-99 years)

ST

13 = 0.00055(0.00027) (GA)
(3 = 0.00097 (0.00035) (AL)
13 = 0.00083 (0.00033) (MO)
(3 = 0.00135 (0.00059) (TX)

Asthma Onset

Tetreault et al., 2016

Children (0-17
years)

LT

(3 = 0.04367 (0.00088)

Allergic Rhinitis

Parker et al., 2009

Children (3-
17)

LT

13 = 0.02546 (0.00962)

Lung Cancer

Gharibvand et al.,
2017

Adults and
older adults
(>29 years)

LT

13 = 0.03784(0.01312)

Alzheimer's
Disease

Kioumourtzoglou et
al., 2016—ICD 331.0

Older adults
(>64 years)

LT

13 = 0.13976 (0.01775)

Parkinson's
Disease

Kioumourtzoglou et
al., 2016—ICD 332

Older adults
(>64 years)

LT

13 = 0.07696 (0.01891)

Asthma
Symptoms

Rabinovitch et al.,
2006

Children (6-17
years)

ST

13 = 0.00200 (0.00148)

Minor
Restricted
Activity Days

Ostro and Rothschild,
1989

Adults and
older adults
(18-64 years)

N/A

13 = 0.00741(0.0007)

Work Loss Days

Ostro, 1987

Adults and
older adults
(18-64 years)

N/A

(3 = 0.0046 (0.00036)

ST- short-term; LT- long-term, (3- beta risk estimate; ICD- International Statistical Classification of Diseases
Notes: Horizontal lines separating studies within an endpoint indicates that the studies are not intended to be pooled.

1 Risk estimates have been mathematically converted to beta coefficients, which include the increment of pollutant change and
allow for more direct comparisons of risk estimates within health endpoints.

60


-------
2.3.2.2 03

Table 11. Set of Risk Estimates for Main 03 Benefits Assessments

Endpoint

Study

Information

Ages

Exposure (Duration;
Season; Metric)

Beta Coefficient
(SE)1

Respiratory
Mortality

Zanobetti and
Schwartz, 2008

Children,
adults, and
older adults
(0-99 years)

ST; June-August;
DA8

|B = 0.00083 (0.00023)
(warm season)

Katsouyanni et
al., 2009

Children,
adults, and
older adults
(0-99 years)

ST; April-September;
MDA1

|B = 0.00073 (0.00057)
(warm season)

Turner et al.,
2016—ICD 460-
519

Adults and
older adults
(30-99 years)

LT; April-September;
MDA8

|B = 0.007696 (0.00118)
(warm season)

Hospital

Admissions,

Respiratory

Katsouyanni et
al., 2009—ICD
460-519

Older adults
(65-99 years)

ST; April-September;
MDA1

P = 0.00028 (0.00018)
(warm season)

Emergency
Department
Visits,
Respiratory

Barry et al.,
2019—ICD 493,
786.07, 460-
466, 477, 491,
492, 496, 480-
486, 466.1,
466.11, 466.19

Children,
adults, and
older adults
(0-99 years)

ST; January-
December; MDA8

P = 0.00118 (0.00040)
(Atlanta, GA)
P = 0.00118 (0.00059)
(Birmingham, AL)
P = 0.00195 (0.00049)
(Dallas, TX)
P = 0.00118 (0.00040)
(Pittsburgh, PA)
P = 0.00079 (0.00030)
(St. Louis, MO-IL)

Asthma Onset

Tetreault et al.,
2016

Children (0-
17 years)

LT; June-August;
MDA8

P = 0.02075 (0.00146)
(warm season)

Asthma
Symptoms

Lewis et al.,
2013

Children (5-
17 years)

ST; January-
December; MDA8

P = 0.00708 (0.00372)
(Cough)

P = 0.00764 (0.00410)
(Wheeze)

P = 0.01140 (0.00505)
(Chest tightness)
P = 0.00423 (0.00386)
(Shortness of breath)

Allergic Rhinitis

Parker et al.,
2009

Children (3-
17 years)

LT; May-September;
DA24

P = 0.01655 (0.00390)
(warm season)

Minor Restricted
Activity Days

Ostro and
Rothschild, 1989
(MRADs)

Adults and
older adults
(18-64 years)

ST; April-September;
MDA1

P = 0.0022 (0.000658)

School Loss Days

Gilliland et al.,
2001

Children (5-
17 years)

ST; January-June;
DA8

P = 0.0078 (0.0044)

61


-------
ST- short-term; LT- long-term, (3- risk estimate (beta); ICD- International Statistical Classification of Diseases; DA8- daily 8-hour
average; MDA8- maximum daily 8-hour average; MDA1- maximum daily 1-hour average; DA24- daily 24-hour average
Notes: Horizontal lines separating studies within an endpoint indicates that the studies are not intended to be pooled
1 Risk estimates have been mathematically converted to beta coefficients, which include the increment of pollutant change and
allow for more direct comparisons risk estimates within health endpoints.

62


-------
3 Baseline Incidence and Prevalence Estimates

A baseline incidence rate is an estimate of the number of new cases in the assessment location over a
specific timespan, typically one year. For example, in 2018 the mortality rate was 868 deaths per
100,000 people in the U.S.54 The baseline incidence of the health effect is necessary to convert the
relative risk of a health effect provided by epidemiologic studies into an estimated number of cases. To
derive the total baseline incidence per year, this rate must be multiplied by the corresponding
population. Continuing with the above example, there were 327 million people in the U.S. in 2018,
leading to a total baseline incidence of 2.8 million deaths in that year.

Prevalence rates are the proportion of the population experiencing a health endpoint at a point in time.
This rate is important when estimating impacts of chronic illnesses, such as asthma, in order to exclude
those already diagnosed from the population at risk. For example, if the prevalence of asthmatic
children is 8%, only the remaining 92% are at risk of developing asthma.

EPA develops either daily or annual baseline incidence and prevalence rates at the most geographically-
and age-specific levels feasible for each health endpoint assessed. For many locations within the U.S.,
these data are available resolved at the county- or state-level, providing a better characterization of the
geographic distribution of hospital and emergency department visits than the national rates. For this
update, we focused on developing baseline incidence rates for new health endpoints. Detailed
information on baseline incidence data developed previously can be found in Appendix D of the
BenMAP-CE User Manual (U.S. EPA, 2018). Importantly, when applying either the daily or annual
baseline incidence rates to a health impact estimate, the temporal scale over which the health endpoint
was assessed within each study is taken into account. For example, if a long-term 03 exposure study
associated annual deaths with warm-season exposures, full-year baseline incidence rates will be used
when estimating benefits.55

Table 12 summarizes the sources of baseline incidence rates and provides national average (where
used) incidence rates for the endpoints included in the analysis. For both baseline incidence and
prevalence data, we used age-stratified rates where available. We applied risk estimates to individual
age groups and then sum them over the relevant age range to estimate total population benefits. In
some cases we used a single national incidence rate, due to a lack of more spatially disaggregated data,
time, or resources. In these cases, whenever possible we used national average rates, because these
data are most applicable to a national assessment of benefits. For some studies, however, the only
available incidence information comes from the studies themselves; in these cases, incidence in the
study population is assumed to represent typical incidence at the national level.

54	CDC WONDER mortality data; https://www.cdc.gov/nchs/fastats/deaths.htm.

55	Turner et al., 2016 and Tetreault et al., 2016 risk estimates of long-term 03-attributable health impacts use full-
year baseline incidence rates, even though the exposure period is restricted to the warm season. As such, our
baseline incidence rate estimates also reflect the full year for those health endpoints.

63


-------
Table 12. Baseline Incidence Rates for Use in Impact Functions

Endpoint

Parameter

Rates





Value

Source

Mortality1

Daily or annual projected
incidence to 2060 in 5-
year increments (0-99)

Age-, cause-, race-, and
county-stratified rates

CDC WONDER (2012-
2014)

U.S. Census Bureau,
2012

Hospitalizations2

Daily incidence rates for
all ages

Age-,

region/state/county-, and
cause- stratified rates

2011-2014 HCUP data
files and data
requested from and
supplied by individual
states

Emergency
Department Visits2

Daily emergency
department visit incidence
rates for all ages

Age-, region-, state-,
county-, and cause-
stratified rates

2011-2014 HCUP data
files and data
requested from and
supplied by individual
states

Nonfatal Acute

Myocardial

Infarction

Daily nonfatal AMI
incidence rate per person
aged 18-99

Age-, region-, state-, and
county- stratified rates

AHRQ, 2016

Asthma Symptoms

Daily incidence among
asthmatic children







Wheeze (ages 5-12)
Cough (ages 5-12)
Shortness of breath (ages
5-12)

Albuterol use (ages 6-13)

Age- and race- stratified
rates

2.2 puffs per day

Ostro et al., 2001

Rabinovitch et al.,
2006

Asthma Onset

Annual incidence
0-4
5 -11
12 -17

0.0234
0.0111
0.0044

Winer et al., 2012

Alzheimer's
Disease

Daily incidence rates for
all ages

Age-, region-, state-, and
county- stratified rates

2011-2014 HCUP data
files

Parkinson's Disease

Annual incidence

18-44

45-64

65-84

85-99

0.0000011
0.0000366
0.0002001
0.0002483

HCUPnet

Allergic Rhinitis

Respondents aged 3-17
experiencing allergic
rhinitis/hay fever
symptoms within the year
prior to the survey

0.192

Parker et al., 2009

64


-------
Endpoint

Parameter

Rates





Value

Source

Cardiac Arrest

Daily nonfatal incidence
rates



Ensor et al., 2013,
Rosenthal et al., 2008,



0 -17

0.00000002

Silverman et al., 2010



18-39

0.00000009





40-64

0.00000056





65-99

0.00000133



Lung Cancer

Annual nonfatal incidence
25-34



NCI, 2015 and
Gharibvand et al.,



35-44

0.000001746

2017



45-54

0.000014919





55-64

0.000067463





65-74

0.000208053





75-84

0.000052370





95-99

0.000576950
0.000557130



Stroke

Annual nonfatal incidence
in ages 65-99

0.00446

Kloog et al., 2012

Work Loss Days

Daily incidence rate per
person (18-64)



Adams et al., 1999,
Table 41; U.S. Census



Aged 18-24

0.00540

Bureau(2000)



Aged 25-44

0.00678





Aged 45-64

0.00492



School Loss Days

Rate per person per year,
assuming 180 school days
per year

9.9

Adams et al., 1999,
Table 47

Minor Restricted-

Daily MRAD incidence rate

0.02137

Ostro and Rothschild,

Activity Days

per person (18-64)



1989, p. 243

CDC-Centers for Disease Control; NHS-National Health Interview Survey

1Mortality rates are only available in 5-year increments. The Healthcare Cost and Utilization Program (HCUP) database contains
individual level, state and regional-level hospital and emergency department discharges for a variety of International
Classification of Diseases (ICD) codes (AHRQ, 2016).

2Baseline incidence rates now include corrections from the states of Indiana and Montana.

3.1 Mortality

Baseline incidence rate estimates for mortality remain the same as they were for previous benefits
assessments (U.S. EPA, 2018). However, information is provided below for reference. Notably, the
Turner et al., 2016 analysis of long-term 03-attributable health impacts compares warm-season
exposures to full-year baseline incidence rates. As such our baseline incidence rate estimates also reflect
the full year.

65


-------
3.1.1 Mortality Data for 2012-2014

We obtained county-level mortality and population data from 2012-2014 for 11 causes for the
contiguous United States by downloading the data from the Centers for Disease Control (CDC) WONDER
database.56

Since the detailed mortality data obtained from CDC do not include population, we combined them with
U.S. Census Bureau population estimates exported from BenMAP. We then generated age-, cause-, and
county-specific mortality rates using the following formula:

_ £>iJ-fc(2012)+DiJ-fc(2013)+DiJ-fc(2014)
i,],k ~ Pj,fc(2012)+Pj,fc(2013)+Pj,fc(2014)

where R,-jis the mortality rate for age group /', cause j, and county k; D is the death count; and P is the
population. Additional details about the translation of the CDC WONDER data to age-, cause-, and
county-specific mortality rates are provided in the BenMAP-CE User's Manual (U.S. EPA, 2018).

56 http://wonder.cdc.gov

66


-------
Table 13. National Mortality Rates (per 100 people per year) by Health Endpoint and Age Group, 2012-2014

Mortality Category

ICD-10 Codes

Infant*

1-17

18-24

25-34

35-44

45-54

55-64

65-74

75-84

85+

Mortality, All
Cause

All

0.59396

0.01951

0.07804

0.10665

0.17264

0.40542

0.86162

1.79670

4.62837

13.58034

Mortality, Non-
Accidental

A00-R99

0.55495

0.00949

0.01874

0.04112

0.10876

0.33084

0.79395

1.73208

4.49595

13.20867

Mortality,
Respiratory

J00-J98

0.01297

0.00102

0.00127

0.00253

0.00570

0.02013

0.06560

0.20585

0.57827

1.42362

Mortality, Chronic
Lung

J40-J47, J67

0.00053

0.00032

0.00040

0.00074

0.00186

0.01033

0.04045

0.13873

0.36008

0.68593

Mortality, Lung
Cancer

C34

0.00002

0.00001

0.00007

0.00033

0.00282

0.02378

0.07992

0.19701

0.32952

0.31820

Mortality, Ischemic
Heart Disease

120-125

0.00033

0.00004

0.00039

0.00234

0.01242

0.04854

0.12174

0.25698

0.68000

2.27271

Mortality, Cardio-
pulmonary

100-178, J10-J18, J40-J47,
J67

0.00539

0.00069

0.00099

0.00214

0.00502

0.01794

0.05877

0.18453

0.51055

1.26213

Mortality, NCD +
LRI

**

0.18459

0.00618

0.01168

0.02751

0.08129

0.26214

0.63767

1.37694

3.44731

9.47467

Mortality, Lower

Respiratory

Infection

A48.1, A70, B97.4-B97.6,
J09-J15.8, J16, J20-J21,
P23.0-P23.4, U04

0.00269

0.00618

0.01168

0.00030

0.00062

0.00112

0.00196

0.00300

0.00758

0.02693

Mortality, Cerebro-
vascular

G45-G46.8, 160-163.9,
165-166.9, 167.0-167.3,
167.5-167.6, 168.1-168.2,
169.0-169.3

0.00116

0.00012

0.00034

0.00096

0.00314

0.00809

0.01455

0.02892

0.08553

0.20863

Mortality, COPD

J40-J44, J47

0.00048

0.00005

0.00004

0.00015

0.00102

0.00904

0.03888

0.13689

0.35661

0.67457

*We estimate post-neonatal mortality (deaths after the first month) for infants because the health impact function (see Appendix E) estimates post-neonatal mortality.

**For a full list of codes for non-communicable diseases (NCD) and lower respiratory infections (LRI), see the IHME GBD Code mapping: http://ghdx.healthdata.org/record/ihme-

data/gbd-2017-cause-icd-code-mappings.

67


-------
3,1.2 Mortality Rate Projections 2015-2060

To estimate age- and county-specific mortality rates in years 2015 through 2060, we calculated annual
adjustment factors, based on a series of Census Bureau projected national mortality rates (for all- cause
mortality), to adjust the age-, county-, and cause-specific mortality rates calculated using 2012-2014
data as described above.57 We used the following procedure:

For each age group, we obtained the series of projected national mortality rates from 2013 to 2050 (see
the 2013 rate in Table 14) based on Census Bureau projected life tables.

We then calculated, separately for each age group, the ratio of Census Bureau national mortality rate in
year Y (Y = 2014, 2015, ..., 2060) to the 2013 rate, which is assumed to be representative of the 2012-
2014 data and used for the base "year." These ratios are shown for selected years in Table 15.

Finally, to estimate mortality rates in year Y (Y = 2015, 2020, ..., 2060) that are both age-group-specific
and county-specific, we multiplied the county- and age-group-specific mortality rates for 2012-2014 by
the appropriate ratio calculated in the previous step. For example, to estimate the projected mortality
rate in 2015 among ages 18-24 in Wayne County, Ml, we multiplied the mortality rate for ages 18-24 in
Wayne County in 2012-2014 by the ratio of Census Bureau projected national mortality rate in 2015 for
ages 18-24 to Census Bureau national mortality rate in 2013 for ages 18-24.

Table 14. All-Cause Mortality Rate (per 100 people per year), by Source, Year, and Age Group

Source and Year

Infant

1-17

18-24

25-34

35-44

45-54

55-64

65-74

75-84

85+

Calculated CDC 2012-2014

0.5941

0.020

0.078

0.107

0.173

0.405

0.862

1.797

4.628

13.580

Census Bureau 20132

0.654

0.029

0.088

0.102

0.183

0.387

0.930

2.292

5.409

13.091

1The Census Bureau estimate is for all deaths in the first year of life. EPA benefits assessments uses post-neonatal mortality
(deaths after the first month, i.e., 0.23 per 100 people) because the health impact function (see Appendix E) estimates post-
neonatal mortality. For comparison purpose, we also calculated the rate for all deaths in the first year, which is 0.684 per 100
people.

2For a detailed description of the model, the assumptions, and the data used to create Census Bureau projections, see the
working paper, "Methodology and Assumptions for the 2012 National Projections," which is available on
http://www.census.gov/population/projections/files/methodology/methodstatementl2.pdf

57 All-cause mortality projections are applied to each cause-specific mortality rate.

68


-------
Table 15. Ratio of Future Year All-Cause Mortality Rate to 2013 Estimated All-Cause Mortality Rate, by
Age Group

Year

Infant

1-17

18-24

25-34

35-44

45-54

55-64

65-74

75-84

85+

2015

0.93

0.93

0.96

1.02

0.96

0.96

1.01

1.02

1.03

1.00

2020

0.94

0.94

0.98

1.04

0.97

0.98

1.02

1.03

1.03

1.00

2025

0.85

0.81

0.74

0.80

0.75

0.77

0.85

0.91

0.93

0.97

2030

0.81

0.75

0.66

0.70

0.67

0.69

0.78

0.86

0.89

0.92

2035

0.76

0.70

0.58

0.62

0.60

0.62

0.71

0.81

0.87

0.87

2040

0.73

0.65

0.51

0.53

0.53

0.56

0.64

0.76

0.84

0.86

2045

0.70

0.60

0.45

0.46

0.46

0.50

0.58

0.71

0.80

0.86

2050

0.67

0.56

0.39

0.40

0.40

0.44

0.53

0.66

0.77

0.87

2055

0.64

0.52

0.34

0.35

0.35

0.39

0.48

0.62

0.73

0.88

2060

0.61

0.48

0.30

0.30

0.31

0.34

0.43

0.58

0.70

0.87

3.1,3 Race-Stratified Incidence Rates

To estimate race-stratified and age-stratified incidence rates at the county level, we downloaded all-
cause mortality data from 2007 to 2016 from the CDC WONDER mortality database.58 Race-stratified
incidence rates were calculated for the following age groups: < 1 year, 1- 4 years, 5-14 years, 15-24
years, 25-34 years, 35-44 years, 45-54 years, 55-64 years, 65-74 years, 75-84 years, and 85+ years. We
stratified the data into two race categories, White and Non-White, and follow all methods outlined in
section D.l.l of the BenMAP-CE User Manual (U.S. EPA, 2018). To properly impute incidence rates for
suppressed and unreliable counties, we downloaded data at the state, regional, and national scales.

3.2 Hospitalizations

The approach for estimating hospitalization baseline incidence rates for new health endpoints is based
on HCUP data, developed to match the granularity and timeframe of other hospitalization endpoints
used in benefits assessments. New hospitalization endpoints are comprised of new sets of ICD-9 codes
that correspond to newer studies evaluating air pollution-attributable hospitalizations. Detailed
information is provided below and available in the BenMAP-CE User Manual (U.S. EPA, 2018).

Hospitalization rates were calculated using data from the Healthcare Cost and Utilization Project (HCUP).
HCUP is a family of health care databases developed through a Federal-State-Industry partnership and
sponsored by the Agency for Healthcare Research and Quality (AHRQ). HCUP products include the State
Inpatient Databases (SID), the State Emergency Department Databases (SEDD), the Nationwide Inpatient
Sample (NIS), and the Nationwide Emergency Department Sample (NEDS).

The level of hospitalization data available differs by state. While many states provide granular discharge-
level data, others may only provide county- or state level-data. Also, 14 states, mostly in the southeast,

58 http://wonder.cdc.gov

69


-------
do not provide data to HCUP. For these states, regional statistics from HCUPnet59 were used to estimate
baseline hospitalization rates.

HCUP categorizes hospital admissions in various ways. Hospitalization admission types used when
reporting data to HCUP include emergency (admitted from the emergency department), urgent
(admitted from another hospital), elective (admitted from another health facility, including long-term
care), newborn (admitted for delivery), trauma (not used by all states), and other/missing/invalid. As
PM2.5 and 03 exposure predominantly leads to cardiovascular and respiratory health effects, we provide
some information on the proportion of these types of hospitalizations, based on an analysis of
hospitalizations from the state of Florida in 2014. Florida was selected for this analysis as it was the most
populated state providing details regarding hospital admission type.

•	Emergency hospital admissions comprise approximately 80% of cardiovascular and 85% of
respiratory admissions

•	Urgent hospital admissions comprise approximately 10% of cardiovascular and 8% of respiratory
admissions

•	Elective hospital admissions comprise approximately 10% of cardiovascular and 7% of
respiratory admissions

•	Newborn hospital admissions comprise no cardiovascular and respiratory admissions

•	Trauma hospital admissions comprise approximately 0.1% of cardiovascular and respiratory
admissions

•	Other/missing/invalid hospital admissions comprise no cardiovascular or respiratory admissions

All hospital admission baseline incidence data used in this analysis (and input into BenMAP-CE) reflects
total hospital admissions, due to time constraints limiting the ability to separate types (e.g., emergency,
urgent, elective, etc) within HCUP data by various states and regions. However, the breakdown of
hospital admission types generally reflects the types of health endpoints associated with air pollution
exposures, with the majority of effects falling into the emergency and urgent types (e.g., heart or
asthma attack) with a small subset potentially leading to elective hospital admissions (e.g., exacerbation
of heart failure).

Health endpoints in hospitalization studies are defined using different combinations of ICD codes
corresponding to specific diagnoses. Some span large categories of diagnoses, such as all cardiovascular
or all respiratory admissions, while others reflect specific conditions, including Alzheimer's disease and
Parkinson's disease.60 For each ICD code combination, unique baseline incidence rates are developed.

59	HCUPnet is a free, on-line query system based on data from HCUP. It provides access to summary statistics at the
state, regional and national levels.

60	Parkinson's disease incidence rates were developed in a slightly different manner, due to time and resource
limitations. We develop regional and age-specific incidence rates for Parkinson's disease hospital admissions using
the HCUPnet SID, which provides the total number of hospital visits in the U.S. by age group and region,
separately. We first calculate the distribution of annual hospital visits across HCUPnet's 6 age groups: less than 1,1
to 17,18 to 44, 45 to 64, 65 to 84, and above 85 years old. Since Parkinson's disease typically affects older adults,
hospitalization counts are unavailable for the age groups below 18 years old. We apply the national age
distribution to the regional hospitalization totals to estimate the annual number of hospital visits by region and
age. We then divide the regional and age-specific counts by the regional and age-specific population and by 365 to

70


-------
3.3	Emergency Department Visits

As new studies evaluating air pollution-attributable emergency department utilizing new sets of ICD-9
codes were identified for use in benefits assessment here, we developed corresponding new emergency
department baseline incidence rates. Similar to hospitalization baseline incidence rates, the approach
for estimating emergency department visit baseline incidence rates also utilizes HCUP data and remains
the same as in previous benefits assessments, details for which can be found in the BenMAP-CE User
Manual (U.S. EPA, 2018). Information is provided below for reference.

Similar to hospitalization rates, the data source for emergency department/room visits is also HCUP,
(i.e., SID, SEDD, and NEDS), states vary by level of data provided (i.e., discharge-, county-, state, and
regional-level), and unique baseline incidence rates are generated for each health endpoint ICD code
combination.

3.4	Health Endpoint Onset/Occurrence

Baseline incidence estimates for health endpoint onset or occurrences are described below, listed in
alphabetical order. Onset indicates the development of a health endpoint (e.g., asthma diagnosis),
whereas occurrence refers to an instance of that health endpoint (e.g., asthma attack).

3,4.1 Acute Myocardial Infarctions (AMIs)

Baseline incidence rate estimates for AMIs remain the same as they were for previous benefits
assessments. However, detailed information is provided below for reference.

The relationship between short-term particulate matter exposure and heart attacks was originally
quantified in a case-crossover analysis by Peters et al., 2001 and supplemented with evidence found in
more recent single and multi-city studies (Pope III et al., 2006, Sullivan et al., 2005). The population in
the original study was identified from heart attack survivors in a medical clinic. Therefore, the applicable
population to apply to the risk estimate is all individuals surviving a heart attack in a given year. Several
data sources are available to estimate the number of heart attacks per year. For example, several cohort
studies have reported estimates of heart attack incidence rates in the specific populations under study.
However, these rates depend on the specific characteristics of the populations under study and may not
be the best data to extrapolate nationally. The American Heart Association reports approximately
785,000 new heart attacks per year (Roger et al., 2012). Exclusion of heart attack deaths reported by
CDC Wonder yields approximately 575,000 nonfatal cases per year.

An alternative approach to the estimation of heart attack rates is to use data from the Healthcare Cost
and Utilization Project (HCUP), assuming that all heart attacks that are not instantly fatal will result in a
hospitalization. Details about HCUP data are described in Section D.2 of the BenMAP-CE User Manual
(U.S. EPA, 2018). According to the 2014 HCUP data there were approximately 608,795 hospitalizations
due to heart attacks (acute myocardial infarction: ICD-9 410, primary diagnosis). We estimated baseline
rates based on HCUP data rather than extrapolating from cohort studies because HCUP is a national
database with a larger sample size intended to provide reliable national estimates. The incidence rate
calculation is also described in Section D.2 of the BenMAP-CE User Manual and the incidence rates for

calculate the daily incidence rates. To generate county level incidence rates, we assume that each county has the
same incidence rate as the region it falls within.

71


-------
AMI hospitalization are presented in Table D-5. An alternative approach to the estimation of AMI rates is
to use data from HCUP and assume that all AMIs that are not instantly fatal will result in a
hospitalization.

It is important to note that when calculating the incidence of nonfatal AMIs, the fraction of fatal heart
attacks is subtracted to ensure that there is no double-counting with mortality estimates. Specifically,
we apply an adjustment factor in the risk estimate to reflect the probability of surviving a heart attack.
The adjustment factor comes from Rosamond et al., 1999, which reported that approximately 6% of
male and 8% of female hospitalized AMI patients die within 28 days (either in or outside of the hospital).
Therefore, we applied a factor of 0.93 to the estimated number of PM-related AMIs to exclude the
number of cases that result in death within the first month. Note that we did not adjust for fatal AMIs in
the incidence rate estimation, due to the way that the epidemiological studies are designed. Those
studies consider total admissions for AMIs, which includes individuals living at the time the studies were
conducted. We use the definition of AMI that matches the definition in the epidemiological studies. Age-
specific baseline incidence rates are based on data from the Agency for Healthcare Research and
Quality's HCUP NIS database (AHRQ, 2016). We identified death rates for adults hospitalized with AMI
stratified by age (e.g., 1.852% for ages 18-44, 2.8188% for ages 45-64, and 7.4339% for ages 65+). These
rates show a clear downward trend over time between 1994 and 2009 for the average adult and thus
replace the 93% survival rate previously applied across all age groups from Rosamond et al., 1999.

3,4.2 Asthma Onset and Symptoms

3.4.2.1 Asthma Onset

Baseline incidence rates for new asthma onset are estimated from Winer et al., 2012. Winer et al., 2012
identify newly diagnosed asthma from the 2006-2008 Asthma Call-Back Survey (ACBS) and Behavioral
Risk Factor Surveillance System (BRFSS) as individuals diagnosed by a doctor, or other health
professional, within the 12 months prior to the surveys. Table 12 details the breakdown, by age, of the
annual national incidence rates for asthma onset.

For the set of endpoints affecting the asthmatic population, in addition to baseline incidence rates,
prevalence rates of asthma in the population are needed to define the applicable population. We derive
asthma prevalence data from the National Health Interview Survey (NHIS).61 For functions with age
ranges that do not align with the ranges reported in the NHIS data table, we develop a weighted-
average prevalence rate for the age range, where the weights are the number of years that overlap with
each NHIS age group. Table 16 provides the breakdown of the 2018 NHIS rates used to calculate the
weighted averages. Table 17 details the resulting weighted averages by study and age group. Note that
these reflect recent asthma prevalence and assume no change in prevalence rates in future years.

61 https://www.cdc.gov/asthma/nhis/2018/data.htm and
https://www.cdc.gov/asthma/most_recent_national_asthma_data.htm

72


-------
Table 16. Asthma Prevalence Rates

NHIS Age Group

Asthma Prevalence Rate

0-4

0.038

5 -11

0.081

5 -14

0.086

12 -17

0.099

15 -19

0.110

20-24

0.081

0 -17

0.075

Table 17. Weighted Average Asthma Prevalence by Study

Endpoint

Ages

Author1

Pollutant

Weighted
Prevalence

Asthma Onset

0-4

Tetreault et al., 2016

PM2.5

0.0380

5 -17

Tetreault et al., 2016

PM2.5

0.0893

0 -17

Tetreault et al., 2016

03

0.0750

Asthma symptoms, albuterol use

6 -13

Rabinovitch et al., 2006

PM2.5

0.0860

Prevalence rate derived for albuterol use must be loaded into BenMAP-CE as part of a separate incidence or
prevalence dataset, unlike the remainder of the rates, which are embedded within the health impact functions.

3.4.2.2	Albuterol Use

We develop incidence rates for albuterol use from the rates presented in Rabinovitch et al., 2006, the
same study from which the risk estimate was developed. As described in the 'Recommended Set of
Health Endpoints and Health Impact Functions' section, Rabinovitch et al., 2006 analyzed the
relationship between short-term PM2.5 exposure and asthma exacerbation in children ages 6 to 13 years
old. The authors use an electronic inhaler to record the number of actuations ('puffs') for each 24-hour
period and calculate an average albuterol use rate of 2.2 'puffs' per child per day.

As described in section 3.4.2.1, in addition to the baseline incidence rates, we apply a weighted-average
asthma prevalence rate of 0.086, based on the 5-14 age group, using the NHIS prevalence data to
identify the applicable population.

3.4.2.3	Asthma Symptoms

We develop incidence rates for asthma symptoms using the estimates presented in Lewis et al., 2013,
the same study from which the concentration-response function was developed. As described in the
'Recommended Set of Health Endpoints and Health Impact Functions' section, Lewis et al., 2013 studied
the effects of short-term 03 exposure on frequency of asthma symptoms in an asthmatic population of
children ages 5 to 12 years old. The authors estimate the incidence of each asthma symptom using the
number of person-days where children reported experiencing the symptom divided by the total number
of person-days monitored for that symptom. The percent of days monitored during which children
experienced each symptom are calculated as 30.1% for cough, 19.4% for wheeze, 18.5% for shortness of
breath, and 12.7% for chest tightness. Therefore, the national incidence rates of asthma symptoms are
0.301 for cough, 0.194 for wheeze, 0.185 for shortness of breath, and 0.127 for chest tightness.

73


-------
'Prevalence rates for asthma symptoms remains the same as in previous benefits assessments
(previously referred to as asthma exacerbation) (Table 16).

3.4.3	Allergic Rhinitis

We develop prevalence rates for hay fever/rhinitis using the estimates presented in Parker et al., 2009,
the same study from which the concentration-response function was developed. As described in the
'Recommended set of Health Endpoints and Health Impact Functions' section, Parker et al., 2009
investigates the associations between long-term ozone exposure and respiratory allergies in children
ages 3 to 17 years old. The authors use prevalence data from the NHIS household interview survey and
define allergic rhinitis as children with reported hay fever, respiratory allergy, or both within the 12
months prior to the survey. Of the eligible population (72,279), 19.2% of respondents experience allergic
rhinitis symptoms within the year prior to the survey, therefore, the national prevalence rate of hay
allergic rhinitis is 0.192.

3.4.4	Lung Cancer

We use the existent baseline incidence rate for lung cancer mortality in combination with the five-year
lung cancer survival rate from NCI, 2015 to develop baseline incidence rates for non-fatal lung cancer.
We first use the five-year lung cancer survival rate to calculate the total incidence of lung cancer (both
fatal and non-fatal) from the baseline mortality rate using the following formula: baseline mortality rate
/ (1 - five-year survival rate). We then calculate the incidence of non-fatal lung cancer as the difference
between total lung cancer incidence and fatal lung cancer incidence (NCI, 2015). presents the baseline
incidence of lung cancer mortality, the SEER five-year survival rate, the estimated total lung cancer
incidence, and the estimated non-fatal lung cancer incidence rate by age group.

Table 18. Lung Cancer Incidence Rates

Age Group

Annual Lung
Cancer Mortality
Incidence
[A]

Five-Year
Survival Rate
[B]

Total Lung Cancer
Incidence
[C] =
[A] / (1 - [B])

Non-fatal Lung
Cancer Incidence
[D] =
[C] - [A]

25-34

0.0000033

34.6%

0.0000050

0.00000175

35-44

0.0000282

34.6%

0.0000431

0.00001492

45-54

0.0002378

22.1%

0.0003053

0.00006746

55-64

0.0007922

20.8%

0.0010003

0.00020805

65-74

0.00019701

21.0%

0.0002494

0.00005237

75-84

0.0032952

14.9%

0.0038722

0.00057695

85+

0.0031820

14.9%

0.0037391

0.00055713

3.4.5 Minor Restricted Activity Days (MRAD)

The incidence estimate for this health endpoint remains the same as in previous benefits assessments.
Ostro and Rothschild, 1989 (p. 243) provide an estimate of the annual incidence rate of MRADs per
person of 7.8.

74


-------
3,4.6 School Loss Days

Baseline incidence rate estimates for school loss days remain the same as they were for previous
benefits assessments. However, detailed information is provided below for reference.

We have two sources of information to use when estimating the baseline incidence rates of missed
school days: the National Center for Education Statistics (NCES), which provided an estimate of all-cause
school loss days, and the National Health Interview Survey (NHIS) (Adams et al., 1999, NCES, 1996, Table
47), which has data on different categories of acute school loss days. Table 19 presents the estimated
school loss day rates. Further detail is provided below on these rates.

Table 19. School Loss Day Rates (per student per year)

Type

Northeast

Midwest

South

West

Respiratory illness-related absences

1.3

1.7

1.1

2.2

Illness-related absences

2.4

2.6

2.6

3.7

All-cause

9.9

9.9

9.9

9.9

*lllness-related school loss day rates were based on data from the 1996 NHIS and an estimate of 180 school days per year,
excluding school loss days due to injuries. All-cause school loss day rates were based on data from the NCES.

3.4.6.1	All-Cause School Loss Day Rate

Based on data from the U.S. Department of Education (1996, Table 42-1), the National Center for
Education Statistics estimates that for the 1993-1994 school year, 5.5% of students are absent from
school on a given day. This estimate is comparable to study-specific estimates from Chen et al., 2000
and Ransom and Pope, 1992, which ranged from 4.5% to 5.1%.

3.4.6.2	Illness-Related School Loss Day Rate

The National Health Interview Survey (NHIS) has regional estimates of school loss days due to a variety
of acute conditions (Adams et al., 1999). NHIS is a nationwide sample-based survey of the health of the
noninstitutionalized, civilian population, conducted by NCHS. The survey collects data on acute
conditions, prevalence of chronic conditions, episodes of injury, activity limitations, and self-reported
health status. However, it does not provide an estimate of all-cause school loss days.

In estimating illness-related school loss days, we started with school loss days due to acute problems
(Adams et al., 1999, Table 47) and subtracted lost days due to injuries, in order to match the definition
of the study used in the risk estimate to estimate illness-related school absences (Gilliland et al., 2001).
We then divided by 180 school days per to estimate illness-related school absence rates per school day.
Similarly, when estimating respiratory illness-related school loss days, we use data from Adams et al.,
1999, Table 47. Note that we estimated 180 school days in a year to calculate respiratory illness-related
school absence rates per year.

3.4.7 Work Loss Days

The incidence estimate for this health endpoint remains the same as in previous benefits assessments.
The yearly work-loss-day incidence rate per 100 people is based on estimates from the 1996 National
Health Interview Survey (Adams et al., 1999, Table 41). They reported a total annual work loss days of
352 million for individuals ages 18 to 65. The total population of individuals of this age group in 1996
(162 million) was obtained from (U.S. Census Bureau, 1998). The average annual rate of work loss days

75


-------
per individual is 2.17. Using a similar approach, we calculated work-loss-day rates for ages 18-24, 25-44,
and 45-64, respectively.

76


-------
4 Demographic Information

Quantified and monetized human health impacts are calculated using information regarding the
demographic characteristics of the population exposed to air pollution; these data include the age, sex,
race/ethnicity and location of the population. We use population projections based on economic
forecasting models developed by Woods and Poole, Inc. (Woods & Poole, 2015). The Woods and Poole
(WP) database contains county-level projections of population by age, sex, and race out to 2050, relative
to a baseline using the 2010 Census data. Projections in each county are determined simultaneously
with every other county in the U.S to take into account patterns of economic growth and migration.

The sum of growth in county-level populations is constrained to equal a previously determined
national population growth, based on Bureau of Census estimates (Hollmann et al., 2000). According to
WP, linking county-level growth projections together and constraining to a national-level total growth
avoids potential errors introduced by forecasting each county independently. County projections are
developed in a four-stage process:

1.	National-level variables such as income, employment, and populations are forecasted.

2.	Employment projections are made for 179 economic areas defined by the Bureau of Economic
Analysis (U.S. BEA, 2004), using an "export-base" approach, which relies on linking industrial-
sector production of non-locally consumed production items, such as outputs from mining,
agriculture, and manufacturing with the national economy. The export-based approach requires
estimation of demand equations or calculation of historical growth rates for output and
employment by sector.

3.	Population is projected for each economic area based on net migration rates derived from
employment opportunities and following a cohort-component method based on fertility and
mortality in each area.

4.	Employment and population projections are repeated for counties, using the economic region
totals as bounds. The age, sex, and race distributions for each region or county are determined
by aging the population by single year of age by sex and race for each year through 2050 based
on historical rates of mortality, fertility, and migration.

77


-------
5 Health Endpoint Valuation

To directly compare benefits estimates associated with a rulemaking to cost estimates, the number of
instances of each air pollution-attributable health impact must be converted to a monetary value. This
requires a valuation estimate for each unique health endpoint, and potentially also discounting if the
benefits are expected to accrue over more than a single year, as recommended by the U.S. EPA, 2014.

As reductions in ambient concentrations of air pollution generally lower the risk of future adverse health
effects by a small amount for a large population, the most appropriate economic measure is the ex ante
(before the effect has occurred) willingness-to-pay (WTP) for changes in risk. WTP values are calculated
by dividing the monetary value an individual is willing to pay for a specific risk reduction by that change
in risk.62 Using this approach, the size of the affected population is automatically taken into account by
the number of incidences predicted by epidemiological studies applied to the relevant population.

There are three primary components of the value to society of an individual's avoidance of a non-fatal
illness: 1) medical costs, 2) lost productivity, and 3) impacts on quality of life (i.e., "pain and suffering").
Estimates of individual WTP are conventionally thought to reflect all three of these components and are
the preferred welfare valuation measure.63 However, WTP values are available for a very limited subset
of health endpoints, such as mortality.64

For health endpoints where WTP estimates are not available, such as hospital admissions, we instead
use the cost of treating or mitigating the effect to estimate the economic value. Cost-of-illness (COI)
estimates are generally considered to be a lower bound estimate of the true value of reducing the risk of
a health effect because they reflect the direct expenditures related to treatment and in some cases
costs such as associated productivity losses, but not the value of avoided pain and suffering (Berger et
al., 1987, Harrington and Portney, 1987, U.S. EPA, 2014). Additionally, COI estimates require additional
parsing of individual health endpoints. For example, a stroke may initially involve an emergency
department visit and hospitalization, but will also likely include additional follow-up medical costs, such
as doctor visits and medications.

To prevent double counting of health impacts, when estimating monetary valuations, health endpoints
are separated into the following non-overlapping categories: mortality (section 5.1), hospital
admissions, emergency department visits (section 5.2), and health endpoint onset/occurrence (section
5.3).

EPA develops valuation estimates at the most age-refined level feasible for each health endpoint
assessed. While we focused on identifying valuation estimates from peer-reviewed and published
literature for new health endpoints, we were also able to update several valuation estimates for
endpoints evaluated in previous benefits analyses, such as stroke, cardiac arrest, and AMIs. New
hospitalizations and emergency department visits health endpoint valuations reflect specific ICD-9 codes

62	For example, suppose a measure is able to reduce the risk of mortality from 2 in 10,000 to 1 in 10,000 (a
reduction of 1 in 10,000). If individual WTP for this risk reduction is $100, then the WTP for an avoided statistical
mortality amounts to $1 million ($100/0.0001 change in risk).

63	WTP estimates may not fully account for medical costs or lost productivity if individuals assume some related
costs would be borne by others (e.g., health insurance providers and employers).

64	Economic theory also argues that WTP for most goods (such as environmental protection) will increase if real
income increases.

78


-------
evaluated by the epidemiologic study. New onset or follow-up/management health endpoints reflect
WTP or COI valuation estimates that exclude death and initial emergency department and
hospitalization costs.

These COI measures represent an update to EPA's previous method to producing COI estimates in three
important respects (U.S. EPA, 2018):

•	Estimates are of the costs of medical treatment, rather than charges by medical providers.

•	Sampling parameters are used in survey data to express statistical uncertainty in mean cost
estimates.

•	More recent data is being used to reflect current treatment and healthcare costs.

When multiple valuation studies are available, the strengths and limitations of each study are
considered, in a manner similar to that described for epidemiologic studies (section 2.1). The criteria for
evaluation of these studies are listed in Table 20. In some cases, judgment is required to identify studies
for valuation estimates when a similar number of preferred attributes are satisfied by multiple studies.

Table 20. Cost of Illness Economic Study Identification Consideration Factors

Criteria3

Prioritization Detail (In order of most to least preferred)

Peer-Reviewed
Research

Peer-reviewed and published literature only

Endpoint Definition

1.	ICD codes align with the epidemiological study

2.	ICD codes overlap with the epidemiological study

Population
Attributes

Prefer studies that match epidemiological study's population (specifically by
age)

Study Period

More recent data are preferred

Measure of Costsb

1.	Total payments

2.	Allowable charges

3.	Cost-adjusted charges

4.	Unadjusted charges

Study Location

1.	Nationwide coverage

2.	Multi-city and/or multi-state coverage

3.	Local study population

Coverage of cost
elements

Studies that account for more cost elements (e.g., treatment settings) and
longer time horizons are preferred

Study Size

Larger study size preferred

a This table focuses on COI because WTP measures are not currently available for the health endpoints of interest. Had WTP
estimates been available, additional criteria would be relevant. It also excludes valuation estimates of hospitalizations and
emergency department visits, which are developed by EPA and described in the appendices to U.S. EPA, 2018.
bOnukwugha et al., 2016 provides more information on these methods.

We provide unit values for health endpoints (along with information on the distribution of the unit
value) in Table 21. All values are in constant year 2015$, adjusted for growth in real income for WTP
estimates out to 2024 using projections provided by Standard and Poor's, which is discussed in further
detail below. Additional detail regarding the development of each health endpoint valuation is also
provided below.

79


-------
Table 21. Unit Values for Economic Valuation of Health Endpoints (2015$)1

Health Endpoint

Type

Central Estimate
of Value Per
Statistical
Incidence (2015$)

Source

Mortality

Value of

3%: $7,800,000

Weibull distribution fitted to 26 published



Statistical

7%: $7,100,000

VSL estimates (5 contingent valuation and 21



Life (VSL)



labor market studies). Underlying studies,
distribution parameters, and other
information are available in Appendix B of the
EPA's Guidelines for Preparing Economic
Analyses (U.S. EPA, 2014). Adjusted for
income growth appropriate to the year of
analysis.

Hospitalizations

Medical
costs and
opportunity
cost of time

Varies by ICD
codes, ranging
between $7,700
and $16,000

HCUP data (details available in section 3.2)

Emergency

Medical

Varies by ICD

HCUP data (details available in section 3.3)

Department

costs

codes, ranging



Visits



between $600
and $1,200



Nonfatal

3-year

3%: $49,000

O'Sullivan et al., 2011

Myocardial

medical

7%: $48,000



Infarction (AMI)3

costsb





Asthma

Medical

$0.35 per

Average prescription costs derived from

Symptom-

costs

albuterol inhaler

Epocrates.com and Goodrx.com accessed

Albuterol Usec



puff

March 19, 2020

Asthma

WTP for 1

$219

Dickie and Messman, 2004

Symptom- Chest

symptom





Tightness, Cough,

day





Shortness of







Breath, or







Wheeze







Asthma Onsetc

Lifetime
medical
costs and
lost

productivity

3%: $17,000
7%: $10,000

Belova et al., 2020

Allergic Rhinitis0

1-year

medical

costs

$600

Soni, 2008

Cardiac Arrestc

3-year

medical

costs

3%: $36,000
7%: $35,000

O'Sullivan et al., 2011

80


-------
Lung Cancerc

5-year

medical

costs

3%: $34,000
7%: $33,000

Kaye et al., 2018

Strokec

1-year

medical

costs3

$34,000

Mu et al., 2017

Work Loss Days

Median
daily wage

U.S. median: $150

lEc, 1993

School Loss Days

Lost

productivity
of parent

$106

US Bureau of Labor Statistics, 2015

Minor Restricted-
Activity Days

Median
WTP

$70

lEc, 1993

3%- three percent real discount rate; 7%- seven percent real discount rate (OMB, 2003); All estimates rounded to two
significant figures.

Valuation estimate has been updated to reflect recent literature.

bExcludes initial emergency department and hospitalization costs, which are captured separately.

Valuation estimate is for a new health endpoint.

5.1 Mortality

Following the advice of the SAB's Environmental Economics Advisory Committee (SAB-EEAC), the EPA
currently uses the value of statistical life (VSL) approach in calculating the core estimate of mortality
benefits, because we believe this calculation provides the most reasonable single estimate of an
individual's willingness to trade money for reductions in mortality risk (Stavins, 2000). The VSL approach
is a summary measure for the value of small changes in mortality risk experienced by a large number of
people.

5.1.1 Value of a Statistical Life (VSL)

The current undiscounted VSL used by EPA is $8.7 million (2015$), or $7.8 million (2015$) using a 3%
discount rate and $7.1 million (2015$) using a 7% discount rate (U.S. EPA, 2014). This estimate is the
mean of a distribution fitted to 26 VSL estimates that appear in the economics literature and that have
been identified in the Section 812 Reports to Congress as "applicable to policy analysis" (U.S. EPA,
2011a). It is a value EPA uses in RIAs as well as in the Section 812 Retrospective and Prospective
Analyses of the Clean Air Act (U.S. EPA, 2011a).

The VSL approach mirrors that of Viscusi, 1992 and uses the same criteria as in his review of value of
statistical life studies. The $8.7 million estimate is consistent with the conclusions of Viscusi, 1992
(updated to 2015$) that "most of the reasonable estimates of the value of life are clustered in the $5.2
to $12.3 million range." Five of the 26 studies are contingent valuation studies, which directly solicit
WTP information from subjects; the rest are wage-risk studies, which base WTP estimates on estimates
of the additional compensation demanded in the labor market for riskier jobs. Because this VSL-based
unit value does not distinguish among people based on the age at their death or the quality of their
lives, it can be applied to all deaths. Table 22 presents the central unit value from the 26 value of
statistical life studies and their underlying distribution.

81


-------
Table 22. Central Unit Value for VSL based on 26-value-of-life studies

Basis for Estimate

Age Range at
Death

Unit Value
(VSL) (2015$)

Distribution of
Unit Value

Parameters of
Distribution

Min

Max

PI

P2

VSL, based on 26
value of statistical life
studies

0

99

8,705,114

Weibull

9,648,168

1.509588

5.2 Hospitalizations and Emergency Department Visits

To value hospitalizations, emergency room visits we develop primary COI estimates using data from the
Healthcare Cost and Utilization Project (HCUP). The 2016 National Inpatient Sample (NIS) and
Nationwide Emergency Department Sample (NEDS) provide recent, nationally representative
information on medical treatment in hospitals and emergency departments. In the case of hospital
admissions, valuation estimates are calculated as a combination of medical costs and the opportunity
cost of time spent at the hospital, measured by lost wages during the hospital stay. In the case of
emergency department visits, valuation estimates include only the medical costs. These cost
components are summarized in Table 23.

Table 23. Hospitalization and Emergency Department Cost Elements by Endpoint

Endpoint

Medical Costs
(Emergency Room)

Medical Costs
(Hospital)

Lost Productivity

Hospitalizations



~

~

Emergency department visits

~





Emergency hospitalizations

~

~

~

The NIS and NEDS datasets include discharge-level observations. That is, each data point represents one
individual being discharged from the hospital (NIS) or emergency department (NEDS). Because
individuals are treated in these settings for a variety of reasons, we use medical billing codes to extract
observations related to each health endpoint. The epidemiological studies described above provide ICD-

9	codes for each illness; however, recent HCUP datasets (including NIS and NEDS) use ICD-10 codes.
Thus, we first crosswalk the relevant ICD-9 codes to associated ICD-10 codes using a mapping provided
by the U.S. Centers for Disease Control.65 We then identify all discharges in the HCUP datasets with ICD-

10	codes that match to a study's ICD-9 code(s).66 Because HCUP datasets often include multiple ICD-10
codes for each discharge, we focus on the principal diagnosis (i.e., the first-listed ICD-10 code). Other
key variables used from HCUP include total charges, cost-to-charge ratio (NIS), and length of stay (NIS).

In the NIS dataset, we convert total charges (i.e., the amount billed to patients, employers, or insurance
providers) into estimates of total costs (i.e., the final reimbursements for medical treatment).
Unadjusted charges are not suitable for use in regulatory analysis because posted prices generally do

65	General Equivalence Mapping Files, FY 2016 release of ICD-10-CM. https://www.cdc.gov/nchs/icd/icdlOcm.htm.

66	For emergency hospitalizations, we further restrict the sample to (1) hospitalizations designated as "emergency"
and (2) emergency department visits that result in hospitalization.

82


-------
not reflect actual medical costs due, in part, to negotiation between medical providers and payers (e.g.,
insurance companies). We assume that adjusted charges reflect the actual revenue the hospital receives
and thus the actual cost of providing care. This conversion is completed using hospital-specific cost-to-
charge (CCR) ratios provided with NIS. Because CCRs are not available for NEDS, we apply average CCRs
for each endpoint in NIS to the same set of ICD-10 codes in NEDS.

For each health endpoint, mean estimates are calculated using estimation commands for survey data to
account for the sampling design and sample discharge weights of the HCUP data. This results in
estimates of mean costs and a 95% confidence interval, which represents uncertainty in our valuation
estimates of medical costs. The resulting estimates are presented in Table 24. Confidence intervals for
length of stay cannot be accounted for in the valuation methodology because the EPA's current tool is
only capable of reflecting uncertainty in one parameter.

83


-------
Table 24. Medical Costs and Hospital Stay Data from the HCUP Database

Endpoint

Ages

Epidemiologic
Study Author

ICD-9
Codes

Length of Stay in
Days (95% CI)1

Medical Costs
(95% CI) (2015$)2

HA,

0-99

Jones et al., 2015

491, 492,

3.86 (3.82, 3.90)

$7,676 ($7,574,

Respiratory-1





493, 496



$7,778)

HA,

65-99

Bell et al., 2015

490-492,

4.66 (4.62, 4.69)

$9,004 ($8,894,

Respiratory-2





464-466,
480-487,
493



$9,113)

HA, All

0 -18

Ostro et al., 2009

460-519

3.50 (3.37, 3.62)

$9,075 ($8,282,

Respiratory









$9,868)

HA, All Cardiac

0-99

Talbott et al., 2014

390-459

5.05 (5.00, 5.11)

$16,045 ($15,721,

Outcomes









$16,368)

HA,

65-99

Kioumourtzoglou

331.0

7.95 (7.70, 8.21)

$10,696 ($10,400,

Alzheimer's



et al., 2016





$10,992)

Disease











HA, Cardio-,

65-99

Bell et al., 2015

426-

4.82 (4.78, 4.87)

$14,665 ($14,434,

Cerebro- and





427,428,4



$14,896)

Peripheral





30-438,





Vascular





410-414,





Disease





429, 440-
448





ED, Respiratory

0-99

Krall et al., 2013

480-486,
491, 492,
496, 460-
465, 466,
477, 493,
786.07



$875 ($826, $923)

ED, All Cardiac

0-99

Ostro et al., 2016

390-459

-

$1,161 ($1,112,

Outcomes









$1,210)

ED, Respiratory

0-99

Barry et al., 2019

480-486,
491, 492,
496, 460-
465, 466,
477, 493,
786.07



$875 ($826, $923)

Confidence intervals (CIs) associated with the length of hospital stay are presented for information only
and are not used in analyses due to technical limitations. Importantly, the length of stay is a factor in the
overall COI estimate.

2Medical costs reflect the expenditures per treatment episode/event (e.g., per hospitalization) and
confidence intervals (CIs) reflect the 95% CI around the population mean value and not that 95% of
patients observe costs within these bounds. Does not include productivity losses.

84


-------
5.3 Health Endpoint Onset/Occurrence

Monetary valuation estimates for health endpoint onset or occurrences are described below, listed in
alphabetical order. Onset indicates the development of a health endpoint (e.g., asthma diagnosis),
whereas occurrence refers to an instance of that health endpoint (e.g., asthma attack).

5.3.1 Acute Myocardial Infarctions (AMIs)

Economic values for acute myocardial infarctions (AMIs, also known as heart attacks) have been
updated to be derived from O'Sullivan et al., 2011, which estimate three-year medical costs associated
with cardiovascular disease events among adults ages 35 and older in the U.S. The authors rely on
administrative claims data from a large U.S. health plan and develop econometric models to estimate
medical costs for 15 different cardiovascular events, including AMIs. The dataset includes over 20 million
commercial and Medical Advantage members between 2002 and 2006. AMIs are identified using the
ICD-9 code 410. The authors use propensity score matching to develop a control group with which to
compare costs versus individuals that suffered AMIs. We exclude medical costs within the month of the
event in an attempt avoid double counting hospitalization costs, which are captured separately in the
hospitalization valuation endpoints. Over three years, the total medical costs, excluding hospitalization,
are $49,758 (undiscounted, inflated to 2015$), or $48,796 using a 3% discount rate and $47,623 for a 7%
discount rate (Table 25). Although this study analyzed costs associated with individuals ages 35 and
older, we apply the total medical costs to all ages from zero to 99 since only a small portion (<10%) of
annual AMI incidence occurs in the age range below 35.

Table 25. Medical Costs for AMIs (2015$)

Costs

Cumulative Costs

Annual Costs

Undiscounted

3% Discount Rate1

7% Discount Rate1

Month of Event*

$43,523

$43,523

$43,523

$43,523

Year 1

$70,629

$27,106

$27,106

$27,106

Year 2

$82,591

$11,962

$11,614

$11,180

Year 3

$93,281

$10,690

$10,076

$9,337

Years 1-3

$93,281

$49,758

$48,796

$47,623

1Uses end-of-year discounting.

We supplement AMI medical costs with estimates of lost earnings using age-specific estimates from
Cropper and Krupnick, 1990. Using a 3% discount rate, we estimated the following present discounted
values in lost earnings over 5 years due to a heart attack: 0.219 times annual earnings for someone
between the ages of 25 and 44, 3.534 times annual earnings for someone between the ages of 45 and
54, and 1.245 times annual earnings for someone between the ages of 55 and 65. The corresponding
age-specific estimates of lost earnings using a 7% discount rate are 0.203, 3.287, and 1.158 times annual
earnings, respectively. Cropper and Krupnick, 1990 does not provide lost earnings estimates for
populations under 25 or over 65. As such we do not include lost earnings in the cost estimates for these
age groups. These costs, along with the total valuation estimates for AMIs, are presented in Table 26.

85


-------
Table 26. Total Valuation Estimates for AMIs (2015$)

Discount Rate

Age F
Min

lange
Max

Medical Cost

Lost Earnings Multiplier

Total Cost

3%

0

24

$48,796

0

$48,796

25

44

$48,796

0.219

$48,796+ 0.219*earnings

45

54

$48,796

3.534

$48,796+ 3.534*earnings

55

65

$48,796

1.245

$48,796+ 1.245*earnings

66

99

$48,796

0

$48,796

7%

0

24

$47,623

0

$47,623

25

44

$47,623

0.203

$47,623+0.203*earnings

45

54

$47,623

3.287

$47,623+3.287*earnings

55

65

$47,623

1.158

$47,623+ 1.158*earnings

66

99

$47,623

0

$47,623

5.3.2	Allergic Rhinitis (Hay Fever)

Two potential valuation sources for allergic rhinitis were reviewed: Soni, 2008 and Bhattacharyya, 2011.
Both studies utilize data from the Medical Expenditure Panel Survey (MEPS) and identify allergic rhinitis
(also referred to as hay fever) using ICD-9 code 477. Each study analyzes medical expenditures for
differing years, Soni, 2008 for the years 2000 and 2005, and Bhattacharyya, 2011 for the year 2007. Soni,
2008 calculates the cost-of-illness for allergic rhinitis as the mean expenditures for ambulatory care, in-
patient services, and prescription medications per person. Bhattacharyya, 2011 calculates the
incremental difference in annual healthcare expenditures for individuals with and without allergic
rhinitis. Although Bhattacharyya, 2011 uses more recent data, the estimates are not specific to children.
Therefore, we derived our COI estimates from the 2005 data presented by Soni, 2008, which are
stratified by age group. The resulting COI for allergic rhinitis is $600 for ages zero to seventeen (2015$;
Table 21). These COI estimates represent mean annual medical costs for patients with hay fever. Given
that the health impact function for this endpoint relates to allergic rhinitis prevalence, these estimates
are more applicable than values representing only first-year costs.

5.3.3	Asthma Onset

Belova et al., 2020 estimated the lifetime cost of asthma using data from the 2002 to 2010 Medical
Expenditure Panel Survey (MEPS). The authors identify all individuals with current asthma (9,409 out of
158,867 respondents) using the ICD-9 code 493 in the MEPS Medical Conditions Files. Additionally, they
identify the date of asthma onset for these individuals. Using the MEPS Medical Events files, which
capture most types of medical expenditures (e.g., hospitalizations, emergency room visits, outpatient
visits, prescriptions), Belova et al., 2020 estimated annual expenditures by asthma duration and age at
onset. The annual healthcare costs for asthma—as measured by healthcare expenditures by all paying
parties—vary from $700 to $1,800 for children and $800 to $2,200 for adults (2010$). They extrapolate
these values to a lifetime cost stream for an incident chronic asthma case to generate present value
estimates by onset age using discount rates of 3% and 7%. Additionally, the authors consider
productivity impacts that capture 1) the probability of not being able to work due to health reasons, 2)
the impact of asthma on occupational choice, and 3) impact of asthma on weekly earnings.

86


-------
We adapt the Belova et al., 2020 estimates to align with the age groups 0 to 17, 4 to 21, and 35 to 99.67
This calculation entails weighting the Belova et al., 2020 age groups by their relative prevalence and
propagating the standard errors to derive new uncertainty bounds. The results are summarized in Table
27. Confidence intervals are not provided for productivity losses to mirror the valuation functions in
BenMAP-CE, which at present are only capable of reflecting uncertainty in one parameter (in this case,
medical costs) (Table 21).

Table 27. Age-adjusted Belova et al., 2020 Estimates of Lifetime Asthma Costs

Age of asthma onset

Discount rate

Healthcare costs (2015$)

Productivity Loss (2015$)

0-17

3%

$17,232
($16,366, $18,097)

$27,426

0-17

7%

$10,187
($9,643, $10,730)

$17,502

5,3.4 Asthma Symptoms/Exacerbation

5.3.4.1	Albuterol Use

As albuterol use is a new measure of PIVh.s-attributable asthma symptoms, we developed a method for
valuing this health endpoint. We estimate the economic value for albuterol use associated with asthma
symptoms using prescription prices for albuterol inhalers. Epocrates and GoodRx provide cost and
actuation information for four common types of albuterol inhalers in 2020 dollars.68,69 Both online
resources utilize published price lists, purchases, claim records, and pharmaceutical data to provide
clinical statistics. Epocrates and the FDA provide cost and actuation information for one additional, less
common, albuterol inhaler.70 We divide the cost of inhalers by the actuations per inhaler to calculate an
average cost per actuation across all inhaler types. We then adjust the values to 2015$ using the
Consumer Price Index (CPI) for medical care. Since medical cost index data were unavailable for 2020 at
the time of these calculations, we used the most recently available index (2019). The resulting value for
asthma symptoms, albuterol use is $0.35 per actuation (2015$) (Table 21).

5.3.4.2	Cough, Wheeze, Chest Tightness, and Shortness of Breath

While the risk estimates for both PM2.5- and 03-attributable asthma symptoms were updated, the
valuation estimates for cough, wheeze, chest tightness, and shortness of breath are still based on the
previous method, using the Dickie and Messman, 2004 analysis of parents' WTP to relieve asthma
symptoms in children and adults. The authors derive the WTP estimates from an attribute-based,
stated-choice question assessing preferences to avoid acute illness as part of a survey performed in
Hattiesburg, Mississippi in 2000. Survey respondents are asked to identify whether they or their child
have experienced the following asthma symptoms in the past year: cough with phlegm, shortness of
breath with wheezing, chest pain on deep inspiration, and fever with muscle pain and fatigue.

67	These age groups were selected based on the ages pertaining to the PIVh.s-related health impact functions.
These do not currently align directly with the ozone health impact functions for new onset asthma, but the
valuation functions nonetheless cover the age ranges needed to value the ozone health impact functions.

68	https://online.epocrates.com/drugs searched March 19th, 2020.

69	https://www.goodrx.com/albuterol searched March 19th, 2020.

70	https://www.accessdata.fda.gov/drugsatfda_docs/label/2016/205636s006lbl.pdf

87


-------
Respondents were then assigned one of sixteen illness profiles varying by symptom, symptom duration,
in days, as well as discomfort level. Dickie and Messman, 2004 calculate the WTP for children ages zero
to seventeen as $219, for one avoided mild symptom-day (2015$). The authors also provide WTP
estimates by symptom, however, they represent six avoided symptom-days. Therefore, we apply the
same WTP value, for one avoided mild symptom-day, to each asthma symptom endpoint (Table 21).

5.3.5 Cardiac Arrest

The COI for cardiac arrests occurring outside of the hospital is derived from O'Sullivan et al., 2011, who
estimate three-year medical costs associated with cardiovascular disease events among adults ages 35
and older in the U.S. The authors rely on administrative claims data from a large U.S. health plan and
develop econometric models to predict medical costs for 15 different cardiovascular events, including
cardiac arrest, referred to as resuscitated cardiac arrest. The dataset includes over 20 million
commercial and Medical Advantage members between 2002 and 2006. Cardiac arrests are identified
using the ICD-9 code 427.5. The authors use propensity score matching to develop a control group with
which to compare costs versus individuals that suffered cardiac arrest. Medical costs occurring within
the month of the event were excluded to avoid double counting hospitalization costs, which are
separately captured by the hospitalization valuation functions. Over three years, the total medical costs,
excluding hospitalization, are $36,142 (undiscounted, inflated to 2015$), or $35,753 using a 3% discount
rate and $35,282 for a 7% discount rate (Table 28 and Table 21).

Table 28. Valuation Estimate for Cardiac Arrests (2015$)

Costs

Cumulative Costs

Annual Costs

Undiscounted

3% Discount Rate

7% Discount Rate

Month of Event*

$43,904

$43,904

$43,904

$43,904

Year 1

$71,901

$27,997

$27,997

$27,997

Year 2

$74,701

$2,800

$2,718

$2,617

Year 3

$80,046

$5,345

$5,038

$4,668

Years 1-3

$80,046

$36,142

$35,753

$35,282

5.3.6 Lung Cancer

The unit value for non-fatal lung cancer incidence is derived from the direct medical costs of lung cancer
treatment estimated by Kaye et al., 2018. This COI value incorporates only direct medical costs and not
lost earnings associated with lung cancer incidence because the average age of lung cancer diagnosis is
approximately 70 and it is assumed that those aged 65 and older are retired and thus have exited the
labor market. Lung cancer treatment costs depend to a large extent on the phase of care, with costs in
the initial year of treatment (e.g., $17,422 for males) far exceeding the continuing costs of treatment in
subsequent years (e.g., $3,269 for males). We calculate costs over a five-year span, beginning with the
initial onset which is occurs with a delay after exposure. The specific lag periods between exposure and
onset are discussed in Section 6.4.2. The initial year's treatment cost is summed with four years of
continuing annual costs discounted by 3% and 7%.

Furthermore, Kaye et al., 2018 provides separate treatment cost estimates for men and women. The
distribution of new lung cancer cases by sex in the United States from Siegel et al., 2019 is

88


-------
approximately 51% male and 49% female. This distribution of new lung cancer cases was used to weight
the sex-specific cost estimates from Kaye et al., 2018 to obtain a combined five-year cost estimate for
both sexes. In order to adjust the cost estimate to 2015$ using a medical cost index, we assume that
costs presented by Kaye et al., 2018 are in 2010$ as an approximate midpoint of the data years 2007-
2012. Altogether, the cost of non-fatal lung cancer incidence over a five-year period is estimated to be
$33,809 using a 3% discount rate or $32,548 using a 7% discount rate (Table 21).

For an outcome such as lung cancer, there is an expected time lag between changes in pollutant
exposure in a given year and the total realization of health effect benefits, commonly referred to in
regulatory analyses as the "cessation lag." The time between exposure and diagnosis can be quite long,
on the order of years to decades, to realize the full benefits of the air quality improvements. This latency
period is important in order to properly discount the economic value of these health benefits.

To estimate the latency period, we performed a literature search using the keywords "non-fatal lung
cancer," "lung cancer," "PM2.5," "latency," and "incidence." Five papers that estimate the risk of lung
cancer incidence from PM2.5 exposure using a latency period were identified. The latency period length
and country of the identified papers are summarized in Table 29. Based on estimates of lung cancer
latency from the literature, 10 years was the most common latency period estimate found in the
literature (i.e., the mode).

Table 29. Latency Periods Used in Lung Cancer Risk Assessment Papers

Study

Latency Period (years)

Location

Gogna et al., 2019

5

Canada

Bai et al., 2020

4; 10

Canada

Kulhanova et al., 2018

10

France

Coleman et al., 2020

10; 15

US

Harrison et al., 2004

20

US

To account for the latency period between air pollution reductions and avoided lung cancer diagnoses in
our economic valuation estimates, we developed an age-at-diagnosis cessation lag distribution method
based on an approach previously used to estimate avoided cases of kidney cancer in analyses of water
quality rules (U.S. EPA, 2017). The method uses lung and bronchus cancer diagnosis age-distribution
from the Surveillance, Epidemiology, and End Results Program (SEER). For this model, we assumed that
the case reduction distribution would follow the age-pattern of cancer diagnosis between the age at
which the exposure change occurs and 99 years. Table 30 shows an example case reduction distribution
calculation for an exposure change experienced at 55. SEER estimates 92.2% of lung and bronchus
cancer cases occur in individuals 55 years and older. Dividing the percentages in the remaining age bins
by 92.2% (the percent of lung and bronchus cancer diagnoses between the age of exposure change and
end of lifetime), we find that there is a 24% chance that the risk reductions for a 55-year-old occur
between ages 55 and 64, a 37% chance that the case reductions occurs between ages 65 and 74, etc. For
distributing avoided cases within an age bin, we assume an equal incidence distribution across years
within each bin.

89


-------
Table 30. Percent Lung and Bronchus Cancer Incidence by Age and Distribution of Risk Reduction by Age
for an Exposure Change at 55

Age
Group

Percent New Cases per Year by
Age*

Percent of New Cases Occurring at or After Age

551

0-20

0

NA

20-34

0.2

NA

35-44

0.9

NA

45-54

6.6

NA

55-64

21.8

24

65-74

34.1

37

75-84

26.6

29

85-99

9.7

11

55-99

92.2

100

*May not sum to 100% due to rounding

1Calulcated as the percentage in column 2 divided by 92.2%, where 92.2% is the percentage of lung and bronchus
incidence between age 55 and 99.

This and other potential cessation lag distribution models for lung cancer onset are described and
compared in section 6.4.2.

5.3,7 Minor Restricted Activity Days (MRADs)

Due to their definition, for the purposes of benefits estimation minor respiratory-restricted activity days
(MRRAD) are assumed to constitute all MRADs (Ostro and Rothschild, 1989). While no peer-reviewed
studies estimating WTP to avoid a MRRAD are available, a central estimate and upper and lower bounds
of WTP to avoid a MRRAD were developed by lEc (lEc, 1993).71 When estimating benefits associated
with an MRAD, we use a triangular distribution centered at the estimate.

Any estimate of mean WTP to avoid a MRRAD (or any other type of restricted activity day other than
Work Loss Day (WLD)) will be somewhat arbitrary because the endpoint itself is not precisely defined.
Many different combinations of symptoms could presumably result in some minor or less minor
restriction in activity. Krupnick and Cropper, 1992 argued that mild symptoms will not be sufficient to
result in a MRRAD, so that WTP to avoid a MRRAD should exceed WTP to avoid any single mild
symptom. A single severe symptom or a combination of symptoms could, however, be sufficient to
restrict activity. Therefore, WTP to avoid a MRRAD should, these authors argue, not necessarily exceed
WTP to avoid a single severe symptom or a combination of symptoms. The "severity" of a symptom,
however, is similarly not precisely defined; moreover, one level of severity of a symptom could induce
restriction of activity for one individual while not doing so for another. The same is true for any
combination of symptoms.

Given that there is inherently a substantial degree of arbitrariness in any point estimate of WTP to avoid
a MRRAD (or other kinds of restricted activity days), the reasonable bounds on such an estimate must be

71 lEc, 1993 derived this estimate of WTP to avoid a MRRAD using WTP estimates from Tolley et al., 1986
for avoiding a three-symptom combination of coughing, throat congestion, and sinusitis. This estimate
of WTP to avoid a MRRAD, so defined, is $69.58 in 2015$.

90


-------
considered. By definition, a MRRAD does not result in loss of work. WTP to avoid a MRRAD should
therefore be less than WTP to avoid a WLD. At the other extreme, WTP to avoid a MRRAD should exceed
WTP to avoid a single mild symptom. The highest lEc midrange estimate of WTP to avoid a single
symptom is $28.51 (2015$), for eye irritation. The point estimate of WTP to avoid a WLD in the benefit
analysis is $110.62 (2015$). If all the single symptoms evaluated by the studies are not severe, then the
estimate of WTP to avoid a MRRAD should be somewhere between $28.51 and $110.62. Because the lEc
estimate of $69.58 (2015$) falls within this range (and acknowledging the degree of arbitrariness
associated with any estimate within this range), we use the lEc estimate of $69.58 (2015$) (Table 21).

5.3.8	School Loss Days

There is currently one unit value available in BenMAP for school loss days, based on (1) the probability
that, if a school child stays home from school, a parent will have to stay home from work to care for the
child, and (2) the value of the parent's lost productivity. We first estimated the proportion of families
with school-age children in which both parents work, and then valued a school loss day as the
probability of a work loss day resulting from a school loss day (i.e., the proportion of households with
school-age children in which both parents work) times a measure of lost wages.

From the U.S. Bureau of Labor Statistics (2015) we obtained the rate of participation in the workforce of
women with children under 18 years of age. We multiplied this rate (69.9%) by the estimated daily lost
wage (if a mother must stay at home with a sick child), based on the median full-time weekly wage
among women 25 and older in 2015.72 This median weekly wage is $759 (2015$).73 Dividing by five work
days per week gives an estimated median daily wage of $152. The expected loss in wages due to a day
of school absence in which the mother would have to stay home with her child is estimated as the
probability that the mother is in the workforce times the daily wage she would lose if she missed a day =
69.9% of $152, or $106. We currently have insufficient information to characterize the uncertainty
surrounding this estimate.

A unit value based on the approach described above is likely to understate the value of a school loss day
in four ways. First, it omits WTP to avoid the symptoms/illness which resulted in the school absence.
Second, it effectively gives zero value to school absences which do not result in a work loss day. Third,
the approach may use a wage rate that is too low by assuming that men do not stay at home with sick
children. Fourth, does not account for deleterious effects on student learning and subsequent utility or
productivity. The unit value of $106 is therefore considered an "interim" value until such time as
alternative means of estimating this unit value become available (Table 22).

5.3.9	Stroke

Mu et al., 2017 estimates COI of non-fatal stroke incidence using direct medical costs incurred during
initial hospitalization and the 360 days following hospital discharge. The study identifies individuals
experiencing a first-time stroke using ICD-9 codes 434 and 436. The authors analyze medical claims from
January 2006 to March 2015 utilizing the retrospective IMS LifeLink PharMetrics Plus database for
individuals ages 18 to 65, and Medicare Advantage and Medicare Supplemental Claims for individuals
above the age of 65. The authors present acute care and long-term care costs stratified by three

72	Does not include benefits rate for lost work time.

73	2015 median wages were the most recently available data at the time of update. However, many valuation
estimates account for income growth, approximating 2020 wages.

91


-------
discharge classifications: dead at discharge, discharged with disability, and discharged without disability.
We estimate the average costs for non-fatal cases by weighting the costs for individuals discharged with
disability and without disability by their prevalence (23 and 77 percent, respectively). The resulting COI
for non-fatal stroke incidence is $33,962 (2015$) (Table 21). This value reflects one-year medical costs
following stroke and does not include hospitalization costs, as these costs are separately captured by
hospitalization valuation functions. We reviewed several studies that estimate longer-term medical
costs (Goodwin et al., 2011, Lee et al., 2007, Luengo-Fernandez et al., 2012, Nicholson et al., 2016) and
concluded that roughly three quarters of costs are incurred in first year after stroke occurrence.74

5.3.10 Work Loss Days (WLDs)

Work loss days are valued at a day's wage. BenMAP calculates county-specific median daily wages from
county-specific annual wages by dividing by (52*5), on the theory that a worker's vacation days are
valued at the same daily rate as workdays. This estimate does not include benefits rate for lost work
time. The resulting COI for work loss days varies by county, but has a median value of $150 (2015$) (lEc,
1993).(Table 21).

5.4 Developing Income Growth Adjustment Factors for Health Endpoint
Onset/Occurrence

Chapter 4 of the BenMAP-CE User Manual provides instructions for formatting and adding income
growth data (U.S. EPA, 2018). These values are used to adjust WTP estimates for growth in real income.
As discussed in that chapter, evidence and theory suggest that WTP should increase as real income
increases. When reviewing the economic literature to develop income growth adjustment factors, it is
important to have an economist assist. For an overview of valuation, see Chapter 7 of the BenMAP-CE
User Manual, "Aggregating, Pooling, and Valuing".

Adjusting WTP to reflect growth in real income requires three steps:

1.	Identify relevant income elasticity estimates from the peer-reviewed literature.

2.	Calculate changes in future income.

3.	Calculate adjustments to WTP based on changes in future income and income
elasticity estimates.

1. Identifying income elasticity estimates

Income elasticity estimates relate changes in demand for goods to changes in income. Positive income
elasticity suggests that as income rises, demand for the good also rises. Negative income elasticity
suggests that as income rises, demand for the good falls. We do not adjust COI estimates according to
changes in income elasticity due to the fact that COI estimates the direct cost of a health outcome;
instead we adjust this metric using inflation factors described above. We include income elasticity
estimates specific to the type of health endpoint associated with the WTP estimate for three types of
health effects: minor, severe and mortality. Minor health effects are those of short duration. Severe, or
chronic, health effects are of longer duration. Consistent with economic theory, the peer reviewed

74 We did not include the additional 25% of medical costs incurred after the first year post-stroke due to the lack of
information on the timing of those additional costs. Without information on when they would be incurred we
cannot appropriately discount the estimated medical costs.

92


-------
literature indicates that income elasticity varies according to the severity of the health effect. A review
of the literature revealed a range of income elasticity estimates that varied across the studies and
according to the severity of health effect. Table 31 summarizes the income elasticity estimates for minor
health effect, severe health effect and mortality. Here we have provided a lower, upper, and central
elasticity estimate for each type of health endpoint.

Table 31. Income Elasticity Estimates for Minor Health Effects, Severe Health Effects, and Mortality

Health Endpoint

Lower Bound

Central Estimate

Upper Bound

Minor Health Effect

0.04

0.15

0.30

Severe and Chronic
Health Effects

0.25

0.45

0.60

Mortality

0.08

0.40

1.00

2.	Calculating changes in future income

The next input to the WTP adjustment is annual changes in future income. The Congressional Budget
Office's (2016) ten-year projections of US Gross Domestic Product (GDP) are used to estimate changes in
future income. Historical GDP data came from the U.S. Bureau of Commerce's Bureau of Economic
Analysis. GDP values were adjusted for inflation as needed using the Implicit Price Deflator annual index,
published by the Economic Research Division of the Federal Reserve Bank of St. Louis. We divided the
projected change in GDP by the Woods & Poole (2015) projected change in total US population to
produce an estimate of the future GDP per capita.

3.	Calculating changes in WTP

The income elasticity estimates from Table 31 and the estimated changes in future income may then be
used to estimate changes in future WTP for each health endpoint. The adjustment formula follows four
steps:

93


-------
1)

AWTP

~WTP~	(WTP2 - WTP\) X (J2 + /,)

£ A/	(/2 - /j) x (WTP2 + VKrPi)
/

2) eI2WTP2 + e12WTP1 - shWTP2 - eI1WTP1 = I2WTP2 + I1WTP2 - I2WTP1 - 11WTP1

Table 32 summarizes the income-based WTP adjustments used within BenMAP-CE for minor health
endpoints, severe health endpoints, and premature mortality. BenMAP-CE applies the "mid" income
growth adjustment to the WTP for each corresponding health endpoint. The "low" and "upper" are
provided for bounding the "mid" estimate. More information on the uncertainties associated with the
choice of income elasticity is provided in section 6.4.3.

3)

WTP2 X ( eI2 — eh - I2 - h) = WTP1 X ( eh - eI2 - h ~ h)

4)

WTP2 = WTP\ X

Eh ~ £h ~ h ~ h
eI2 — Eh ~ h~ h

94


-------
Table 32. Income-Based WTP Adjustments by Health Effect and Year

Minor Health Endpoint

Severe Health Endpoint

Mortaiity

Year

Low

Mid

Upper

Low

Mid

Upper

Low

Mid

Upper

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018

2019

2020

2021

2022

2023

2024

2025

2026

1

0.99943614
1.00027895
1.00083518
1.001928926
1.00252676
1.003553152
1.004830718
1.006105304
1.007476926
1.008633733
1.008626103
1.008962274
1.009722471
1.010831594
1.011770725
1.012430668
1.01275146
1.012262344
1.01079917
1..011459589
1.011786222
1.012354255
1.012712717
1.013344169
1.014022827
1.014274955
1.014827728
1.015322924
1.015639186
1.015908599
1.016283393
1.016681671
1.017086029
1.017486334
1.017879009
1.018263578

1

0.997887194
1.0010463
1.003135324
1.007252812
1.009508491
1.013389945
1.018236518
1.023088813
1.028329492
1.032765627
1.032736301
1.034028053
1.036953092
1.041232586
1.044866562
1.047425628
1.048671246
1.04677248
1.041107416
1.043661356
1.044926405
1.047128916
1.048520923
1.050976157
1.053619623
1.054602981
1.056761265
1.058697701
1.059935808
1.060991406
1.062461257
1.064024925
1.065614223
1.067189336
1.068736196
1.070252538

1

0.995778859
1.002093554
1.006280541
1.014558434
1.019107819
1.026960373
1.036808729
1.046717048
1.057473063
1.066622734
1.066562176
1.069233894
1.075297475
1.084201217
1.091792464
1.097154975
1.099770427
1.095784903
1.083940148
1.089271665
1.091917515
1.096532583
1.099454761
1.104618788
1.110193372
1.11227107
1.116838455
1.120945096
1.123574972
1.125819683
1.128949761
1.132284641
1.135679722
1.139050007
1.142365217
1.145620227

1

0.99648118
1.00174439
1.005231023
1,012117267
1.01589787
1.022416711
1.030580044
1.038779855
1.047665954
1.055212379
1.05516243
1.057363987
1.062356591
1.069678664
1.075912952
1.080312252
1.08245647
1.079188704
1.069464207
1.073843718
1.076015592
1.079801798
1.082197666
1.086428881
1.090992332
1.092692018
1.096426606
1.099781871
1.101929545
1.103761792
1.106315613
1.109035254
1.11180234
1.114547729
1.117246628
1.119895339

1

0.993674994
1.003141999
1.009435654
1.021917462
1.028799176
1.040715098
1.055725813
1.070903659
1.087466359
1.101626635
1.101532698
1.105680108
1.115113139
1.129017591
1.140922546
1.149360299
1.153483868
1.147202373
1.12860918
1.136964202
1.141119123
1.148379683
1.15298593
1.161142945
1.169972539
1.173269868
1.180531025
1.187074065
1.191271544
1.194858909
1.199867845
1.205213666
1.210665345
1.216086745
1 221428633
1.226682782

1

0.991575539
1.004191637
1.01260066
1.029330373
1.038584113
1.054657698
1.074997663
1.095668674
1.118347049
1.137836933
1.137707233
1.143433094
1.156486392
1.175803781
1.192415953
1.204231024
1.210017681
1.201206207
1.175235033
1.186884999
1.192690849
1.202856183
1.209318399
1.220786929
1.233237505
1237896562
1.248175263
1.257459402
1.263426423
1.268532872
1.275673389
1.283308029
1.291108131
1.298879623
1.306551456
1.314110994

1

0.998872638
1.000557899
1.001670957
1.003861666
1.005059958
1.00711906
1.00968492
1.012248039
1.01500988
1.017342448
1.01732707
1.018005252
1.019539833
1.021781206
1.023680925
1.025016785
1.025666475
1.024675965
1.021715641
1.023051262
1.023712158
1.024862051
1.025588155
1.026867747
1.02824378
1.028755188
1.029876947
1.030882478
1.031524897
1.032072306
1.032834172
1.03364408
1.034466624
1.035281181
1.036080599
1.036863804

1

0.994375765
1.002792478
1.008382797
1.019458413
1.025558352
1.036109447
1.049381137
1.062778115
1.077371955
1.089827657
1.089745045
1.093389511
1.10167253
1.11386621
1.12429142
1.131672144
1.135276675
1.12978518
1.113508224
1.120826483
1.124463439
1.130814672
1.134841442
1.141967177
1.149673343
1.152549148
1.158878565
1.164577603
1.168231487
1.171352863
1.175709248
1.180356026
1.185091853
1.189798594
1.194433689
1.198989868

1

0.985998511
1.006995797
1.021089673
1.049366593
1065144777
1.092786551
1.128201008
1.164700031
1.205345869
1.24079144
1.240554214
1.251058578
1.275163889
1.31124568
1.342671752
1.36525023
1.376378059
1359451532
1.310176253
1.332167268
1.343194842
1.362613082
1.375030875
1.397211194
1.421498775
1.430643797
1.45092845
1.469381213
1.48130703
1491554141
1.505947828
1.521420956
1.53731966
1.553251266
1.56906867
1.584743023

95


-------
6 Characterizing Uncertainty and Evaluating Sensitivity to Alternate
Assumptions

Complex analyses such as the one presented in the final RIA for the Revised CSAPR Update rule use
many estimated parameters and inputs. The approach for estimating PM2.5 and 03 benefits includes
health effect risk estimates from epidemiologic studies, population data, population growth estimates,
economic data for monetizing benefits, and assumptions regarding the future state of the world (i.e.,
on-the-books regulations). When the uncertainties from each stage of the analysis are compounded,
even small uncertainties can have large effects on the total quantified benefits.

After reviewing the EPA's approach to quantifying benefits, the National Research Council (NRC) (2002,
2008) highlighted the need to conduct rigorous quantitative analyses of uncertainty and to present
benefits estimates to decision makers in ways that foster an appropriate appreciation of their inherent
uncertainty. Since the publication of these reports, the EPA has continued improving its techniques for
characterizing uncertainty in the estimated air pollution-attributable benefits.

In light of these recommendations, we incorporate new quantitative and qualitative characterizations of
uncertainty. Where possible, we quantitatively assess uncertainty in each input parameter (for example,
we characterize statistical uncertainty by performing Monte Carlo simulations). We invest the time and
resources in performing the most comprehensive uncertainty analyses for those input parameters that
most greatly influence on the size of the estimated health impacts.75

In some cases, this type of quantitative analysis is not possible due to lack of data, so we instead
characterize the sensitivity of the results to alternative plausible input parameters. And, for some inputs
into the benefits analysis, such as the air quality data, we lack the data to perform either a quantitative
uncertainty analysis or sensitivity analysis.

Sections 6.1 and 6.2 quantitatively describe the uncertainty associated with estimated PM2.5 and 03-
attributable incidence. Section 6.3 provides information on the sensitivity to more granular baseline
incidence rates. Section 6.4 quantitatively discusses the influence of uncertainty in the economic
valuation functions. Lastly, section 6.5 qualitatively discusses the various potential sources of
uncertainty, sometimes including sources of uncertainty touched upon quantitatively.

6.1 Quantitative Characterization of PM2.5 Uncertainty and Evaluating Sensitivity to
Alternate PM2.5 Assumptions

Below we describe our approach for characterizing uncertainty in the estimated PM2.5-related effects.
We start first with the role of Monte Carlo assessment in generating a quantitative distribution of
results. We next describe how alternative risk estimates76 can be useful for assessing the sensitivity of

75	Uncertainties that we expect will have the greatest influence on health impacts are 1) those associated with
mortality impacts given the severity of the outcome and the associated economic valuation and 2) quantitative
and qualitative uncertainty characteristics likely to most strongly impact the magnitude of bias.

76	Alternate risk estimates are a means to quantitatively understand uncertainties around the main risk estimate.
Alternate risk estimates are based on a different set of input parameters, which may come from the same study or
different studies. Alternate risk estimates can be used to assess the sensitivity of the risk estimate to alternative
assumptions and input parameters, such as modeling choices, populations, or statistical techniques.

96


-------
the estimated PM2.5-related mortality and morbidity to plausible alternative input parameters; this gives
insight to the influence of the functional form of the model or alternative epidemiologic approaches.
Quantitative sensitivity analyses using alternative or additional risk estimates are included for the
following PM2.5-attributable health endpoints: mortality in adults (section 6.1.1), asthma onset in
children (section 6.1.3), cardiovascular hospital admissions (section 6.1.4), and respiratory hospital
admissions (section 6.1.5).

6.1.1	Statistical Uncertainty Around the Risk Estimate (Monte-Carlo Assessment)

For all endpoints analyzed, we use a Monte Carlo simulation in which we sample from the standard
error associated with each risk estimate and present the resulting 2.5th and 97.5th percentile values from
this distribution as a 95th percentile confidence interval around the estimated health impact and
monetized health benefits. Monte Carlo methods are a well-established means of characterizing random
sampling error associated with the risk estimates from epidemiological studies. This approach randomly
samples from a distribution of incidence and valuation estimates to characterize the effects of
uncertainty in those inputs on output variables. The reported standard errors in the epidemiological
studies determined the distributions for individual effect estimates for endpoints estimated using a
single study. The confidence intervals around the monetized benefits incorporate the epidemiology
standard errors as well as the distribution of the valuation function. These confidence intervals do not
reflect other sources of uncertainty inherent within the estimates, such as baseline incidence rates,
populations exposed, and transferability of the effect estimate to diverse locations. As a result, the
reported confidence intervals and range of estimates give an incomplete picture about the overall
uncertainty in the benefits estimates.

6.1.2	Adult All-Cause Mortality77

Two studies of all-cause, long-term PM2.5 exposure and mortality were identified as best characterizing
U.S. risk in adults, Di et al., 2017b and Turner et al., 2016. Additional information regarding the cohort
concentration exposure distributions (section 6.1.2.1) and additional risk estimates potentially providing
insight into the effect of various potential sources of uncertainty, such as different exposure estimation
techniques (section 6.1.2.2), confounding by 03 (section 6.1.2.3), statistical regression techniques and
methods to control for confounders (section 6.1.2.4), and effect modification by individual risk factors
(section 6.1.6).

6.1.2.1 Low Concentration Exposures

Each epidemiological risk estimate is based on a distribution of air quality concentrations experienced by
the original cohort population. As such, it is important to consider the relationship between the
concentrations from which the mortality estimates are derived and the concentrations at which the
estimates are subsequently applied in future policy scenarios in which concentrations are likely to be
lower due to decreasing air pollution trends. When estimating health impacts, we are most confident in
results estimated using projected air quality concentrations that closely align with those observed in the
epidemiological study from which the risk estimate was obtained (i.e., we are less confident applying
risk estimates to pollutant concentrations that do not match the original cohort due to changes in air
pollutant concentrations over time). To address the potential mismatch between projected air quality

77 As estimates of infant mortality incidence are relatively small, we do not perform quantitative uncertainty
analyses for that health endpoint.

97


-------
levels and those in the epidemiologic study, we include air quality information from the original
epidemiologic studies where feasible.

Additional information was requested from mortality study authors regarding the ambient
concentrations used to estimate exposure of the original cohort.78 Study authors provided cohort
specific PM2.5 concentration data at varying levels of detail. PM2.5 concentrations for the two long-term
exposure epidemiologic cohort studies examining mortality, ACS CSP-II and Medicare are presented in
Figure 13 (Di et al., 2017b, Turner et al., 2016). We also included the distribution of PM2.5 concentrations
from a recent analysis of the CanCHEC cohort in order to compare to some of the lowest reported
concentrations in North America (Crouse et al., 2015). Points reflect cohort specific PM2.5 concentration
data, with connecting lines estimating missing data.

PM Concentration (|jg/mA3)

Figure 13. Cumulative Percentile of PM2.5 Cohort Exposure from the ACS CSP-II, Medicare, and CanCHEC
Cohorts

As air pollution concentrations continue to decline an increasing fraction of the population will be
exposed to PM2.5 concentrations at the lower end of the air quality distribution experienced by the study
cohort. The distribution of PM2.5 concentrations for each of three large, long-term exposure cohorts are

78 For morbidity studies, author-reported air quality information such as the average or mean, standard deviation,
and maximum and minimum concentrations were collected and is available in the corresponding Study
Information Table.

98


-------
provided in Table 33. For comparison, the lowest reported PM2.5 concentrations from previous studies
(Krewski et al., 2009, Lepeule et al., 2012) in which risk estimates were used to estimate all-cause
mortality attributed to long-term PM2.5 exposure were 5.8 and 8.0 ng/m3, whereas the recent studies
identified as best characterizing long-term PM2.5 exposure and the risk of all-cause mortality include
PM2.5 concentrations below 3 ng/m3 (Di et al., 2017b, Turner et al., 2016).

Table 33. Low Concentration PM2.5 Exposures from the ACS CSP-II, Medicare, and CanCHEC Cohorts

Cohort

Percentile of Cohort Exposure (ng/m3)

0.0%

1.0%

1.5%

2.0%

2.5%

3.0%

3.5%

4.0%

4.5%

5.0%

10.0%

ACS CSP-II

2.8

5.8



6.3

6.5

6.6



7.0



7.4



Medicare

0.0

5.5

5.9

6.1

6.3

6.5

6.7

6.8

6.9

7.1

7.9

CanCHEC

0.0

3.2



3.5

3.6









4.0

4.7

We note that PM2.5 concentrations reported in cohort studies are not equivalent to NAAQS design
values (DVs). Information relating PM2.5 concentrations from cohort studies discussed within this section
to PM2.5 DVs can be found in section 3.2 of the 2020 PM PA (U.S. EPA, 2020c).

6.1.2.2 Estimating and Assigning Exposures in Epidemiology Studies

New developments in exposure assessment, including hybrid spatiotemporal models that incorporate
satellite observations of AOD, land use variables, surface monitoring data from monitors, and chemical
transport models, have led to improvements in the spatial resolution and extent of pollutant
concentration surfaces. After reviewing the current state of exposure science, the 2019 PM ISA stated
that "a number of studies demonstrate that the positive associations observed between long-term PM2.5
exposure and mortality are robust to different methods of assigning exposure" and the 2020 03 ISA
articulated that "hybrid methods have produced lower error predictions of ozone concentration
compared with spatiotemporal models using land use and other geospatial data alone but may be
subject to overfitting given the many different sources of data incorporated into the hybrid framework."

Although these advancements may reduce bias and uncertainty in risk estimates, the accuracy of hybrid
exposure estimates can be difficult to confirm in areas lacking monitors. On the other hand, studies
using monitor data as the only exposure information have increasing exposure uncertainty the farther
people live from the monitor site.

Di et al., 2017a provided PM2.5-attributable mortality risk estimates based on either a hybrid exposure
estimation approach combining photochemical air quality modeling with ground-level monitoring data
or only on monitoring data. Comparing these two estimates aids in understanding how sensitive long-
term, all-cause estimates of PM2.5-attributable mortality are to exposure estimation method. A
comparison of the risk estimates using either the hybrid or monitor-based exposure estimates is
available in Table 34. The italicized risk estimate was identified for use in the main PM2.5 benefits
assessment.

99


-------
Table 34. Di et al., 2017a PIVh.s-Attributable Mortality Risk Estimates per 10 ng/m3 from Different
Exposure Estimation Techniques

Exposure Technique

Risk Estimate

Hybrid Exposure Estimate

1.073 (1.071,1.075)

Monitor-Based Exposure Estimate

1.061 (1.059, 1.063)

Turner et al., 2016 and Pope et al., 2015 analyzed the same ACS CSP-II population over the same time
period but used different hybrid exposure estimation techniques. Turner et al., 2016 used the
hierarchical Bayesian space-time model (HBM) approach, which combines ambient measurement data
with gridded estimates from the CMAQ photochemical model. Pope et al., 2015 used a land use
regression model with Bayesian Maximum Entropy kriging of residuals (LURBME). Sensitivity of the risk
estimate to the exposure estimation technique is available in Table 35, including the estimate identified
for the main benefits assessment in italics.

Table 35. PM2.5-Attributable ACS CSP-II Mortality Risk Estimates per 10 ng/m3 from Different Exposure
Estimation Techniques

Exposure Technique

Risk Estimate

HBM

1.06 (1.04-1.08)

LURBME

1.07 (1.06-1.09)

6.1.2.3 Confounding by O3

When considering the relationship between pollutant exposure and health effects, it can be informative
to consider whether risk estimates are subjected to confounding when including other pollutants in
copollutant models, especially when health impacts of more than two highly correlated pollutants are
being estimated concurrently.79 Regarding long-term exposures, the 2019 PM ISA concluded that
"positive associations observed between long-term PM2.5 exposure and total mortality remain relatively
unchanged after adjustment for 03, N02, and PM10-2.5"

Both Turner et al., 2016 and Di et al., 2017a provided single-pollutant and two-pollutant (including 03 as
a copollutant) PM2.5-attributable mortality risk estimates. Although the 2019 PM ISA found that, in
general, PM2.5 risk estimates were relatively unchanged to the inclusion of 03 in copollutant models, a
comparison of risk estimates that either do or do not include 03 as a copollutant is included to clarify
this potential sensitivity with respect to all-cause PM2.5-attributable mortality. Differences in the
magnitude of risk estimates including or excluding 03 as a copollutant are provided in Table 36. Italicized
risk estimates were identified for use in the main benefits assessment.

Table 36. Single- and Two-Pollutant (Including 03 as a Copollutant) PM2.5-Attributable Mortality Risk
Estimates per 10 ng/m3



Turner et al., 2016

Di et al., 2017a

Two-Pollutant

1.06 (1.04-1.08)

1.073 (1.071, 1.075)

Single-Pollutant

1.06 (1.04-1.08)

1.084 (1.081, 1.086)

79 Modeling more than two correlated pollutants can be problematic due to collinearity issues.

100


-------
6.1.2.4 Statistical Technique

Di et al., 2017a provided mortality risk estimates using two different statistical methods to adjust for
covariates, potentially providing insight into model uncertainties associated with statistical regression
techniques. A comparison of the risk estimates using either the generalized estimating equation (GEE)
approach, which the authors identified as the main analysis, or the mixed-effects model (COXME) can be
found in Table 37.

Table 37. Di et al., 2017a PIVh.s-Attributable Mortality Risk Estimates per 10 ng/m3 from Different
Statistical Techniques

Statistical Technique

Risk Estimate

GEE

1.073 (1.071, 1.075)

COXME

1.081 (1.078, 1.083)

6.1.3 Asthma Onset in Children

For a number of health endpoints we identified plausible alternative risk estimates to characterize the
sensitivity of the main risk estimate to alternative assumptions and/or input parameters. Below we
detail: 1) the endpoints for which we considered alternative risk estimates; and 2) the studies from
which we drew the alternative risk estimates. This type of sensitivity assessment is also performed for
other PM2.5 and 03 health endpoints in sections 6.1.4, 6.1.5, and 6.2.4.

The study identified as best characterizing risk for this health endpoint took place in Canada (Tetreault
et al., 2016). Even though comparatively Tetreault et al., 2016 was preferred in all identification criteria
to other available studies (e.g., study size, exposure estimation technique, study period, etc.) other than
location, we thought it useful to include the available U.S.-based risk estimates as uncertainty analyses.
An overall comparison of the main risk estimate and 95% confidence interval from Tetreault et al., 2016
and the alternative risk estimates and confidence intervals from McConnell et al., 2010 and Nishimura et
al., 2013 can be found in Table 38. Details about the two studies providing alternate risk estimates is
below.

101


-------
Table 38. Potential Sensitivity of Estimated Instances of Asthma Onset

Potential Source of
Uncertainty

Potential Insights Gained from Quantitative Uncertainty Analyzes2

Application of Risk Estimates
to Other Locations and
Populations

Tetreault et al., 2016 included only Canadians whereas Nishimura et
al., 2013 included five U.S. urban areas and McConnell et al., 2010 was
restricted to southern CA

Study Size

Tetreault et al., 2016 included the largest study size, approximately
twenty-five times the size of either Nishimura et al., 2013 or
McConnell et al., 2010

Study Period

Tetreault et al., 2016 evaluated the most recent health study period
(1996-2011) compared to 2002-2006 for McConnell et al., 2010 and
1986-2005 for Nishimura et al., 2013

Exposure Estimate

Tetreault et al., 2016 used hybrid exposure estimates whereas sever
states, whereas Nishimura et al., 2013 and McConnell et al., 2010 used
monitor-based estimates

Statistical Technique

Tetreault et al., 2016 and McConnell et al., 2010 use time-varying and
multilevel Cox proportional hazard models, respectively, whereas
Nishimura et al., 2013 uses logistical regression models

Two of the five ISA-identified studies of asthma onset took place in the U.S. (McConnell et al., 2010,
Nishimura et al., 2013). McConnell et al., 2010 examined the association between long-term traffic-
related air pollution (PM2.5, PM10, 03, and N02) exposure and incident asthma in children. The authors
collected data for three years from a cohort of 2,497 kindergarten and first-grade children aged 4-9 who
entered the Southern California Children's Health Study without asthma or wheeze. McConnell et al.,
2010 defined new-onset asthma as physician-diagnosed asthma reported by parents on a yearly
questionnaire. While the primary focus of the study was traffic-related air pollution from local vehicle
emissions, the authors also utilized ambient air pollution exposure data from central site monitors in
each of the 13 communities in the Southern California Children's Health Study. The authors used a
multilevel Cox proportional hazards model to estimate the association between ambient air pollution
exposure and new-onset asthma, controlling for race/ethnicity, secondhand smoke exposure, and pets
in the home. The identified hazard ratio of 1.66 (95% CI: 0.91-3.05) for a 17.4 ng/m3 (range of exposure
in the 13 communities) increase in annual average PM2.5 exposure came from a single pollutant model.

The other study, Nishimura et al., 2013, investigated the relationship between long-term early-life
pollution exposure (PM2.5, PM10, 03, N02, and S02) and asthma onset in 3,343 Latino and African
American children in five urban areas (Chicago, IL; Bronx, NY; Houston, TX; San Francisco, CA; Puerto
Rico). The authors obtained data from the Genes-environments and Admixture in Latino Americans
(GALA II) Study and the Study of African Americans, Asthma, Genes and Environments (SAGE II). GALA II
and SAGE II are case-control studies that enrolled children with and without asthma. The studies defined
case subjects as children with physician-diagnosed asthma plus two or more symptoms of coughing,
wheezing, or shortness of breath in the two years before study enrollment while control subjects were
children with no reported history of asthma, lung disease, or chronic illness, and no reported symptoms
of coughing, wheezing, or shortness of breath in the two years before study enrollment. The authors
estimated annual average pollution exposures during the first year of life as well as the first three years
of life from self-reported residential histories by calculating inverse distance-squared weighted averages

102


-------
from the four closest U.S. EPA Air Quality System monitoring stations within 50 km. The authors first
used regional- and study-specific logistic regression models to estimate the association between asthma
diagnosis and pollution exposure, controlling for demographics and socioeconomic status and
subsequently combined the regression coefficients into a multi-region estimate using a random-effects
meta-analysis. The identified odds ratio of 1.03 (95% CI: 0.90-1.18) for a 1 ng/m3 increase in average
annual PM2.5 levels at the residential address during the first year of life came from a single pollutant
model. Beta effect coefficients from the main (italicized) and sensitivity analyses are available in Table
39.

Table 39. Beta Coefficients and Standard Errors (SE) from Studies of Examining Long-term PM2.5
Exposure and New Onset Asthma in Children

Study

Age Range

Beta Coefficient (SE)

Tetreault et al., 2016

0-17

0.044 (0.0009)

McConnell et al., 2010

4-17

0.029 (0.017)

Nishimura et al., 2013

7-21

0.030 (0.069)

6,1.4 Cardiovascular Hospital Admissions

Bell et al., 2015 was identified as best characterizing risk across the U.S. for benefits assessment
purposes as it included the largest study size, most recent time period, and a nationally representative
geographic area. However, it was restricted to ages >64 and based exposure estimates solely on
monitoring data. There was also another large study of PM2.5-attributable cardiovascular hospital
admission impacts that included all ages and incorporated hybrid exposure estimation techniques
(Talbott et al., 2014). Differences in the age ranges and ICD-9 codes prevented pooling of the two
estimates but comparing the two estimates could provide insights into uncertainties associated with
epidemiologic estimates of this health endpoint (Table 40). Therefore, we include a risk estimate of
cardiovascular hospital admission impacts from long-term PM2.5 exposure from Talbott et al., 2014 as a
sensitivity analysis of this health endpoint (Table 41). Please note that Talbott et al., 2014 provides
individual risk estimates for each state, which will be pooled into a single estimate to compare with Bell
et al., 2015.

Talbott et al., 2014 assessed daily PM2.5 concentrations and hospitalizations for cardiovascular disease in
Florida, Massachusetts, New Hampshire, New Jersey, New Mexico, New York, and Washington from
2001 to 2008. The authors gathered hospital discharge data from each state's respective data stewards.
Talbott et al., 2014 conducted a time-stratified case crossover study using hospitalization data for all
cardiovascular diseases (ICD-9 390-459) and for several specific cardiovascular diseases within the ICD-9
390-459 range. Authors used a downscaling Bayesian space-time modeling approach to combine air
monitoring data and air gridded numerical outputs from CMAQto predict daily PM2.5 concentrations.
The authors gathered meteorological data from the Centers for Disease Control (CDC) Wonder North
America Land Data Assimilation System Daily Air Temperatures and Heat Index. Risk estimates were
presented for a 10 ng/m3 increase in PM2.5 for each state and season across three single-day lags (0, 1,
and 2) and a three-day lag average (0-2) by diagnosis. All-year risk estimates were identified over
season-specific estimates and estimates of multiday average lag period were identified over single-day

103


-------
lag estimates.80 The seven state-specific risk estimates identified as sensitivity analyses were pooled
using the random or fixed effects algorithm in BenMAP-CE.81 The seven risk estimates reflect a mix of
positive and negative values. State-specific risk estimates identified from Talbott et al., 2014 come from
a two-pollutant multivariable model including 03 of ICD-9 codes 390-459: 1.005 (95% CI: 0.998-1.012)
for Massachusetts; 1.011 (95% CI: 1.007-1.016) for New Jersey; 1.011 (95% CI: 0.973-1.050) for New
Mexico; and 1.011 (95% CI: 1.008-1.014) for New York. Each odds ratio is for a 10 ng/m3 increase in the
averaged daily mean PM2.5 concentration 0-, 1-, and 2-day lags (Talbott et al., 2014, Table 3). Beta effect
coefficients from the main (italicized) and sensitivity analyses are available in Table 41.

Table 40. Potential Sensitivity of Estimated Cardiovascular Hospital Admissions

Potential Source of Uncertainty

Potential Insights Gained from Quantitative Uncertainty
Analyzes

Application of Risk Estimates to
Other Locations and Populations

Talbott et al., 2014 included all ages whereas Bell et al., 2015
was restricted to ages >64

Confounding by Individual Risk
Factors (Location)

Talbott et al., 2014 was restricted to seven states, Bell et al.,
2015 included all states

Confounding by Other Pollutants

Talbott et al., 2014 included the copollutant 03

Exposure Estimate

Talbott et al., 2014 used hybrid exposure estimates whereas
sever states, Bell et al., 2015 used monitor-based estimates

Table 41. PM2.5-Attributable Cardiovascular Hospital Admissions Beta Estimates

Study

Beta Coefficient (SE)

Bell et al.,
2015

0.00065 (0.00009)

Talbott et al.,
2014, MA

MA: 0.00050 (0.00035), NJ: 0.00109 (0.0002), NM: 0.00109 (0.0019), NY: 0.00109
(0.00015), FL: -0.00040 (0.0003), NH: -0.00121 (0.0012), WA: -0.00090 (0.0005)

6.1.5 Respiratory Hospital Admissions

Similar to cardiovascular hospital admissions, there was an estimate of PM2.5-attributable respiratory
hospital admissions that included all ages and utilized a hybrid exposure estimation approach, but was
geographically limited, in this case to a single state. However, we thought it useful to include this
estimate as a sensitivity analysis due to the contrasts between it and the italicized main benefits
estimates (Table 42 and Table 43). As compared to PM2.5-attributable mortality and cardiovascular
hospital admission impact estimates, there may be greater uncertainty associated with estimates of
PM2.5-attributable respiratory hospital admissions (Table 43).

80	Lag period preference identification criteria is more fully described in 2019 PM ISA Appendix Table A-l.

81	Random or fixed effects pooling is a method to combine two or more distributions into a single new distribution,
allowing for the possibilities that either 1) a single true underlying relationship exists between the component
distributions, and that differences among estimated parameters are the result of sampling error, or 2) the
estimated parameter from different studies may in fact be estimates of different parameters, rather than just
different estimates of a single underlying parameter, and weights for the pooling are generated via inverse
variance weighting, thus giving more weight to the studies that exhibit lower variance and less weight to the input
distributions with higher variance.

104


-------
Jones et al., 2015 encompassed all ages, races, and ethnicities with a case-crossover analysis in New
York state, using 24-hour average PM2.5 concentrations from CMAQand meteorological data from the
National Climactic Data Center. The authors assessed hospital discharge data from the New York State
Department of Health State Planning and Research Cooperative System through principle diagnosis
categorized by ICD-9 code (chronic bronchitis (ICD-9 491), emphysema (ICD-9 492), asthma (ICD-9 493),
and chronic airway obstruction (ICD-9 496)). Authors used a single pollutant conditional logistic
regression model to analyze the respiratory hospital admission and PM2.5 chemical constituent data over
time and by season. The authors calculated hazard ratios using SAS (version 9.2) with 95% confidence
intervals from the regression models. The estimate best characterizing U.S. risk comes from the 4-day
lag all-year PM2.5 estimate in Figure 2a: 1.006 (1.004-1.009). Please note, this risk estimate was derived
from the figure, as the exact numbers were not provided in the paper and the authors did not respond
to our request.

Table 42. Potential Respiratory Hospital Admission Sensitivity Insights

Potential Source of Uncertainty

Potential Insights Gained from Quantitative Uncertainty
Analyzes2

Application of Risk Estimates to Other
Locations and Populations

Jones et al., 2015 included all ages whereas Bell et al., 2015
was restricted to ages >64

Confounding by Individual Risk Factors
(Location)

Jones et al., 2015 was restricted to a single state, Bell et al.,
2015 included all states

Exposure Estimate

Jones et al., 2015 used hybrid exposure estimates, whereas
Bell et al., 2015 used monitor-based estimates

Table 43. PM2.5-Attributable Respiratory Hospital Admissions Beta Risk Estimates

Study

Age Range

Beta Coefficient (SE)

Bell et al., 2015

65-99

0.00025 (0.0001)

Ostro et al., 2016

0-18

0.00275 (0.0008)

Jones et al., 2015

0-99

0.00080 (0.0002)

6.1.5.1 Emergency Hospital Admissions (EHAs)

Interestingly, a substantial subset of the ISA-identified recent epidemiologic literature evaluating
respiratory hospitalizations restricted analyses to emergency hospital admissions (EHAs), defined as
hospitalizations admitted from the emergency department (section 3.2). Due to time and resource
requirements, we were unable to develop county-level baseline incidence data for EHAs, in addition to
total hospital admissions. However, as we were interested in how estimates of EHAs compared to total,
we include a risk estimate of respiratory EHAs from Zanobetti et al., 2009 using national baseline
incidence data. Though the EHA estimate came from a smaller and older study then the main analysis
respiratory hospital admission study, the EHA estimate is nearly an order of magnitude larger than the
risk estimate included in the main estimate (italicized).

105


-------
Table 44. Comparison of the PM2.5-Attributable Respiratory Hospital Admissions Beta Risk Estimate to
the EHA Respiratory Estimate

Study

Beta Coefficient (SE)

Bell et ai, 2015

0.00025 (0.0001)

Zanobetti et al., 2009

0.00204 (0.0004)

6.1.6 Effect Modification of Health Impacts in At-Risk Populations82
ISAs typically include an assessment of the weight of evidence demonstrating that certain
subpopulations experience increased mortality or morbidity risks from air pollutant exposure compared
to other groups. This is also known as effect modification and occurs when the measure of an effect
changes across levels of a variable other than PM2.5 exposure. The 2019 PM ISA examined toxicological,
controlled human exposures, and epidemiologic literature considering whether certain populations and
lifestages might be at increased risk of air pollutant-related health effects (U.S. EPA, 2019c).

The ISAs categorize relationships between exposure and effect modification for various population and
lifestages into the following four groups:

•	Adequate evidence: There is substantial, consistent evidence within a discipline to conclude that
a factor results in a population or lifestage being at increased risk of air pollutant-related health
effect(s) relative to some reference population or lifestage. Where applicable, this evidence
includes coherence across disciplines. Evidence includes multiple high-quality studies.

•	Suggestive evidence: The collective evidence suggests that a factor results in a population or
lifestage being at increased risk of air pollutant-related health effect(s) relative to some
reference population or lifestage, but the evidence is limited due to some inconsistency within a
discipline or, where applicable, a lack of coherence across disciplines.

•	Inadequate evidence: The collective evidence is inadequate to conclude whether a factor results
in a population or lifestage being at increased risk of air pollutant-related health effect(s)
relative to some reference population or lifestage. The available studies are of insufficient
quantity, quality, consistency, and/or statistical power to permit a conclusion to be drawn.

•	Evidence of no effect: There is substantial, consistent evidence within a discipline to conclude
that a factor does not result in a population or lifestage being at increased risk of air pollutant-
related health effect(s) relative to some reference population or lifestage. Where applicable, the
evidence includes coherence across disciplines. Evidence includes multiple high-quality studies.

Presenting PIVh.s-attributable benefit estimates striated by the value of another covariate can provide
insight into risk within various population subgroups. To accomplish this, we reviewed relevant chapters
from the 2019 PM ISA in order to compile and assess studies cited in support of the Agency's
determinations, focusing on studies referenced in Table 12-3 for population characteristics with either
"adequate evidence" (i.e., substantial, consistent evidence) or "suggestive evidence" (i.e., limited

82 Analyses of effect modification will not be included in the main analyses, so as to avoid the possibility of double-
counting impacts. This potential uncertainty could also be described as the effect modification of individual risk
factors.

106


-------
evidence due to inconsistency or a lack of coherence in research) of increased risk (sections 6.1.6.1 and
6.1.6.2) (U.S. EPA, 2019c). The factors with "adequate evidence" for PM2.5 are lifestage (children) and
race (nonwhite populations), while the factors with "suggestive evidence" are pre-existing disease
(cardiovascular disease, respiratory disease, and obesity), genetic factors (variant genotypes), low
socioeconomic status, and smoking.

6.1.6.1 Study and Risk Estimate Identification Criteria for Populations At-Riskfor PM2.5 Exposures
We identified all studies in the related section of ISA Chapter 12 for each at-risk factor listed above,
resulting in a set of 123 studies for at-risk populations. This collection includes the following number of
studies, with some studies duplicated for multiple endpoints.

•	8 studies for lifestage (children),

•	25 studies for race (nonwhite populations),

•	67 studies for pre-existing disease across disease types,

•	18 studies for genetic factors, 25 studies for socioeconomic status, and

•	14 studies for smoking.

We then focused our review on the risk factors with "adequate evidence", due to stronger supporting
evidence as well as because they could be evaluated using currently available data. We extracted study
information from all studies with "adequate evidence" and applied initial screening criteria to identify
peer-reviewed, epidemiological studies focused on PM2.5 conducted in the US or Canada. We also
documented the mortality and/or morbidity health endpoints included in each study, focusing on all-
cause mortality and respiratory morbidity endpoints. We then evaluated the group of remaining studies
based on additional identification criteria built off the criteria for the primary analysis, described in
Table 45.

Table 45. PM2.5 At-Risk Study Identification Criteria

Criterion

Notes

Peer-Review

Peer-reviewed research exclusively

Study design

Epidemiological study

PM2.5 Study

PM2.5 exposure rather than other PM10 or other sizes

Study Location

US or Canada

Study Duration

Long-term studies preferred

Population Attributes

Presents risk estimates for clearly defined at-risk groups for which data
currently exist in BenMAP-CE

Causal or Likely Causal
Health Effects

Adequate evidence for at-risk groups in ISA

Economically Valuable
Health Effects

Health endpoints for which economic values have been or could
reasonably be developed

Baseline Incidence Data

Must be able to identify baseline incidence data for subpopulations

6.1.6.2 All-Cause Mortality

For the mortality endpoint, seven all-cause mortality studies for the nonwhite population met our
criteria. No mortality studies for the child at-risk group met the initial screening criteria. Of the seven
studies of nonwhite populations, three were short-term exposure studies relating daily PM2.5 exposure

107


-------
and daily deaths and four were long-term exposure studies relating annual PM2.5 exposure and annual
mortality. Consistent with the main benefits assessment, we focused on the following four long-term
studies as long-term exposure studies may include some effects of short-term exposures (section
2.2.5.1): Di et al., 2017b, Kioumourtzoglou et al., 2016, Parker et al., 2018, and Wang et al., 2017.

We evaluated specific details of risk estimates provided by each study to determine if sufficient
information exists for use in a quantitative sensitivity analysis. Of the studies, only Di et al., 2017b
provided sufficient information to apply risk models quantifying increased risks to nonwhite groups,
including non-Hispanic white, Black, Asian, Native American, and Hispanic-white populations. Additional
detail on the study can be found in section 2.2.5.1.3.1.2 or in the associated Study Information Table.

We applied similar criteria to morbidity endpoints for the child and nonwhite at-risk groups. No studies
cited for the child subgroup met our criteria for inclusion, and one endpoint, emergency room visits for
asthma, was chosen for quantification for the nonwhite populations at-risk factor group. The study we
chose to evaluate was Alhanti et al., 2016, which presents risk information for white and pooled
nonwhite populations disaggregated into five age groups.

We developed BenMAP-ready Health Impact Functions for each at-risk group described by Di et al.,
2017b and Alhanti et al., 2016, summarized in Table 46.

Table 46. Identified PM2.5 At-Risk Beta Coefficients and Standard Errors

At-Risk Factor

Endpoint

Study

Subgroup

Beta Coefficient
(SE)

Race, nonwhite
populations

Morality, All Cause

Di et al.,
2017b

Non-Hispanic
White

0.0061 (0.0001)

Hispanic White

0.0110 (0.0008)

Black

0.0189 (0.0004)

Asian

0.0092 (0.0010)

Native
American

0.0095 (0.0019)

Race, nonwhite
populations

Emergency Room
Visits, Asthma

Alhanti et al.,
2016

White, age 0-4

0.0025 (0.0019)

Nonwhite, age
0-4

0.0037 (0.0012)

White, age 5-18

0.0025 (0.0016)

Nonwhite, age
5-18

0.0049 (0.0012)

108


-------
6.2 Quantitative Characterization of O3 Uncertainties and Evaluating Sensitivity to
Alternate O3 Assumptions

6.2.1	Statistical Uncertainty Around the Risk Estimate (Monte-Carlo Assessment)

For all endpoints analyzed, we use a Monte Carlo simulation in which we sample from the standard
error associated with each risk estimate and present the resulting 2.5th and 97.5th percentile values from
this distribution as a 95th percentile confidence interval around the estimated health impact and
monetized health benefits. Monte Carlo methods are a well-established means of characterizing random
sampling error associated with the risk estimates from epidemiologic studies. This approach randomly
samples from a distribution of incidence and valuation estimates to characterize the effects of
uncertainty on output variables. The reported standard errors in the epidemiologic studies determined
the distributions for individual effect estimates for endpoints estimated using a single study. For
endpoints estimated using a pooled estimate of multiple studies, the confidence intervals reflect both
the standard errors and the variance across studies. The confidence intervals around the monetized
benefits incorporate the standard errors from the epidemiologic risk estimate as well as the distribution
of the valuation function. These confidence intervals do not reflect other sources of uncertainty
inherent within the estimates, such as baseline incidence rates, populations exposed and transferability
of the effect estimate to diverse locations. As a result, the reported confidence intervals and range of
estimates give an incomplete picture about the overall uncertainty in the benefits estimates.

6.2.2	Respiratory Mortality

6.2.2.1.1 Confounding by PM2.5

When considering the relationship between pollutant exposure and health impacts, it can be
informative to consider whether risk estimates are changed when other pollutants are included in
copollutant models, especially when health impacts of multiple pollutants are being estimated
concurrently. While no conclusions were formed regarding the impact of copollutant confounding on
long-term exposure-related respiratory mortality, the 2020 03 ISA found that "positive associations
observed between long term 03 exposure and total mortality remain relatively unchanged after
adjustment for PM2.5 and N02."

Turner et al., 2016 provided single- and multipollutant (including PM2.5 as a copollutant) 03-attributable
respiratory mortality risk estimates. A comparison of risk estimates that either do or do not include
PM2.5 as a copollutant is included to clarify this potential sensitivity with respect to 03-attributable
respiratory mortality. Differences in the magnitude of risk estimates including or excluding PM2.5 as a
copollutant are provided in Table 47.

Table 47. Single- and Two-Pollutant (Including PM2.5 as a Copollutant) Long-Term 03-Attributable
Respiratory Mortality Risk Estimates per 10 ppb

Study

Single-Pollutant

Multipollutant

Turner et al., 2016

1.14(1.10-1.18)

1.12 (1.08-1.16)

109


-------
6.2.2.2 Short-Term O3 Exposures

6.2.2.2.1	Potential Threshold Analysis

The 2020 final 03 ISA evaluated a number of studies examining the shape of the concentration-response
relationship for short-term 03 exposure and total/nonaccidental mortality, which we use to inform the
long-term 03-attributable respiratory mortality relationship. The ozone ISA found that "studies that used
different statistical approaches and ozone averaging times (i.e., 24 hour avg and 8 hour max) provide
evidence of a linear concentration-response relationship, with less certainty in the shape of the curve at
lower concentrations [i.e., ...30 ppb for 8 hour max], [although] an examination of whether a threshold
exists in the ozone mortality concentration-response relationship provided no evidence of a
concentration below which mortality effects do not occur when examining 5 ng/m3 (~2.55 ppb)
increments across the range of 1 hour max concentrations reported in the U.S. and Canadian cities
included in [a large cohort]." As the Zanobetti and Schwartz, 2008 risk estimate uses the MDA8 metric, it
can be used to quantitatively assess the effect of an 03 threshold at 30 ppb would have on benefits
estimates. For context, approximately 3.7% of the contiguous U.S. population is projected to reside in
areas where MDA8 03 concentrations are annually below 30 ppb in 2024 (U.S. EPA, 2020b). Clinical
evidence provides little indication that adverse effects occur at extremely low levels in most individuals.
Epidemiologic evidence is qualitatively discussed further in section 6.5.14.2.

6.2.2.2.2	Confounding by PM

Regarding short-term exposures, the 2020 03 ISA found that "the few recent multicity studies that
examined potential copollutant confounding provide evidence supporting that 03 mortality risk
estimates are relatively unchanged or slightly attenuated, but remain positive, in copollutant models
with PM2.5, PM10, and N02."

Katsouyanni et al., 2009 provided single- and two-pollutant (including PM10 as a copollutant) short-term
03-attributable respiratory mortality risk estimates for a subset of 15 of the 86 cities analyzed. A
comparison of risk estimates that either do or do not include PM10 as a copollutant is included to clarify
this potential sensitivity with respect to 03-attributable respiratory mortality. Differences in the
magnitude of risk estimates including or excluding PM2.5 as a copollutant are provided in Table 48.

Please note, as distributed lag risk estimates were not provided for the two-pollutant analyses, in
additional to the inclusion of PM2.5 as a copollutant and the number of cities analyzed, there is a
difference in the lag duration between the estimates in Table 48.

Table 48. Single- and Two-Pollutant (Including PMi0 as a Copollutant) Short-Term 03 Exposure 03-
Attributable Excess Premature Respiratory Mortality Risk Estimates per 10 ppb

Study

Single-Pollutant

Two-Pollutant

Katsouyanni et al., 2009

0.77% (0.17%, 1.37%)

0.99% (-0.33%, 2.31%)

6.2.3 All-Cause Mortality

When estimating air pollutant-attributable health impacts, EPA focuses on endpoints for which the
underlying scientific evidence is strongest. That is, when evaluating evidence across scientific disciplines
(i.e., clinical, animal toxicological, and epidemiologic) there is often consistency of effects within a
discipline, coherence of effects across disciplines, and evidence of biological plausibility. Such an
approach gives us greater confidence in the relationship between exposure and health outcome. For

110


-------
criteria pollutants, EPA typically relies upon the causality determinations in the latest ISA or equivalent,
which are made using a weight-of evidence approach. Generally, to estimate the pollutant-attributable
human health benefits in which we are most confident, we at least assess health effects identified as
having a 'causal' or 'likely to be causal' relationship with the pollutant of interest in the most recently
published ISA. This is not to imply there may not be benefits associated with endpoints having a
"suggestive of, but not sufficient to infer, a causal relationship," but rather that there is greater
uncertainty in these potential benefits.

Because of the significance of the endpoint, we include a limited quantitative sensitivity analysis of total
mortality associated with long-term 03 exposure. While the 2020 03 ISA concluded that evidence was
"suggestive of, but not sufficient to infer," a causal relationship between long-term exposures and total
mortality, the reduction of this risk is likely still valuable to society. As such, for this sensitivity
discussion, we include risk estimates of long-term, all-cause 03-attributable total mortality from the two
studies used to estimate PM2.5-attributable mortality risk (Table 49). Please note these long-term, all-
cause risk estimates include respiratory mortality estimates and should not be added to the respiratory
mortality estimates.

Table 49. Long-Term 03-Attributable Total Mortality Risk Estimates per 10 ppb

Study

Risk Estimate (per 10
ppb increase in 03)

Risk Estimate Details

Turner et al., 2016

1.02 (1.01-1.03)

Fully adjusted HBM multipollutant estimate from Table
E9, ages 35-99

Di et al., 2017b

1.011 (1.010, 1.012)

GEE two-pollutant main analysis estimate from Table 2,
ages 65-99

6,2.4 Asthma Onset in Children

The study identified as best characterizing risk for asthma onset in children was conducted in Canada
(Tetreault et al., 2016). Even though comparatively Tetreault et al., 2016 was preferred in identification
criteria to other available studies (e.g., study size, exposure estimation technique, etc.) other than
location and study period, we thought it useful to include the largest and most recent U.S.-based risk
estimates as a sensitivity analysis. An overall comparison of the main risk estimate from Tetreault et al.,
2016 and the alternative risk estimates from Garcia et al., 2019 can be found in Table 50. Details about
the study providing an alternate risk estimate is below.

Ill


-------
Table 50. Potential Sensitivity of Estimated Instances of Asthma Onset

Potential Source of Uncertainty

Potential Insights Gained from Quantitative Uncertainty
Analyses2

Application of Risk Estimates to
Other Locations and Populations

Tetreault et al., 2016 included only Canadians whereas Garcia et
al., 2019 was restricted to southern CA

Study Size

Tetreault et al., 2016 included the largest study size,
approximately twenty-five times the size of Garcia et al., 2019

Study Period

Garcia et al., 2019 evaluated a more recent and longer health
study period (1993-2014) compared to 2002-2011 for Tetreault et
al., 2016

Exposure Estimate

Tetreault et al., 2016 used hybrid exposure estimates whereas
sever states, whereas Garcia et al., 2019 used monitor-based
estimates

Statistical Technique

Tetreault et al., 2016 used time-varying Cox proportional hazard
models, whereas Garcia et al., 2019 uses Poisson log-linear
regression models

Three of the four ISA-identified studies of long-term 03-attributable asthma onset took place in the U.S.,
although only one included a study period more recent than 2005 (Garcia et al., 2019). Garcia et al.,
2019 examined the associations between long-term ozone exposure and asthma onset in children (aged
nine-18 years) with no history of asthma in Southern California. The authors followed three waves of
participants from the Children's Health Study for eight years between 1993 and 2014. Garcia et al., 2019
obtained health and demographic data from parents, guardians, or participants, who completed
questionnaires annually. In order to calculate annual mean, community-level ozone exposure, the
authors acquired daily eight-hour mean 03 concentrations through ambient air pollution monitors.
Multi-level Poisson regression models with one-year lag showed no statistically significant associations
between long-term 03 exposure and asthma onset in children. Models adjusted for demographic
variables as well as factors pertaining to family medical history, environmental factors, and near-
roadway pollution.

The magnitudes of main and alternate risk estimates of long-term 03 exposure and asthma onset in
children provided in Table 51.

Table 51. Long-Term 03-Attributable Asthma Beta Coefficients

Study

Beta Coefficient

Age Range

Tetreault et al., 2016

0.020754

0-17

Garcia et al., 2019

0.01695

9-18

112


-------
6.2,5 Understanding the Effect Modification of Health Impacts in At-Risk Populations83
Effect modification was also investigated with regard to 03 exposures. We reviewed relevant chapters
from 2020 03 ISA and used a similar methods to that described for PM2.5 to compile and assess studies
cited in support of the Agency's determinations (section 6.1.6). As the 2020 03 ISA only presents an
evaluation of at-risk groups in summary form and extensively references the findings from the 2013 03
ISA, we focused on the detailed chapter from that previous document in identifying the at-risk factors
and studies to review (U.S. EPA, 2013).Factors with "adequate evidence" are genetic factors, asthma,
children, older adults, diet, and outdoor workers. Factors with "suggestive evidence" are sex, SES, and
obesity. Considering feasibility and our review criteria, we focused on studies addressing increased risks
based on age in the adequate evidence group and note that some health functions already applied in
the primary analysis focus on asthmatic subpopulations. We also elected to include illustrative
calculations for some risk factors with "suggestive evidence", specifically those for sex.

6.2.5.1 Study and Risk Estimate Identification Criteria for Populations At-Risk for O3 Exposures
We compiled epidemiologic studies from the related section of Chapter 8 of the 2013 03 ISA for the
following at-risk factors, excluding all other study types (e.g. toxicological studies), for a total of 28
studies. This collection includes the following number of studies, with some studies duplicated for
multiple endpoints.

•	9 studies for children,

•	10 studies for older adults, and

•	18 studies for sex

We excluded the genetic factors population from our analysis, as we do not currently have the capability
to estimate health impacts among variant genotypes. We excluded diet, outdoor workers, and obesity
for similar reasons, as we have no representative dataset for use in analysis with these risk factors.
Effects on asthmatics were not included in this analysis because we currently lack highly resolved spatial
data on asthma prevalence, in part because effects on asthmatic populations are included in the main
analysis.84 We also excluded the SES group as the studies associated with the SES group for 03 were
associated with other methodological challenges. We coded the identified studies into a spreadsheet
and applied the initial screening criteria described previously. We collected information on mortality
and/or morbidity endpoints assessed in each study and focused on all-cause mortality and respiratory
morbidity endpoints. The remaining studies were evaluated based on the additional identification
criteria described in Table 52.

83	Analyses of effect modification will not be included in the main analyses, so as to avoid the possibility of double-
counting impacts. This potential uncertainty could also be described as the effect modification of individual risk
factors.

84	Effects on asthmatics using national level prevalence estimates are estimated in the main analysis.

113


-------
Table 52. 03 At-Risk Study Identification Criteria

Criterion

Notes

Peer-Review

Peer-reviewed research exclusively

Study Design

Epidemiologic study

Ozone Study

Research on ozone exposure is used

Study Location

U.S. or Canada

Population Attributes

Presents risk estimates for clearly defined at-risk groups for which data
currently exist in BenMAP-CE

Exposure Duration

Both short- and long-term exposure studies

Causal or Likely Causal
Health Effects

Adequate or suggestive evidence for at-risk groups in ISA

Economically Valuable
Health Effects

Health endpoints for which economic values have been or could
reasonably be developed

Baseline Incidence Data

Must be able to identify baseline incidence data for subpopulations

Season

All year exposure or warm season exposure

Ozone Exposure Metrics

MDA8, or able to be converted to MDA8

Lag Structure

Choose model that most clearly represents the relationship between
ozone exposure and the physiologic changes for the health endpoint

6.2.5.2 Total Mortality

Regarding the health endpoint of mortality, three studies for older adults met our criteria: Medina-
Ramon and Schwartz, 2008, Zanobetti and Schwartz, 2008, and Katsouyanni et al., 2009. All three
studies provided sufficient details to apply risk model information for short-term all-cause mortality
among adult age groups.85 There were no mortality studies for the child at-risk group in either the 2013
or 2020 03 ISAs (U.S. EPA, 2013, U.S. EPA, 2020a). For the at-risk population stratified by sex, two studies
met our initial criteria: Medina-Ramon and Schwartz, 2008, which evaluated short-term 03 exposure and
all-cause mortality, and Jerrett et al., 2009, which evaluated long-term 03 exposure and respiratory
mortality. Both studies provided sufficient data to apply risk model information to male and female
subpopulations. We developed health impact functions for these studies.

The 03-mortality risk estimates for at-risk subpopulations reported in Medina-Ramon and Schwartz,
2008 required additional modification in order to use those results to develop health impact functions.
The authors presented excess risk estimates for each subpopulation as the additional percent change in
mortality for persons who have the condition, compared to persons without the condition. For our
populations of interest, these subgroups were persons age 65 or older compared to those younger than
65, and females relative to males. However, they did not report the risk estimate for these comparison
groups, so in order to estimate the total excess risk for each exposed at-risk group, we needed to first
back-calculate the excess risk for the comparison group without the factor of interest. We accomplished
this by assuming that the authors' overall reported excess risk for the full sample of 0.65% (95%
confidence interval = 0.38% to 0.93%) could be expressed as a weighted average of the unreported
excess risk ("x") and the full excess risk for the at-risk group, which would be expressed as the sum of x

85 Calculations required to apply risk model information from Medina-Ramon and Schwartz, 2008 are described in
the following paragraph.

114


-------
and the reported excess risk from Medina-Ramon and Schwartz, 2008 Table 2, where the weights are
calculated using the total and at-risk group sample sizes in Table 1 of that paper. For example, to
calculate the total excess risks for the females in the sample, we used the following equation:

„D _ ERMaie(PopMale) + ERFemale (P°PFemale)

ERTotal —	t}

P°PTotal

where ERTotai is the full sample excess risk of 0.65%; ERFemaie is the excess risk of ozone exposures for
females; ERMaie is the excess risk of ozone exposures for males; PopTotai is the total sample population;
and PopFemaie and PopMaie are the size of the female and male subsets of the sample population,
respectively. We also know from Table 2 of that paper that:

ERFemaie = ERjy[aie + 0.58 %

Substituting and using the available information from Medina-Ramon and Schwartz, 2008 Tables 1 and
2, we can solve for ERMaie and then ERFemaie:

n ,rn/ ERMale(l,365,937) + (0.58% + ERMale)(l,363,703)

0.65% = 	

2,729,640

ERMaie — 0.36 %
and

ERFemaie = 0.36% + 0.58% = 0.94%

We then used the full excess risk value for the female subpopulation to derive a health impact function
for ozone-related mortality for females.

6.2.5.3 Respiratory Morbidity

We applied the same identification criteria described in section 6.2.5.1 to respiratory morbidity
endpoints for the child and sex at-risk groups. Three studies for children met our criteria for inclusion:
Mar and Koenig, 2009, Paulu and Smith, 2008, and Villeneuve et al., 2007. Each study evaluated
emergency room visits for asthma and provided sufficient risk model information stratified by age. No
studies cited for the older adult population met our criteria for inclusion. Three studies for sex met our
identification criteria: Paulu and Smith, 2008, Cakmak et al., 2006, and Lin et al., 2005. These studies
evaluated emergency room visits for asthma, all respiratory hospital admissions, and hospital
admissions for lower respiratory infection, respectively. Each study provided sufficient risk model
information for male and female subpopulations. We developed health impact functions for all studies
identified above. All the at-risk studies we identified are summarized in Table 53.

115


-------
Table 53. Identified 03 At-Risk Beta Coefficients and Standard Errors

At-Risk
Factor

Endpoint

Study

Subgroup

Beta

Coefficient
(SE)1

Lifestage,
older adults

Mortality, All Cause

Medina-Ramon and Schwartz,
2008 and Zanobetti and
Schwartz, 2008

Age 0-64

-0.0001
(0.0001)

Age 65+

0.0010
(0.0002)

Lifestage,
older adults

Mortality, All Cause

Katsouyanni et al., 2009

Age 0-74

0.0008
(0.0002)

Age 75+

0.0007
(0.0003)

Lifestage,
older adults

Mortality, All Cause

Zanobetti and Schwartz, 2008

Age 0-20

0.0001
(0.0003)

Age 21-
30

0.0001
(0.0004)

Age 31-
40

0.0001
(0.0002)

Age 41-
50

0.0001
(0.0002)

Age 51-
60

0.0005
(0.0002)

Age 61-
70

0.0004
(0.0001)

Age 71-
80

0.0005
(0.0001)

Age 81+

0.0003
(0.0001)

Sex

Mortality, All Cause

Medina-Ramon and Schwartz,
2008 and Zanobetti and
Schwartz, 2008

Female

0.0009
(0.0002)

Male

0.0004
(0.00004)

Sex

Mortality, Respiratory

Jerrett et al., 2009

Female

0.0044
(0.0011)

Male

0.0011
(0.0014)

Lifestage,
children

Emergency Room
Visits, Asthma

Mar and Koenig, 2009

Age 0-17

0.0104
(0.0044)

Age 18+

0.0039
(0.0027)

Lifestage,
children

Emergency Room
Visits, Asthma

Paulu and Smith, 2008

Age 2-14

0.0104
(0.0050)

Age 15-
34

0.0148
(0.0035)

Lifestage,
children

Emergency Room
Visits, Asthma

Villeneuve et al., 2007

Age 2-4

0.0032
(0.0033)

116


-------






Age 5-14

0.0073
(0.0024)

Age 15-
44

0.0058
(0.0018)

Age 45-
64

0.0063
(0.0033)

Age 65-
74

0.0073
(0.0055)

Age 75+

-0.0006
(0.0067)

Sex

Emergency Room
Visits, Asthma

Paulu and Smith, 2008

Female

0.0113
(0.0027)

Male

0.0104
(0.0032)

Sex

Hospital Admissions, All
Respiratory

Cakmak et al., 2006

Female

0.0013
(0.0004)

Male

0.0017
(0.0003)

Sex

Hospital Admissions,
Lower Respiratory
Infection

Lin et al., 2005

Female

0.0087
(0.0060)

Male

0.0040
(0.0052)

1 Beta coefficients and SEs in this table have been converted to MDA8 for comparability

6.3 Quantitative Characterization of Baseline Incidence Rate Uncertainties

When available from HCUP, we incorporate county-level hospital admissions and emergency
department visit baseline incidence data. Comparisons of the county-level data (box and whisker plot) to
the national-level data (red circles) are available in Figure 14.

Figure 14. Example County-Level and National-Level Emergency Department Visit and Hospital
Admission Baseline Incidence Data

Emergency
Department Visits

Cardiovascular
Respiratory

i	

—I • i	1-





Hospital Admissions

Cardio-, Cerebro- and Peripheral Vascular Disease
Respiratory Illness

h • i	

	\	







0.0000 0.0002

0.0004 0.0006 0.0008
Baseline Incidence Rates

00010 00012

We performed several quality assurance checks to ensure the incidence rates accurately reflect

observed health outcomes in the underlying counties. These checks included:

•	Examining data inputs to ensure the endpoints reflect the specified set of ICD codes from the
epidemiological studies;

•	Reviewing data processing scripts to ensure all calculations implement the intended procedures, as
documented in the BenMAP-CE User Manual (U.S. EPA, 2018);

•	Re-processing existing incidence rates in BenMAP-CE's "Other Incidence (2014)" to confirm that
changes to data processing to incorporate new endpoints have no or minimal impact on incidence
rate data for existing endpoints from BenMAP-CE's 2017 data update;

117


-------
•	Comparing the relative magnitude of related endpoints to ensure that incidence rates for broader
endpoints (e.g., HA, All Respiratory) are greater than incidence rates for endpoints with a narrower
set of ICD codes (e.g., HA, Asthma);

•	Comparing national baseline incidence counts when using county-level incidence rates and nation-
level incidence rates to ensure that, in aggregate, the two datasets produce similar results;86 and

•	Examining the geographic distribution of incidence rates to ensure no counties, states, or regions,
are characterized by anomalously low or high incidence.

We identified no systematic errors or bias in the raw data or data processing steps. The main source of
uncertainty in these data is related to imputation of rates where county data for specific endpoints were
suppressed due to statistical reliability or privacy concerns. The state or regional rates used to substitute
for these suppressed values may under- or over-estimate individual county rates.

6.4 Quantitative Characterization of Economic Valuation Estimate Uncertainties

6.4.1	Mortality Cessation Lag

Following advice from the Health Effects Subcommittee of EPA's independent Science Advisory Board
(SAB-HES), the agency typically assumes that some amount of time lapses between when air pollution is
reduced and when PM-attributable mortality is reduced fully. Within the context of benefits analyses,
this term is often referred to as "cessation lag." The duration of this lag affects how changes in PM-
attributable mortality associated with long-term (i.e., years-long) exposure are valued. Economic theory
suggests that the value of these future impacts should be discounted. The primary analysis included in
recent RIAs assumes that this lag is distributed over a 20-year period, with 30% of deaths reduced in
year 0, 50% occur in years 1-5, and the remaining 20% occur in years 6-20. This approach is generally
support by SAB recommendations (Cameron, 2001, Cameron, 2004, Hammitt and Bailar, 2010).

Based on SAB requests and recommendations, we previously performed several quantitative
uncertainty analyses with the goal of better understanding potential impacts of different cessation lag
distribution assumptions (U.S. EPA, 2012b). Although it was determined that certain extreme lag
structure assumptions may substantially impact monetized benefits, potentially increasing or decreasing
monetized impacts by up to 25%, for most reasonable distributed lag model structures, differences in
the specific shape of the lag function had relatively small impacts on overall PM2.5 benefits estimates.

We do not know how long-term 03 exposure-related respiratory deaths are distributed over time.

Hence, when discounting the value of 03-attributable deaths we use two lag structures originally
developed for PM2.5 (the 20-year segmented lag used for PM2.5 and an assumption of zero lag) as
sensitivity analyses.

6.4.2	Lung Cancer Cessation Lag

For a given health effect attributable to air pollution exposure, EPA reports the number of avoided cases
associated with the estimated pollutant reduction in a specified year. However, for some health effects,
there is an expected time lag between changes in pollutant exposure in a given year and the total
realization of health effect benefits, commonly referred to in regulatory analyses as the "cessation lag"

86 Aggregated county-level baseline incidence counts for all hospitalization and emergency department visit
endpoints included in the main benefits estimates were within 10% of the national baseline incidence counts.

118


-------
(section 6.4.1). For an outcome such as lung cancer, where the time between exposure and diagnosis
can be quite long, it may take decades to realize the full benefits of the air quality improvements.
Properly estimating the time course over which lung cancer health benefits are realized is critical for
proper discounting of the economic value of these health benefits.

Following guidance from EPA's Science Advisory Board (Ostro, 2004, EPA RIAs have applied a 20-year
distributed cessation lag model to estimate the temporal distribution of reductions in mortality risks,
including fatal lung cancer cases. In the 20-year distributed lag model, 30 percent of the total mortality
risk reductions occur in the first year following the exposure reduction, 50 percent are distributed
evenly among years two through five, and the remaining 20 percent are distributed evenly among years
six through 20. This structure reflects mortality risks from a variety of causes, with the mortality risk
reductions occurring later representing mortality risks from lung cancer.

For non-fatal cancer incidence, we considered a similar cessation lag approach based on estimates of
the lung cancer "latency period," or the time that passes between exposure and diagnosis, when
diseases processes may be occurring undetected and not yet resulting in observable symptoms. Based
on findings in the 2019 PM ISA, EPA has recently developed a health impact function based on
Gharibvand et al., 2017 for non-fatal lung cancer incidence. To support the new non-fatal lung cancer
risk estimate, we applied an age-at-diagnosis cessation lag distribution for the main analysis as it
accounts for age-specific latency period, instead of assuming a single latency duration. However, other
potentially applicable distribution models are available that also take into account the latency between
exposure and lung cancer diagnosis, such as the adapted 20-year distribution (section 6.4.2.1) and the
latency-based triangular distribution (section 6.4.2.2). All potential lag cessation distributions, including
the traditional 20-year lag distribution, are compared in section 6.4.2.3.

6.4.2.1	Adapted 20-Year Distribution

We adapted the 20-year distributed lag model applied to VSL estimates in previous EPA RIAs using the
estimated 10-year latency period. Following the latency period, the adapted 20-year model has zero
cancer case reductions in years one through five and an even distribution of case reductions in years six
through 20, resulting in scaling factors of 0.71 for a 3% discount rate or 0.46 for a 7% discount rate.

6.4.2.2	Latency-Based Triangular Distribution

A continuous probability distribution shaped like a triangle may better assess lung cancer lag cessation.
Triangular distribution based on a search of lung cancer latency periods from the peer-reviewed
literature. Using the most common latency period of 10-years observed in the literature (Table 29), we
estimated a triangular distribution that spans from five to 20 years, with the peak of the distribution at
ten years, the most common latency period estimate found in the literature (i.e., the mode). We
identified a triangular distribution to reflect the uncertainty of latency period duration found in the
literature, given the limited amount of information available to establish the shape and form of an
uncertainty distribution. We used the cumulative probability function for this distribution to estimate
the incremental annual number of cases likely to be diagnosed year to year by subtracting the
cumulative probability from the previous year from the cumulative probability of the current year. We
then used the resulting percentages to create a cessation lag model, allocating cases avoided in the
years following an exposure change according to the corresponding yearly percentages.

119


-------
6.4.2.3 Comparison of Lung Cancer Lag Cessation Distribution Models

The effect of each potential cessation lag distribution model was converted into scaling factors (Table
54). The scaling factors of the adjusted 20-year lag distribution estimate falls between that of the
traditional cessation lag and the triangular distribution lag estimates. Also, the adjusted cessation lag
distribution underestimates as compared to the age-at-diagnosis distribution.

Table 54. Scaling Factors for Various Lung Cancer Lag Cessation Distribution Models

Discount Rate

Age Range

Scaling Factor

Lag Cessation Distribution Model

3%

65-99

0.668980939

Traditional VSL cessation lag, 3% DR

7%

65-99

0.398456232

Traditional VSL cessation lag, 7% DR

3%

30-99

0.70711338

Adjusted 20-Year Distributed Lag Adjustment Factor

7%

30-99

0.463225599

Adjusted 20-Year Distributed Lag Adjustment Factor

3%

30-99

0.72192703

Triangular Adjustment Factor

7%

30-99

0.480176715

Triangular Adjustment Factor

3%

30-34

0.350285148

SEER Age-Distribut

on Adjustment Factor

3%

35-44

0.427591186

SEER Age-Distribut

on Adjustment Factor

3%

45-54

0.553445022

SEER Age-Distribut

on Adjustment Factor

3%

55-64

0.675599356

SEER Age-Distribut

on Adjustment Factor

3%

65-74

0.775053763

SEER Age-Distribut

on Adjustment Factor

3%

75-84

0.843760064

SEER Age-Distribut

on Adjustment Factor

3%

85-99

0.916741635

SEER Age-Distribut

on Adjustment Factor

7%

30-34

0.108397669

SEER Age-Distribut

on Adjustment Factor

7%

35-44

0.168901798

SEER Age-Distribut

on Adjustment Factor

7%

45-54

0.294643444

SEER Age-Distribut

on Adjustment Factor

7%

55-64

0.445786107

SEER Age-Distribut

on Adjustment Factor

7%

65-74

0.590393871

SEER Age-Distribut

on Adjustment Factor

7%

75-84

0.702750875

SEER Age-Distribut

on Adjustment Factor

7%

85-99

0.82379138

SEER Age-Distribut

on Adjustment Factor

Using the lung cancer incidence risk estimates and a hypothetical scenario, we compared the three
potential lung cancer cessation lag models. The annual reduction in cancer cases was estimated from
zero to 100 years after the exposure change. For the triangular and adjusted 20-year distributed lag, all
annual reductions occur within 20 years after exposure change and for the age of diagnosis distribution,
all annual reductions fall within 67 years after exposure change (Figure 15) with 90% occurring by year
26.

120


-------
Figure 15. Lung Cancer Cases Cessation Lag Distribution by Model

O

(/)
05

Sg 100

150-

50

0-

Distribution
~~ Trianglar

Adjusted 20-Year Distributed Lag
SEER Age of Diagnosis

0

25

50

Year

75

100

A potential limitation of the triangular distribution and adjusted 20-year distributed lag is that the same
latency period is used for all ages. For an exposure change experienced at 30, both the triangular
distribution and adjusted 20-year distributed lag estimate that reductions occur between ages 35 and
50. However, SEER data indicates that less than 5% of lung and bronchus cancer diagnoses occur during
this period. Conversely, for an exposure change experienced at 90, the reductions are realized from ages
95 to 110 (greater than life expectancy).

A limitation of the age-of-diagnosis distribution methods is that the highest reductions occur in the first
five years for the age-of-diagnosis distribution and not all ages display latency periods (Figure 15). The
factor used in this method estimates the time pattern of benefits based on the percentage of cancer
incidence remaining in the life and results in older age bins without latency periods (Figure 16). While an
age-dependent latency period may more accurately reflect the diagnosis data, the age-of-diagnosis
distribution method may overestimate case reductions in earlier years by assuming all reduced cases for
a change in exposure at later ages are realized by the end of life (age 99). At the same time, some cases
are delayed by two to five decades, beyond latency values reported in the literature for lung cancer.

121


-------
Figure 16. Lung Cancer Cases Reduction Distribution

Age at Exposure Change: 33

Age at Exposure Change: 40

Distribution

SEER Age of Diagnosis



Distribution

SEER Age of Diagnosis

0 25 50 75 100
Age

Age at Exposure Change: 50

0 25 50 75 100
Age

Age at Exposure Change: 60

Distribution	g

CO

SEER Age of Diagnosis ¦§

Distribution

SEER Age of Diagnosis

0 25 50 75 100

0 25 50 75 100
Age

Age at Exposure Change: 70

Age at Exposure Change: 80

Distribution

SEER Age of Diagnosis

Distribution

SEER Age of Diagnosis

L

0 25 50 75 100
Age

Age at Exposure Change: 93

0 25 50 75 100
Age

0.4

s 03

J5

to

.Q

£ 0 2
0.1

Distribution

— SEER Age of Diagnosis

0 25 50 75 100
Age

6.4.3 Income Elasticity of Willingness to Pay

The degree to which one's WTP to reduce the risk of adverse effects changes in proportion to future
changes in income is uncertain. We previously evaluated the potential impact of this factor on the
monetized benefits in a sensitivity analysis (U.S. EPA, 2012b). Results are available below.

Our estimates of monetized benefits account for growth in real gross domestic product per capita by
adjusting the WTP for individual endpoints based on the central estimate of the adjustment factor for
each of the categories (minor health effects, severe and chronic health effects, mortality, and visibility).

122


-------
We previously examined how sensitive the estimate of total benefits is to alternative estimates of the
income elasticities. Table 55 lists the ranges of elasticity values used to calculate the income adjustment
factors, while Table 56 lists the ranges of corresponding adjustment factors. The results of this
sensitivity analysis, giving the monetized benefit subtotals for the four benefit categories, are presented
in Table 57.

Table 55. Ranges of Elasticity Values Used to Account for Projected Real Income Growth3

Benefit Category

Lower Sensitivity Bound

Upper Sensitivity Bound

Minor Health Effect

0.04

0.30

Mortality

0.08

1.00

derivation of these ranges can be found in Kleckner and Neumann, 1999. COI estimates are assigned an
adjustment factor of 1.0.

Table 56. Ranges of Adjustment Factors Used to Account for Projected Real Income Growth3

Benefit Category

Lower Sensitivity Bound

Upper Sensitivity Bound

Minor Health Effect

1.018

1.147

Mortality

1.037

1.591

3Based on elasticity values reported in Table 55, U.S. Census population projections, and projections of real GDP
per capita.

Table 57. Sensitivity of Monetized Benefits to Alternative Income Elasticities3

Benefit Category

Benefits Incremental to Baseline (Millions of 2006$)

Lower Sensitivity Bound

Upper Sensitivity Bound

Minor Health Effect

$30

$31

Mortality15

$3,600

$3,800

3AII estimates rounded to two significant digits.

bUsing mortality effect estimate from Krewski et al., 2000 and 3% discount rate. Results using Laden et
al. (2006) or a 7% discount rate would show the same proportional range.

Consistent with the impact of mortality on total benefits, the adjustment factor for mortality has the
largest impact on total benefits. The value of mortality in 2020 ranges from 96% to 108% of the main
estimate based on the lower and upper sensitivity bounds on the income adjustment factor. The effect
on the value of minor health effects is much less pronounced, ranging from 86% to 133% of the main
estimate for minor effects.

6.4.4	Statistical Estimates of VSL

EPA relies on published peer-reviewed studies to provide statistical estimates of the value of avoided
statistical mortality risk (VSL). These studies provide a range of differ estimates due to varying study
design and different statistical samples. EPA uses a distribution of values fit to the studies' estimates as
described in Section 5.1.1 and Table 22.

6.4.5	Alzheimer's Disease and Parkinson's Disease Onset Lifetime Costs

The epidemiologic study from with the risk estimates for Alzheimer's and Parkinson's disease were
identified used time to first hospital admission as the health endpoint readout. As the authors note that
this is not necessarily indicative of disease onset, we only include valuation estimates of associated

123


-------
hospital admissions costs in the main benefits assessment. However, we include information here
regarding how the benefits estimates would increase if the first hospital admission were used as a
surrogate for disease onset.87

6.4.5.1 Alzheimer's Disease

Potential valuation sources of Alzheimer's disease lifetime medical costs were available from the
Alzheimer's Association, 2020 report and Jutkowitz et al., 2017. Using Alzheimer's Association, 2020, we
first developed an estimate of incremental annual medical expenses for Medicare beneficiaries living
with Alzheimer's Disease (Table 58). Then, using the estimated life expectancy duration of 5 year from
Jutkowitz et al., 2017, 3% and 7% discounted costs were extrapolated (Table 59). We note that the
average/median age of Alzheimer's disease diagnosis/onset is after the age of 65, at which we assume
retirement, so any potential lost wages are not included in this valuation estimate. Lifetime medical
costs, excluding hospitalization, are estimated at $156,920 using a 3% discount rate or $145,946 using a
7% discount rate in 2015$ Alzheimer's Association, 2020.

Table 58. Annual Alzheimer's Disease Valuation Estimate Calculation

Service

Beneficiaries with Alzheimer's
Disease or Other Dementia

Beneficiaries without
Alzheimer's Disease or Other
Dementia

Inpatient hospital

$11,465

$3,703

Medical provider

$5,762

$3,589

Skilled nursing facility

$7,213

$493

Nursing home

$16,523

$800

Hospice

$2,126

$161

Home health care

$2,661

$386

Prescription Medications

$3,481

$2,986

Annual Medical Expenses ($2019)

$49,231

$12,118

Annual Medical Expenses ($2015)

$44,128

$10,862

Incremental Annual Medical
Expenses for Medicare
beneficiaries with AD ($2015)

$33,266



Table 59. Lifetime Alzheimer's Disease Valuation Estimate Calculation (2015$)

Year

3% Discount Rate

7% Discount Rate

0

$33,266

$33,266

1

$32,297

$31,090

2

$31,357

$29,056

3

$30,443

$27,155

4

$29,557

$25,379

Total Lifetime Costs

$156,920

$145,946

87 Baseline incidence and prevalence data would need to be updated to estimate impacts of disease onset.

124


-------
Jutkowitz et al., 2017 provided information needed to separately develop a lifetime Alzheimer's Disease
cost estimate with a 3% discount rate, but not with a 7% discount rate (Table 60. Additional Lifetime
Alzheimer's Disease Valuation Estimate Calculation with a 3% Discount Rate (2015$)). As the 3%
discount rate estimate of $156,920 from Alzheimer's Association, 2020 is fairly similar to the lifetime 3%
discount rate estimate of $184,500 from Jutkowitz et al., 2017, we have additional confidence in the
validity of the Alzheimer's Association, 2020 estimates (Table 21).

Table 60. Additional Lifetime Alzheimer's Disease Valuation Estimate Calculation with a 3% Discount
Rate(2015$)

Service

Base-Case (83-year-old

Counterfactual

Incremental Increase



incident dementia case)

Dementia Free

in Lifetime Costs

Value of informal caregiving

$135,300

$2,460

$132,840

Out-of-pocket expenditures

$89,840

$64,720

$25,120

Medicaid expenditures

$44,090

$37,450

$6,640

Medicare expenditures

$52,540

$32,650

$19,890

Total value

$321,780

$137,280

$184,500

6.4.5.2 Parkinson's Disease

Yang et al., 2020 provided estimates of lifetime costs, including direct, indirect, and non-medical costs.
Using Yang et al., 2020, we first developed an annual estimate of excess costs associated with living with
Parkinson's Disease for one year (Table 61). Then, using the estimated life expectancy duration of 14.6
years from De Pablo-Fernandez et al., 2017, the 3% and 7% discounted costs were extrapolated (Table
62). Lifetime medical costs are estimated at $567,285 using a 3% discount rate or $445,792 using a 7%
discount rate in 2015$ (Table 21).

125


-------
Table 61. Annual Parkinson's Disease Valuation Estimate Calculation

Service

Excess Cost per
Person with
Parkinson's Disease

Description

Direct Medical Costs

Non-acute institutional care

$6,888

Quantify excess healthcare cost of
each person with Parkinson's
Disease compared with 10 matched
individuals without Parkinson's
Disease

Hospital inpatient

$6,932

Outpatient

$5,308

Physician office

$1,182

Durable medical equipment

$140

Prescription medication

$3,988

Direct Medical Costs Subtotal

$24,438

Indirect Medical Costs

Paid daily non-medical care

$3,709

Home caretakes/long-term care
facilities

Home modification

$2,151



Motor vehicle modification

$897



Other expenses

$508



Indirect Medical Costs Subtotal

$7,265



Non-Medical Costs

Reduced employment

$2,579

Reduced labor market participation
due to early retirement

Absenteeism

$4,869

Lost work days

Presenteeism

$2,841

Lost work productivity at work

Social productivity loss in volunteer
work

$997

Supplemental security income (SSI)

$541

SS disability supplemental income

Social security disability insurance
(SSDI)

$1,617

Other disability income

$2,431

Includes other disability income
sources such as VA disability, gov't
employee disability, & state
disability insurance or personal
disability insurance payments

Non-Medical Costs Subtotal

$18,293



Annual Medical Expenses ($2017)

$47,578



Annual Medical Expenses ($2015)

$44,718



126


-------
Table 62. Lifetime Parkinson's Disease Valuation Estimate Calculation

Year

3% Discount Rate

7% Discount Rate

0

$44,718

$44,718

1

$43,416

$41,793

2

$42,151

$39,059

3

$40,924

$36,503

4

$39,732

$34,115

5

$38,574

$31,883

6

$37,451

$29,798

7

$36,360

$27,848

8

$35,301

$26,026

9

$34,273

$24,324

10

$33,275

$22,732

11

$32,305

$21,245

12

$31,364

$19,855

13

$30,451

$18,556

14

$29,564

$17,343

14.6

$17,427

$9,992

Total Lifetime Costs (14.6 yr survival)

$567,285

$445,792

6.5 Qualitative Characterization of Uncertainties

There are several uncertainties we are unable to fully or partially quantitatively assess, but qualitatively
discuss below, in alphabetical order.

6,5.1 Applying Risk Estimates to Locations and Populations not Specified in the Epidemiologic
Study

EPA regulatory actions often affect portions of the country and populations that differ from those
considered in the epidemiologic studies providing the risk estimates. EPA commonly transfers risk
estimates from one location or population to another, following a procedure called benefits transfer, a
potential source of uncertainty. When available, risk estimates based on nationwide studies reflecting
the overall population demographics of U.S. residents will be used when estimating health benefits.
Epidemiologic studies exploring the relationship between air pollution and the risk of mortality often
consider populations whose characteristics are broadly representative of the U.S. (e.g., Medicare-based
estimates will be applied to those >64). By contrast, epidemiologic studies examining morbidity
outcomes may focus on population subsets, such as those residing in specific geographic regions, a
single sex, or selected races/ethnicities. In this context, two or more epidemiologic studies may report
risk estimates that, when pooled, can better characterize risks experienced by U.S. populations.
However, in some cases it may be scientifically inappropriate to pool risk estimates—for example, a
nationwide analysis of populations ages 65-99 and a less-geographically diverse analysis of populations
ages 0-99. In a situation such as this, the estimate best characterizing risk in the U.S. will be included in
the main benefits assessment and the others will be included in quantitative sensitivity analyses
(sections 6.1 and 6.1.6).

127


-------
6.5.2	Causality Determination

When estimating air pollutant-attributable health impacts, EPA focuses on endpoints for which the
underlying scientific evidence is strongest. This approach is based on evaluating evidence across
scientific disciplines (i.e., clinical, animal toxicological, and epidemiologic) with regard to consistency of
effects within a discipline, coherence of effects across disciplines, and evidence of biological plausibility.
Such an approach gives us greater confidence in the relationship between exposure and health
outcome. For criteria pollutants, EPA typically relies upon the causality determinations in the latest ISA
or equivalent, which are made using a weight-of evidence approach. These causality determinations are,
however, made for categories of health effects and not for specific endpoints. Thus, the extent to which
the relationship exists for the specific endpoint and the exposure circumstances of interest in a benefits
assessment is a source of uncertainty.

An expert elicitation sponsored by EPA to characterize the uncertainty in the relationship between PM2.5
and mortality, including causal uncertainty, was released in 2006 and reviewed by the Advisory Council
on Clean Air Compliance (Hammitt, 2008, lEc, 2006). Although the 12 expert-defined concentration-
response functions provide useful information on the sensitivity of the health benefits estimates,
additional epidemiology literature which addresses some of the weaknesses identified by the expert
elicitation has since become available, such as improved exposure estimation techniques and the use of
cohorts more representative of the U.S population. For these reasons we do not include the expert-
derived results as a sensitivity analysis here but consider it as qualitative support for the relationship
between long-term PM2.5 exposures and all-cause mortality impacts.

Causal inference is another method of establishing a causal connection that evaluates associations
under changing conditions. The 2019 PM ISA stated that "overall, the results of these causal inference
studies contribute to the body of epidemiologic evidence that informs the causal relationship between
long-term (one month to years) PM2.5 exposure and total mortality (U.S. EPA, 2019c). Observing
consistent results for this relationship across studies using different analytic techniques (i.e., difference-
in-difference approach) increases our confidence in the relationship."

6.5.3	Estimating and Assigning Exposures in Epidemiology Studies

New developments in exposure assessment, including hybrid spatiotemporal models that incorporate
satellite observations of aerial optical density, land use variables, surface monitoring data from
monitors, and chemical transport models, have led to improvements in the spatial resolution extent of
pollutant concentration surfaces. After reviewing the current state of exposure science, the 2019 PM ISA
stated that "a number of studies demonstrate that the positive associations observed between long-
term PM2.5 exposure and mortality are robust to different methods of assigning exposure" and the 2020
03 ISA articulated that "hybrid methods have produced lower error predictions of ozone concentration
compared with spatiotemporal models using land use and other geospatial data alone but may be
subject to overfitting given the many different sources of data incorporated into the hybrid framework."

Although these advancements may reduce bias and uncertainty in risk estimates, the accuracy of hybrid
exposure estimates is difficult to confirm in areas lacking monitors. On the other hand, studies using
monitor data as the only exposure information have increasing exposure uncertainty the farther people
live from the monitor site.

128


-------
Uncertainties related to PM2.5 and 03 exposure estimation vary. For example, the PM2.5 HBM method
had Pearson R's ranging from 0.91 to 0.94 when applied across the U.S. at a 36-km resolution,
depending on the geoimputation approach of the CMAQ data (U.S. EPA, 2019c). The evaluation of the
03 HBM method has been relatively limited. However, overall conclusions regarding long-term 03
exposure estimates that include fixed-site monitor measurements are that "the true effect of long-term
exposure to ambient ozone may be underestimated or overestimated by the model" and that it "is
much more common for the effect estimate to be underestimated, and the bias is typically small in
magnitude" (U.S. EPA, 2020a).

6.5.4	Differential Toxicity of PM2.5 According to Chemical Composition

PM2.5 is a heterogenous mixture of solid and liquid particles suspended in air and can vary with regards
to size, composition, and source. The 2020 PM ISA found that "across exposure durations and health
effects categories...many PM2.5 components and sources are associated with many health effects, and
the evidence does not indicate that any one source or component is consistently more strongly related
to health effects than PM2.5 mass;" although, it was also noted that "most studies that examine PM
sources and components focused on PM2.5" (U.S. EPA, 2019c).

Since the 2019 PM ISA concluded that "recent studies continue to demonstrate that no individual PM2.5
component or source is a better predictor of mortality than PM2.5 mass" and "many PM2.5 components
and sources are associated with many health effects and that the evidence does not indicate that any
one source or component is consistently more strongly related with health effects than PM2.5 mass" we
continue to assume that all fine particles, regardless of their chemical composition, are equally potent in
causing mortality and do not quantitatively assess uncertainties related to potential differences in PM2.5
toxicity or composition. A qualitative discussion of this uncertainty as it relates to respiratory effects,
cardiovascular effects, and mortality can be found in section 1.5.4 of the 2019 PM ISA (U.S. EPA, 2019c).

6.5.5	Different Long-Term Exposure Windows

The delay between changes in exposure and changes in health is an empirical challenge in estimating
potential health effects associated with air pollution exposure. For example, if health impacts of high
pollutant exposures have a long latency, risk estimates attributing to lower pollutant concentrations
experienced more recently may be biased away from the null. However, the 2019 PM ISA states that
"new evidence from recent studies continues to support the previous conclusion that health benefits
from reducing air pollution could be expected with a few years of intervention" (U.S. EPA, 2019c). This
issue is likely less relevant to 03 exposure-attributable mortality, as those studies often have very
similar, if not overlapping, health and air quality data.

6.5.6	Discounting Future Benefit Estimates

Discounting reflects that people prefer benefits presently more than in the future. When appropriately
applied, discounting can allow for the direct comparison of future benefits to costs. However, there are
potential uncertainties associated with discounting future benefit estimates. EPA bounds discounted
benefits and costs using an estimate of the consumption rate of interest and a rate of return on private
capital given that the share of capital that is displaced by a regulation is unknown. OMB currently
recommends, and EPA uses, a real consumption rate of discount of 3% and a real rate of 7% for the
opportunity cost of private capital based on prior empirical estimates (OMB, 2003). These values are

129


-------
estimates and therefore introduce uncertainty. Additional detail on discounting can be found in the EPA
Guidelines for Preparing Economic Analyses (U.S. EPA, 2014).

6.5.7	Statistical Estimates of WTP

EPA relies on published peer-reviewed studies to provide statistical estimates of the value of avoided
pain and suffering (WTP). While most of these studies provide estimates of the uncertainty due to
statistical sampling, there are other important sources of error. First, the statistical models used to
produce these estimates may be incorrect, termed modeling error. Second, the statistical samples used
to produce these estimates may be selectively chosen and unlike the population of interest, leading to
selection error. Third, WTP values are unavailable for many health endpoints of interest. Assigning a
value of zero is clearly incorrect, but the EPA has no basis on which to assign other values.

6.5.8	Confounding by Individual Risk Factors

Interindividual variability in both physiological responses and exposures to ambient air pollution can
affect the size of reported risk estimates in epidemiologic studies. Well-designed epidemiology studies
account for individual risk factors as covariates in their models88, and so all else being equal we identify
risk estimates adjusted to control for the most covariates that could reasonably impact the risk
estimate. However, confounding by individual risk factors remains a potential source of uncertainty as
additional relevant covariates may exist that are not included as covariates in epidemiological risk
estimates used for health benefits assessment. Unfortunately, we are currently unable to quantitively
assess this area of uncertainty but will include qualitative discussions when possible.

6.5.9	Confounding by Other Pollutants

When considering the relationship between pollutant exposure and health impacts, it can be
informative to consider whether risk estimates are changed when other pollutants are included in
copollutant models, especially when health impacts of multiple pollutants are being estimated
concurrently. Regarding long-term exposures, the 2019 PM ISA concluded that "positive associations
observed between long-term PM2.5 exposure and total mortality remain relatively unchanged after
adjustment for...N02 and PM10-2.5" and the 2020 03 ISA found that "positive associations observed
between long term 03 exposure and total mortality remain relatively unchanged after adjustment
for...N02." However, confounding due to the effects of copollutants other than 03 and PM2.5 are a
potential source of uncertainty.

6.5.10	Risk Attributable to Long-Term and Short-Term Exposures

Long- and short-term exposures may follow similar or divergent biological pathways. When pathways
are similar, estimates of impacts from long-term exposures may include short-term impacts and vice-
versa. However, if pathways diverge, long- and short-term impacts may be the sum, or even greater
than the sum, of the two exposure duration impacts. As there is little research directly comparing long-
and short-term effects, we are currently unable to quantitatively assess this area of uncertainty for
either PM2.5- or 03-attributable health effects.

88 Common covariates include education level, marital status, body mass index, cigarette smoking, diet,
occupational exposures, income, percentage of minority, unemployment, and poverty.

130


-------
6.5.11	Heterogeneity of Risk Estimates

Epidemiologic studies often differ according to study design, geographic locations, age groups,
population attributes, study size, methods for estimating exposure, range of pollutant concentrations,
time periods, study sizes, and follow-up durations. These differences in turn influence the magnitude
and standard error of study-reported risk estimates. The diversity of identified risk estimates could
reflect either the variability across the populations studied or uncertainty around the risk estimates.
Importantly, heterogeneous risk estimates are not necessarily indicative of bias, but could also result
from variability of the underlying input parameters.

6.5.12	03 Metrics89

03 exposure metrics used to develop risk estimates can take many forms, though the most widely used
metrics are the maximum daily 8-hour average (MDA8), daily average 24-hour (DA24), daily average 8-hr
from 10AM to 6PM (DA8), and maximum daily 1-hour average (MDA1) metrics.

Historically, if epidemiologic studies developed risk estimates based on 03 metrics other than MDA8,
EPA would adapt the risk estimates based on average conversion ratios to be appropriate for use with
air quality surfaces projected in the MDA8 metric (Anderson and Bell, 2010). This approach brings with it
uncertainties associated with the simplifying assumptions used to develop the conversion ratios. In most
cases, the day to day variation in different metrics (e.g., DA24 vs MDA8) is highly correlated. As such, the
relationships between health impacts and different ozone metrics will also be highly correlated.
However, when we apply risk estimates derived from time series results to evaluate the impacts of a
specific policy scenario, we focus most on the shift in the overall distribution of 03 concentrations over
an entire season, instead of on the day to day variation in 03 levels. Because specific policy scenarios
might result in different temporal distributions of ozone concentrations than was observed in the
monitored ozone data used in the studies, it is important to choose an 03 metric that is best suited to
capturing changes in 03 that are likely to occur during hours where populations are likely to be exposed.

6.5.12.1 Converting O3 Risk Estimate Exposure Metrics

When epidemiologic risk estimates are developed using non-MDA8 03 exposure metrics, EPA has
typically converted the beta risk estimates into MDA8 metrics, which brings in a potential source of
uncertainty (Anderson and Bell, 2010). We discuss uncertainties associated with converting various
common 03 exposure metrics into the MDA8 metric below.

6.5.12.1.1 DA24 to MDA8

Currently, air quality projections using the MDA24 metric are unavailable, so a conversion factor from
Anderson and Bell, 2010 is used in order to apply these risk estimates to MDA8 air quality surface
projections. We multiply the beta risk estimate by the inverse of the median summer ratio of MDA8 to
DA24 mean 03 concentrations (i.e., 1 / 1.53 = 0.6536) for studies assessing summer 03 exposure or by
the inverse of the fixed effects average ratio of MDA8 to DA24 mean 03 concentrations (i.e., 1 / 1.53 =
0.6536) for studies assessing all-year 03 exposure. We note that Anderson and Bell, 2010 included a
range of ratios from 1.23-1.83.

89 PM2.5 exposure metrics are not discussed here as the vast majority are based on daily 24-hour average
concentrations and annual exposures are often estimated using daily 24-hour average concentrations. Importantly,
this potential source of uncertainty is not likely to have a large effect on overall PM2.5 benefits estimates results.

131


-------
6.5.12.1.2 DA8 to MDA8

A comparison of the MDA8 and DA8 metrics using 20 years of 03 monitoring data (2000-2019) for the
entire contiguous U.S. resulted in a very high rate of correlation (Figure 17). The correlation was based
on a simple linear regression with zero intercept, meaning that if the MDA8 is 0, then the DA8 mean
must also be zero. The green line is the regression line and the light gray line represents a 1:1
relationship. Please note, the MDA8 cannot exceed the DA8 and the high density of the ~7 million points
shown in the graph cluster near the 1:1 line. In fact, the MDA8 and DA8 metrics are identical
approximately 30% of the time and differ by 2 ppb or less about 80% of the time. Based on this
comparison, the conversion factor from DA8 to MDA8 is 0.97.

Figure 17. Correlation of MDA8 and DA8 03 Exposures Between 2000-2019 (R=0.986)

0	50	100	150	200

Daily 8-hour Mean 03 (ppb)

6.5.12.1.3 MDA1 to MDA8

Due to time and resource limitations, air quality projections using the MDA1 metric are also unavailable
for the Revised CSAPR Update final rulemaking, so a conversion factor from Anderson and Bell, 2010 is
used in order to apply these risk estimates to MDA8 air quality surface projections. We multiply the beta
risk estimate by the median summer ratio of MDA1 to MDA8 03 concentrations (i.e., 1.13) for studies
assessing summer 03 exposure or by the fixed effects average ratio of MDA1 to MDA8 03 concentrations
(i.e., 1.14) for studies assessing all-year 03 exposure. We note that Anderson and Bell, 2010 included a
range of ratios from 1.08-1.26.

132


-------
6.5.13	O3 Season

Studies of 03 vary with regards to 03 season, limiting analyses to various definitions of summer (e.g.,
April-September, May-September or June-August) and exposures over the full calendar year. These
differences can reflect state-specific, EPA-defined 03 seasons or another seasonal definition chosen by
the study author. 03 exposure estimates are arguably more accurate during the summer when
concentrations are typically higher and more monitors are operational. In addition, respiratory effects
associated with short-term exposures are commonly limited to the warm season and therefore reflect
the incidence that occurs during the 5- or 6-month 03 season (U.S. EPA, 2020d). Recently, there are an
increased number of long-term analyses of 03-attributable health impacts over the full calendar year
using hybrid modeling techniques and where 03 monitoring data is collected for the entire year. These
studies likely represent a more complete estimate of 03-attributable health impacts.

While epidemiologic studies assessing all-year 03 exposures would likely present more comprehensive
estimates of health impacts, hybrid 03 surface projections for baseline and policy rulemaking scenarios
are not currently available.90 As such, we identified epidemiologic studies and associated risk estimates
that evaluated associations between exposures and warm season effects when available. There was
some variability amongst the warm season definitions within the list of studies identified in this update
(e.g., April-September and June-August), although only the respiratory emergency department and
asthma symptom risk estimate was based on full year 03 exposures (Barry et al., 2019, Lewis et al.,
2013). It should be noted that the exposures for asthma symptoms among the identified studies were
not evenly distributed across all the seasons (I.e. three in Spring, two in Summer, two in Fall, and one in
Winter).91

There is some variability regarding the definition of the warm season amongst epidemiologic studies
included in the ISAs and the main risk estimates identified here for 03 benefits estimates. When there is
a substantial difference, such as the June-August warm season assessed by Zanobetti and Schwartz,
2008, we develop season-specific air quality projections, when feasible. However, many studies begin
the 5-7 month warm season in either April or May and conclude the season in September or October.
Since projected full year hybrid 03 surfaces are currently not available, epidemiologic risk estimates will
be applied to the air quality projection most closely matching the exposure season in the study (e.g.,
April-September exposures will be applied to May-September air quality projections). We expect this
seasonal mismatch will only have a limited effect on the magnitude of related health incidence.

6.5.14	Shape of the Concentration-Response Relationship
6.5.14.1 PMzs

An important consideration when characterizing uncertainty is whether the concentration-response
relationship is linear across the full concentration range that is encountered, or if there are
concentration ranges where there are departures from linearity. Overall, evidence from the 2019 PM

90	The paucity of O3 monitoring data in winter months potentially complicates the development of full year
projected O3 surfaces, which would need to be subject to comprehensive evaluation prior to use in EPA RIAs.

91	When risk estimates based on full-year, long-term O3 exposures are applied to warm season air quality
projections, the resulting benefits assessment may underestimate impacts, due to a shorter timespan for impacts
to accrue. When risk estimates based on full-year, short-term O3 exposures are applied to warm season air quality
projections, the resulting benefits assessment may also underestimate impacts, as short-term O3 exposure effects
are typically larger during the warm season (U.S. EPA, 2020a).

133


-------
ISA continues to "support a linear, no-threshold concentration-response relationship for long-term
exposure to PM2.5 and total (nonaccidental) mortality, especially at lower ambient PM2.5 concentrations,
with confidence in some studies in the range of 5-8 ng/m3" and "there is less certainty in the shape of
the concentration-response curve at mean annual PM2.5 concentrations generally below 8 ng/m3,
although some studies characterize the concentration-response relationship with certainty down to 4
Hg/m3" (U.S. EPA, 2019c).

Although ten large cohort studies of long-term PIVh.s-attributable mortality observed linear, no-
threshold concentration-response relationships, three Canadian studies presented evidence of
deviations from linearity down to the lowest concentration evaluated. Two studies found evidence of a
supralinear relationship at lower concentrations, although only one was statistically significant. And a
single study found that the best fit for the long-term PM2.5 mortality relationship was in a threshold
model with a threshold at 11 ng/m3.

There are several potential explanations for these results, one of which is that studies may be unable to
adequately evaluate the relationship at low levels without sufficient population exposure at those levels.
Consistent with that hypothesis, the single statistically significant study finding evidence of
supralinearity did have one of the lowest mean PM2.5 concentrations, at 6.3 ng/m3. Another possible
explanation with support from the 2019 ISA is that the shape of the concentration-response relationship
could differ by health outcome.

Although there were no evaluations of the shape of the long-term PM2.5-attributable respiratory
mortality relationship in the 2019 PM ISA, there were several studies of the relationship between long-
term PM2.5exposure and cardiovascular disease. When considering long-term PM2.5-attributable
cardiovascular mortality, again most results "continue to support a linear, no-threshold
relationship...especially at lower ambient concentrations of PM2.5...[with] a number of the
concentration-response analyses include concentration ranges <12 ng/m3." As with total mortality, a
few studies found that risk was greater at lower concentrations, although the deviation from linearity
was not statistically significant. The only evidence of nonlinearity in the long-term PIVh.s-attributable
cardiovascular mortality relationship came from two studies by the same group, which included
exposure from cigarette smoking. They observed that the concentration-response relationship was
much steeper at lower PM2.5 concentrations, such as those due to ambient air pollution, than at the
higher concentrations associated with cigarette smoking.

There were a small number of studies of the relationship between long-term PM2.5 exposure and
cardiovascular morbidity endpoints in the 2019 PM ISA. A study of hypertension and another of ischemic
heart disease incidence found no deviations from linearity. Two studies of coronary artery calcification
found evidence of deviations from linearity, but the direction of the results was inconsistent. One study
found evidence of sublinearity at higher concentrations while the other found evidence of supralinearity
at both high and low concentrations.

The shape of the relationships between PM2.5 exposure and health effects may also include a threshold,
or PM2.5 exposure concentration below which human health is not adversely impacted. Although
evidence does not currently support the existence of a measurable PM2.5 exposure population-level
threshold, prior higher concentration exposures with longer latency periods could make thresholds
difficult to detect. However, the 2019 PM ISA states that "new evidence from recent studies continues

134


-------
to support the previous conclusion that health benefits from reducing air pollution could be expected
with a few years of intervention/' reducing the likelihood of this potential source of uncertainty.

Based on the evidence and lack of nonlinear relationships between long-term PM2.5 exposure and health
impacts, we continue to assume a linear, no-threshold relationship and do not quantitatively assess
uncertainties related to the shape of the concentration-response relationships

6.5.14.2 03

The 2020 final 03 ISA evaluated a number of studies examining the shape of the concentration-response
relationship for long term 03 exposure and mortality using various different statistical techniques,
including linear models and restricted cubic splines, which we use to inform the long-term 03-
attributable respiratory mortality relationship (U.S. EPA, 2020a). The ISA concluded that:

Generally linear, no-threshold relationships exist down to 35-40 ppb, although the results were
not entirely consistent. Some studies observed a sublinear relationship, indicating larger changes
in risk for higher 03 concentrations compared with lower 03 concentrations. Several studies also
included threshold analyses and support the possibility of a threshold near 35 to 40 ppb. (U.S.
EPA, 2020a, section 6.2.7)

The ozone ISA also found that:

Recent multicity studies continue to support a linear [concentration-response] relationship with
no evidence of a threshold between short term ozone exposure and mortality over the range of
ozone concentrations typically observed in the U.S. Studies that used different statistical
approaches and ozone averaging times (i.e., 24 hour avg and 8 hour max) provide evidence of a
linear concentration-response relationship, with less certainty in the shape of the curve at lower
concentrations [i.e., 40 ppb for 24 hour avg and 30 ppb for 8 hour max]. An examination of
whether a threshold exists in the ozone mortality concentration-response relationship provided
no evidence of a concentration below which mortality effects do not occur when examining 5
/ug/m3 (~2.55 ppb) increments across the range ofl hour max concentrations reported in the U.S.
and Canadian cities included in [a large cohort]. (U.S. EPA, 2020a, section 6.1.8)

Collectively, these results continue to support the conclusion of the 2006 Ozone Air Quality Criteria
Document that "if a population threshold level exists in ozone health effects, it is likely near the lower
limit of ambient ozone concentrations in the U.S." and this we assume linear, no-threshold relationships
exist between ozone and health impacts in the main benefits estimate.

In addition, the studies identified as best characterizing respiratory mortality exposures did not provide
threshold models or find evidence supporting a threshold associated with warm-season effects. Turner
et al., 2016 did find "some evidence that a threshold model improved model fit for respiratory mortality
at 35 ppb (P = 0.002) compared with a linear model using year-round but not summertime 03 (HR per 10
ppb using threshold 03 indicator at 35 ppb for respiratory mortality, 1.17; 95% CI, 1.11-1.22)." However,
as we are currently unable to obtain all year air quality projections, we are unable to quantitatively
assess this year-round-specific uncertainty associated with long-term 03 exposures.

135


-------
6.5.15	Short-Term Lag Structure92

Epidemiologic analyses of short-term exposures often present results as health outcomes occurring a
certain time period, or lag days, after exposure. Although there are means of aggregating outcomes that
do not occur simultaneously, such as distributed or multi-day lags, there is a possibility that the full
impact may not be captured by discrete lag periods in short-term study results. Although uncertainty
remains as to whether short-term health impacts are fully captured by discrete lag durations, potentially
biasing results toward the null, we are currently unable perform quantitative uncertainty analyses
regarding this source of uncertainty.

6.5.15.1	PM2.5

The 2019 final PM ISA states that "a number of recent studies conducted systematic evaluations of the
lag structure of associations for the [short-term] PM2.5 [exposure]-mortality relationship by examining
either multiday lags or a series of single-day lags, and these studies continue to support an immediate
effect (i.e., lag 0-1 days) of short-term PM2.5 exposures on mortality." With respect to morbidity effects,
the ISA found that "while recent studies provided evidence of associations in the range of 0-5 days for
respiratory effects, there was evidence of an immediate effect for cardiovascular effects and mortality
(i.e., 0-1 days) with some initial evidence of associations occurring over longer exposure durations (e.g.,
0-4 days)."

6.5.15.2	03

The 2020 final 03 ISA found that "for respiratory health effects, when examining more overt effects,
such as respiratory related hospital admissions and ED visits (i.e., asthma, COPD, and all respiratory
outcomes), epidemiologic studies reported strongest associations occurring within the 1st few days of
exposure (i.e., in the range of 0 to 3 days)."

6.5.16	Statistical Technique/Model Used to Quantify Risks in Epidemiologic Study
Multiple statistical techniques are used in epidemiological analyses, including the Cox proportional
hazards model and the Poisson survival analysis.

6.5.16.1	PM2.5

The 2019 PM ISA compared the use of various statistical techniques, spatial random effects, and fixed93
effect models (U.S. EPA, 2019c). The ISA found that "results from well-studied, highly regarded cohorts
help to reduce uncertainties that the observed associations between long-term PM2.5 exposure and
mortality could be due to the statistical techniques employed or model specification."

6.5.16.2	03

The 2020 03 ISA found that "studies used a number of different statistical techniques to evaluate the
shape of the [long-term exposure] concentration-response function, including linear models and
restricted cubic splines, and generally observed linear, no-threshold relationships down to 35-40 ppb,
although the results are not entirely consistent" (U.S. EPA, 2020a).

92	The 2019 PM ISAs includes Table A-l in its appendix, which describes the lag hierarchy preferences followed
when identifying risk estimates for benefits assessment.

93	Assumes that there is a single true concentration-response relationship and therefore a single true value for the
risk estimate parameter that applies everywhere.

136


-------
6.5.17	Temperature and Weather

Temperature and weather may impact observed associations between air pollution exposure and health
effects in epidemiologic studies, especially in short-term exposure studies. Although a few studies
attempt to disentangle the influence of temperature and/or weather, there is insufficient information
available to perform quantitative assessments of uncertainty.

6.5.17.1	PM2.5

The PM ISA included a number of studies that assessed whether statistical models adequately account
for temporal trends and weather covariates. The ISA found that:

Across studies that evaluated model specification, [short-term] PIVh.s-mortality, associations
remained positive, although in some cases were attenuated, when using different approaches to
account for temporal trends or weather covariates. Seasonal analyses continue to provide
evidence that associations are larger in magnitude during warmer months, but it remains
unclear whether copollutants confound the associations observed. In addition to seasonal
analyses, some studies also examined whether temperature modifies the [short-term] PM2.5-
mortality relationship. Initial evidence indicates that the PIVh.s-mortality association may be
larger in magnitude at lower and higher temperatures, but this observation has not been
substantiated by studies conducted in the U.S. (U.S. EPA, 2019c, section 11.1.12)

6.5.17.2	03

Temperature and weather can also impact epidemiologic results, especially in short-term exposure
analyses. While there is limited evidence of differential 03 mortality associations by season, the 2020 03
ISA determined that the most extensive analyses conducted by recent studies examined whether
temperature (i.e., long-term average temperatures or the distribution of mean daily temperatures)
modifies the 03 mortality association. Analyses focusing on temperature indicate that locations with
lower long-term average temperature have higher 03 mortality risk estimates, which is also reflected by
the observed difference in risk estimates between northern and southern U.S. cities in a single study.
However, as long term average temperature may be a surrogate for air conditioning prevalence and
studies that examined either the joint or stratified effects of 03 and temperature on mortality provided
evidence of 03 mortality associations that are larger in magnitude at temperature extremes, we do not
plan on including quantitative uncertainty analyses for the effect of temperature on ozone effects.

6.5.18	Unquantified Impacts

As with all estimates of benefits, due to the lack of complete data, not all human health impacts
attributable PM2.5 and 03 can be identified and quantified. EPA acknowledges the existence of
unquantified impacts, such as subclinical health endpoints (e.g., hypertension, inflammation, changes in
lung/heart function, etc.) or pollutant-attributable clinical endpoints not evaluated in epidemiologic
studies.

137


-------
7 References

Adams, PF, Hendershot, GE and Marano, MA (1999). Current estimates from the national health
interview survey, 1996. Vital and Health Statistics(Series 10): 1-203.

AHRQ (2016). HCUPnet, Healthcare Cost and Utilization Project. Agency for Healthcare Research
and Quality. Agency for Healthcare Research and Quality. Available at:
https://hcupnet.ahrq.gov/.

Alhanti, BA, Chang, HH, Winquist, A, Mulholland, JA, Darrow, LA and Sarnat, SE (2016). Ambient
air pollution and emergency department visits for asthma: a multi-city assessment of
effect modification by age. Journal of Exposure Science & Environmental Epidemiology
26(2): 180-188.

Alzheimer's Association (2020). 2020 Alzheimer's Disease Facts and Figures. Available at:
https://www.alz.org/media/Documents/alzheimers-facts-and-figures.pdf.

Anderson, GB and Bell, ML (2010). Does one size fit all? The suitability of standard ozone

exposure metric conversion ratios and implications for epidemiology. J Expo Sci Environ
Epidemiol 20(1): 2-11.

Baena-Cagnani, C, Rossi, GA and Canonica, GW (2007). Airway remodelling in children: When
does it start? Curr Opin Allergy Clin Immunol 7(2): 196-200.

Bai, L, Shin, S, Burnett, RT, Kwong, JC, Hystad, P, van Donkelaar, A, Goldberg, MS, Lavigne, E,

Weichenthal, S, Martin, RV, Copes, R, Kopp, A and Chen, H (2020). Exposure to ambient
air pollution and the incidence of lung cancer and breast cancer in the Ontario
Population Health and Environment Cohort. Int J Cancer 146(9): 2450-2459.

Barry, V, Klein, M, Winquist, A, Chang, HH, Mulholland, JA, Talbott, EO, Rager, JR, Tolbert, PE and
Sarnat, SE (2019). Characterization of the concentration-response curve for ambient
ozone and acute respiratory morbidity in 5 US cities. J Expo Sci Environ Epidemiol 29(2):
267-277.

Bell, ML, Son, JY, Peng, RD, Wang, Y and Dominici, F (2015). Ambient PM2.5 and Risk of Hospital
Admissions: Do Risks Differ for Men and Women? Epidemiology 26(4): 575-579.

Belova, A, Fann, N, Haskell, J, Hubbell, B and Narayan, T (2020). Estimating Lifetime Cost of
Illness. An Application to Asthma. Ann Am Thorac Soc 17(12): 1558-1569.

Berger, MC, Blomquist, GC, Kenkel, D and Tolley, GS (1987). Valuing Changes in Health Risks: A
Comparison of Alternative Measures. Southern Economic Journal 53(4): 967-984.

Bhattacharyya, N (2011). Incremental healthcare utilization and expenditures for allergic rhinitis
in the United States. The Laryngoscope 121(9): 1830-1833.

Bleichrodt, H, Courbage, C and Rey, B (2019). The value of a statistical life under changes in
ambiguity. Journal of Risk and Uncertainty 58(1): 1-15.

138


-------
Burnett, R, Chen, H, Szyszkowicz, M, Fann, N, Hubbell, B, Pope, CA, 3rd, Apte, JS, Brauer, M,
Cohen, A, Weichenthal, S, Coggins, J, Di, Q, Brunekreef, B, Frostad, J, Lim, SS, Kan, H,
Walker, KD, Thurston, GD, Hayes, RB, Lim, CC, Turner, MC, Jerrett, M, Krewski, D,
Gapstur, SM, Diver, WR, Ostro, B, Goldberg, D, Crouse, DL, Martin, RV, Peters, P, Pinault,
L, Tjepkema, M, van Donkelaar, A, Villeneuve, PJ, Miller, AB, Yin, P, Zhou, M, Wang, L,
Janssen, NAH, Marra, M, Atkinson, RW, Tsang, H, QuocThach, T, Cannon, JB, Allen, RT,
Hart, JE, Laden, F, Cesaroni, G, Forastiere, F, Weinmayr, G, Jaensch, A, Nagel, G, Concin,
H and Spadaro, JV (2018). Global estimates of mortality associated with long-term
exposure to outdoor fine particulate matter. Proc Natl Acad Sci USA 115(38): 9592-9597.

Cakmak, S, Dales, RE and Judek, S (2006). Respiratory health effects of air pollution gases:
modification by education and income. Archives of Environmental & Occupational
Health 61(1): 5-10.

Cameron, TA. (2001). Letter from Trudy Cameron, Chair, Advisory Council on Clean Air

Compliance Analysis, to Administrator Christine Todd Whitman. Re:Review of the Draft
Analytical Plan for EPA's Second Prospective Analysis—Benefits and Costs of the Clean
Air Act 1990-2020: An Advisory by a Special Panel of the Advisory Council on Clean Air
Compliance Analysis. September 24, 2001. EPA-SAB-COUNCIL-ADV-01-004. Office of the
Administrator, Science Advisory Board U.S. EPA HQ, Washington DC. Available at:
https://yosemite.epa.gov/sab/sabproduct.nsf/572DC1989A3986CC8525718D006BAB8B
/$File/councila01004.pdf.

Cameron, TA. (2004). Letter from Trudy Cameron, Chair, Advisory Council on Clean Air
Compliance Analysis, to Administrator Michael Leavitt. Re: Review of the Draft
Analytical Plan for EPA's Second Prospective Analysis - Benefits and Costs of the Clean
Air Act, 1990-2020: An Advisory by the Advisory Council for Clean Air Compliance
Analysis. May 20, 2004. EPA-SAB-COUNCIL-ADV-04-004. Office of the Administrator,
Science Advisory Board U.S. EPA HQ, Washington DC. Available at:
https://yosemite.epa.gov/sab/sabproduct.nsf/7CCBBFE15CD4C8B185256F17005E3079/
$File/council_adv_04004.pdf.

Canada, EaCC (2016). Canada-United States Transboundary Particulate Matter Science

Assessment 2013. Environment and Climate Change Canada. 978-0-660-04660-0.
https://www.epa.gov/sites/production/files/2016-

09/documents/pm_transboundary_assessment_2013_downloaded_27septl6.pdf.

CBC (2016). Canada by the Numbers. CBC News Online.

Chen, L, Jennison, BL, Yang, W and Omaye, ST (2000). Elementary school absenteeism and air
pollution. Inhal Toxicol 12(11): 997-1016.

Coleman, NC, Burnett, RT, Ezzati, M, Marshall, JD, Robinson, ALand Pope, CA, 3rd (2020). Fine
Particulate Matter Exposure and Cancer Incidence: Analysis of SEER Cancer Registry
Data from 1992-2016. Environ Health Perspect 128(10): 107004.

Cropper, ML and Krupnick, AJ (1990). Social costs of chronic heart and lung disease. Quality of
the Environment Division, Resources for the Future.

139


-------
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, O 3, and NO 2 exposures and associations with mortality
over 16 years of follow-up in the Canadian Census Health and Environment Cohort
(CanCHEC). Environ Health Perspect 123(11): 1180-1186.

De Pablo-Fernandez, E, Tur, C, Revesz, T, Lees, AJ, Holton, JL and Warner, TT (2017). Association
of Autonomic Dysfunction With Disease Progression and Survival in Parkinson Disease.
JAMA Neurol 74(8): 970-976.

Delfino, RJ, Gillen, DL, Tjoa, T, Staimer, N, Polidori, A, Arhami, M, Sioutas, C and Longhurst, J

(2011). Electrocardiographic ST-Segment Depression and Exposure to Traffic - Related
Aerosols in Elderly Subjects with Coronary Artery Disease. Environ Health Perspect
119(2): 196-202.

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. J Am Med Assoc
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 Engl J Med
376(26): 2513-2522.

Dickie, M and Messman, VL (2004). Parental altruism and the value of avoiding acute illness: are
kids worth more than parents? J Environ Econ Manage 48(3): 1146-1174.

Ensor, KB, Raun, LH and Persse, D (2013). A case-crossover analysis of out-of-hospital cardiac
arrest and air pollution. Circulation 127(11): 1192-1199.

EO 12886 (1993). 58 FR 51735. 190.

Freeman III, AM, Herriges, JA and Kling, CL (2014). The measurement of environmental and
resource values: theory and methods. Routledge.

Garcia, E, Berhane, KT, Islam, T, McConnell, R, Urman, R, Chen, Z and Gilliland, FD (2019).

Association of Changes in Air Quality With Incident Asthma in Children in California,
1993-2014. JAMA 321(19): 1906-1915.

Gharibvand, L, Shavlik, D, Ghamsary, M, Beeson, WL, Soret, S, Knutsen, R and Knutsen, SF

(2017). The Association between Ambient Fine Particulate Air Pollution and Lung Cancer
Incidence: Results from the AHSMOG-2 Study. Environ Health Perspect 125(3): 378-384.

Gilliland, FD, Berhane, K, Rappaport, EB, Thomas, DC, Avol, E, Gauderman, WJ, London, SJ,

Margolis, HG, McConnell, R and Islam, KT (2001). The effects of ambient air pollution on
school absenteeism due to respiratory illnesses. Epidemiology: 43-54.

Gogna, P, Narain, TA, O'Sullivan, DE, Villeneuve, PJ, Demers, PA, Hystad, P, Brenner, DR,

Friedenreich, CM, King, WD and Com, PST (2019). Estimates of the current and future
burden of lung cancer attributable to PM2.5 in Canada. Prev Med 122: 91-99.

140


-------
Goodwin, N, Smith, J, Davies, A, Perry, C, Rosen, R, Dixon, A, Dixon, J and Ham, C (2011). A
report to the Department of Health and the NHS Future Forum; Integrated Care for
Patients and Populations: Improving Outcomes by Working Together.

Guerra, S, Wright, AL, Morgan, WJ, Sherrill, DL, Holberg, CJ and Martinez, FD (2004). Persistence
of asthma symptoms during adolescence: role of obesity and age at the onset of
puberty. Am J Respir Crit Care Med 170(1): 78-85.

Hammitt, JK. (2008). Letter from James K. Hammitt, Chair, Advisory Concil on Clean Air

Compliance, to Administrator Stephen Johnson. Re: Characterizing Uncertainty in
Particulate Matter Benefits Using Expert Elicitation. July 11, 2008. EPA-COUNCIL-08-002.
Office of the Administrator, Science Advisory Board U.S. EPA HQ, Washington DC.
Available at:

https://yosemite.epa.gOv/sab/sabproduct.nsf/f697818d4467059f8525724100810c37/4
3 B91173651AED9E85257487004E A6CB/$File/EPA-COU NCI L-08-002-unsigned.pdf.

Hammitt, JK and Bailar, J. (2010). Letter from James Hammitt, Chair, Advisory Council on Clean
Air Compliance Analysis and John Bailar, Chair, Health Effects Subcommittee, to
Administrator Lisa Jackson. Re: Review of EPA's Draft Health Benefits of the Second
Section 812 Prospective Study of the Clean Air Act (June 2010). EPA-COUNCIL-10-001.
Office of the Adminitrator, Science Advisory Board U.S. EPA HQ, Washington DC.
Available at:

https://yosemite.epa.gOv/sab/sabproduct.nsf/9288428b8eeea4c885257242006935a3/7
2 D4EFA39E48CDB28525774500738776/$File/E PA-CO UNCI L-10-001-unsigned.pdf.

Harrington, W and Portney, PR (1987). Valuing the benefits of health and safety regulation.
Journal of Urban Economics 22(1): 101-112.

Harrison, RM, Smith, D and Kibble, A (2004). What is responsible for the carcinogenicity of PM2.
5? Occup Environ Med 61(10): 799-805.

Hollmann, F, Mulder, T and Kalian, JJW, DC: US Bureau of the Census (2000). Methodology and
assumptions for the population projections of the United States: 1999 to 2100
(Population Division Working Paper No. 38). 338.

Honeycutt, M. (2020). Letter from Michael Honeycutt Chair, Scientific Advisory, to Administrator
Lisa Jackson. Re: Science Advisory Board (SAB) Consideration of the Scientific and
Technical Basis of EPA's Proposed Rule titled "Increasing Consistency and Transparency
in Considering Benefits and Costs in the Clean Air Act Rulemaking Process.". Septemeber
30, 2020. EPA-SAB-20-012. Office of the Administrator, Science Advisory Board U.S. EPA
HQ, Washington DC. Available at:

https://yosemite.epa.gov/sab/sabproduct.nsf/LookupWebReportsLastMonthBOARD/0A
312659C8AC185D852585F80049803C/$File/EPA-SAB-20-012.pdf.

lEc. (1993). Memorandum to Jim DeMocker, U.S. Environmental Protection Agency, Office of Air
and Radiation, Office of Policy Analysis and Review. September 30. . U.S. EPA.

141


-------
lEc (2006). Expanded expert judgment assessment of the concentration - response relationship
between PM2. 5 exposure and mortality. Available at:

http://www.epa.gov/ttn/ecas/regdata/Uncertainty/pm_ee_tsd_expert_interview_sum
maries.pdf.

Jerrett, M, Burnett, RT, Pope, CA, 3rd, Ito, K, Thurston, G, Krewski, D, Shi, Y, Calle, E and Thun, M
(2009). Long-term ozone exposure and mortality. N Engl J Med 360(11): 1085-1095.

Jones, MR, Diez-Roux, AV, O'Neill, MS, Guallar, E, Sharrett, AR, Post, W, Kaufman, JD and Navas-
Acien, A (2015). Ambient air pollution and racial/ethnic differences in carotid intima-
media thickness in the Multi-Ethnic Study of Atherosclerosis (MESA). J Epidemiol
Community Health 69(12): 1191-1198.

Jutkowitz, E, Kane, RL, Gaugler, JE, MacLehose, RF, Dowd, B and Kuntz, KM (2017). Societal and
Family Lifetime Cost of Dementia: Implications for Policy. J Am Geriatr Soc 65(10): 2169-
2175.

Katsouyanni, K, Samet, JM, Anderson, HR, Atkinson, R, Le Tertre, A, Medina, S, Samoli, E,

Touloumi, G, Burnett, RT, Krewski, D, Ramsay, T, Dominici, F, Peng, RD, Schwartz, J,
Zanobetti, A and Committee, HEIHR (2009). Air pollution and health: a European and
North American approach (APHENA). Res Rep Health Eff lnst(142): 5-90.

Kaye, DR, Min, HS, Herrel, LA, Dupree, JM, Ellimoottil, C and Miller, DC (2018). Costs of Cancer
Care Across the Disease Continuum. Oncologist 23(7): 798-805.

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.

Kivi, PA and Shogren, JF (2010). Second-order ambiguity in very low probability risks: food safety
valuation. J Agric Resour Econ: 443-456.

Kleckner, N and Neumann, J (1999). Recommended Approach to Adjusting WTP Estimates to

Reflect Changes in Real Income. Memorandum to Jim Democker, US EPA/OPAR, June 3.

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, Anderson, GB, Dominici, 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. Environ Health Perspect 121(10): 1148-1153.

Krewski, D, Burnett, R, Goldberg, M, Hoover, K, Siemiatycki, J, Jerrett, M, Abrahamowich, M and
White, W (2000). Reanalysis of the Harvard six-cities study and the American Cancer
Society study of air pollution and mortality, phase II: sensitivity analysis. Health Effects
Institute 295.

Krewski, D, Jerrett, M, Burnett, RT, Ma, R, Hughes, E, Shi, Y, Turner, MC, Pope III, CA, Thurston, G
and Calle, EE (2009). Extended follow-up and spatial analysis of the American Cancer

142


-------
Society study linking particulate air pollution and mortality. Health Effects Institute
Boston, MA.

Krupnick, AJ and Cropper, ML (1992). The effect of information on health risk valuations. Journal
of Risk and Uncertainty 5(1): 29-48.

Kulhanova, I, Morelli, X, LeTertre, A, Loomis, D, Charbotel, B, Medina, S, Ormsby, JN, Lepeule, J,
Slama, R and Soerjomataram, I (2018). The fraction of lung cancer incidence attributable
to fine particulate air pollution in France: Impact of spatial resolution of air pollution
models. Environ Int 121(Pt 2): 1079-1086.

Lee, WC, Christensen, MC, Joshi, AV and Pashos, CL (2007). Long-term cost of stroke subtypes
among Medicare beneficiaries. Cerebrovascular Diseases 23(1): 57-65.

Lepeule, J, Laden, F, Dockery, D and Schwartz, JJEhp (2012). Chronic exposure to fine particles
and mortality: an extended follow-up of the Harvard Six Cities study from 1974 to 2009.
Environ Health Perspect 120(7): 965-970.

Lewis, TC, Robins, TG, Mentz, GB, Zhang, X, Mukherjee, B, Lin, X, Keeler, GJ, Dvonch, JT, Yip, FY,
O'Neill, MS, Parker, EA, Israel, BA, Max, PT, Reyes, A and Community Action Against
Asthma Steering, C (2013). Air pollution and respiratory symptoms among children with
asthma: vulnerability by corticosteroid use and residence area. The Science of the total
environment 448: 48-55.

Lin, M, Stieb, DM and Chen, Y (2005). Coarse particulate matter and hospitalization for

respiratory infections in children younger than 15 years in Toronto: A case-crossover
analysis. Pediatrics 116(2): e235-e240.

Luengo-Fernandez, R, Gray, AM and Rothwell, PM (2012). A population-based study of hospital
care costs during 5 years after transient ischemic attack and stroke. Stroke 43(12): 3343-
3351.

Mar, TF and Koenig, JQ (2009). Relationship between visits to emergency departments for
asthma and ozone exposure in greater Seattle, Washington. Ann Allergy, Asthma
Immunol 103(6): 474-479.

McConnell, R, Islam, T, Shankardass, K, Jerrett, M, Lurmann, F, Gilliland, F, Gauderman, J, Avol,
E, Kunzli, N, Yao, L, Peters, J and Berhane, K (2010). Childhood incident asthma and
traffic-related air pollution at home and school. Environ Health Perspect 118(7): 1021-
1026.

McGartland, A, Revesz, R, Axelrad, DA, Dockins, C, Sutton, P and Woodruff, TJ (2017). Estimating
the health benefits of environmental regulations. Science 357(6350): 457-458.

Medina-Ramon, M and Schwartz, J (2008). Who is more vulnerable to die from ozone air
pollution? Epidemiology: 672-679.

Mu, F, Hurley, D, Betts, K, Messali, A, Paschoalin, M, Kelley, C and Wu, E (2017). Real-world costs
of ischemic stroke by discharge status. Current medical research and opinion 33(2): 371-
378.

143


-------
NCES (1996). The Condition of Education 1996, Indicator 42: Student Absenteeism and

Tardiness. Office of Educational Research and Improvement. Washington, DC. U.S.
Department of Education. NCES 96-304. Available at:
https://nces.ed.gov/pubs96/96304.pdf.

NCI (2015). SEER, Surveillance, Epidemiology, and End Results Program. National Institutes of
Health, National Cancer Institute (NCI). Bethesda, MD.

Nicholson, G, Gandra, SR, Halbert, RJ, Richhariya, A and Nordyke, RJ (2016). Patient-level costs
of major cardiovascular conditions: a review of the international literature.
ClinicoEconomics and outcomes research: CEOR 8: 495.

Nishimura, KK, Galanter, JM, Roth, LA, Oh, SS, Thakur, N, Nguyen, EA, Thyne, S, Farber, HJ,
Serebrisky, D, Kumar, R, Brigino-Buenaventura, E, Davis, A, LeNoir, MA, Meade, K,
Rodriguez-Cintron, W, Avila, PC, Borrell, LN, Bibbins-Domingo, K, Rodriguez-Santana, JR,
Sen, S, Lurmann, F, Balmes, JR and Burchard, EG (2013). Early-life air pollution and
asthma risk in minority children. The GALA II and SAGE II studies. Am J Respir Crit Care
Med 188(3): 309-318.

NRC (2002). Estimating the public health benefits of proposed air pollution regulations.
0309086094. National Academies Press.

O'Sullivan, AK, Rubin, J, Nyambose, J, Kuznik, A, Cohen, DJ and Thompson, D (2011). Cost

estimation of cardiovascular disease events in the US. Pharmacoeconomics 29(8): 693-
704.

Ochs, M, Nyengaard, JR, Jung, A, Knudsen, L, Voigt, M, Wahlers, T, Richter, J and Gundersen, HJ
(2004). The number of alveoli in the human lung. Am J Respir Crit Care Med 169(1): 120-
124.

OECD (2016). Users' Handbook supplement to the Guidance Document for developing and
assessing Adverse Outcome Pathways. Organisation for Economic Co-operation and
Development. ENV/JM/MONO(2016)12. https://www.oecd-
ilibrary.org/docserver/5jlvlm9dlg32-

en.pdf?expires=1606153032&id=id&accname=guest&checksum=B85E857D295450C88F
C13BD1FFF0DE69.

OMB (2003). Circular A-4. Office of Management and Budget. Washinton, DC. 68 FR 58366.
October 2003. Available at:

https://www.federalregister.gov/documents/2003/10/09/03-25606/circular-a-4-
regulatory-analysis.

Onukwugha, E, McRae, J, Kravetz, A, Varga, S, Khairnar, R and Mullins, CD (2016). Cost-of-lllness
Studies: An Updated Review of Current Methods. Pharmacoeconomics 34(1): 43-58.

Ostro, B (1987). Air pollution and morbidity revisited: A specification test. J Environ Econ
Manage 14(1): 87-98.

144


-------
Ostro, B, Lipsett, M, Mann, J, Braxton-Owens, H and White, M (2001). Air pollution and
exacerbation of asthma in African-American children in Los Angeles. Journal of
Epidemiology 12(2): 200-208.

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.
Am J Epidemiol 184(6): 450-459.

Ostro, B, Roth, L, Malig, B and Marty, M (2009). The effects of fine particle components on

respiratory hospital admissions in children. Environ Health Perspect 117(3): 475-480.

Ostro, BC, T. A. (2004). Letter from Dr. Bart Ostro, Chair, Health Effects Subcommittee and Dr.
Trudy Ann Cameron, Chair, Advisory Council on Clean Air Compliance Analysis to
Honorable Michael O. Leavitt, Administrator, US EPA. Re: Advisory on Plans for Health
Effects Analysis in the Analytical Plan for EPA's Second Prospective Analysis - Benefits
and Costs of the Clean Air Act, 1990-2020. March 2004. EPA-SAB-COUNCIL-ADV-04-002.
Office of the Administrator, Science Advisory Board U.S. EPA HQ, Washington DC.
Available at:

https://yosemite.epa.gov/sab%5CSABPRODUCT.NSF/08E1155AD24F871C85256E540043
3D5D/$File/council_adv_04002.pdf.

Ostro, BD and Rothschild, S (1989). Air pollution and acute respiratory morbidity: an
observational study of multiple pollutants. Environ Res 50(2): 238-247.

Parker, JD, Akinbami, LJ and Woodruff, TJ (2009). Air pollution and childhood respiratory
allergies in the United States. Environ Health Perspect 117(1): 140-147.

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 to 2009
National Health Interview Survey with mortality follow-up through 2011. Circulation
137(16): 1688-1697.

Paulu, C and Smith, AE (2008). Tracking associations between ambient ozone and asthma-
related emergency department visits using case-crossover analysis. Journal of Public
Health Management and Practice 14(6): 581-591.

Peters, A, Dockery, DW, Muller, JE and Mittleman, MA (2001). Increased particulate air pollution
and the triggering of myocardial infarction. Circulation 103(23): 2810-2815.

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. Circul Res 116(1): 108-U258.

Pope III, C, Muhlestein, JB, May, HT, Renlund, DG, Anderson, JL and Home, BD (2006). Ischemic
heart disease events triggered by short-term exposure to fine particulate air pollution.
Circulation 114(23): 2443-2448.

Pringsheim, T, Jette, N, Frolkis, A and Steeves, TD (2014). The prevalence of Parkinson's disease:
a systematic review and meta-analysis. Mov Disord 29(13): 1583-1590.

145


-------
Rabinovitch, N, Strand, M and Gelfand, EW (2006). Particulate levels are associated with early
asthma worsening in children with persistent disease. Am J Respir Crit Care Med
173(10): 1098-1105.

Ransom, MR and Pope, CA, 3rd (1992). Elementary school absences and PM10 pollution in Utah
Valley. Environ Res 58(2): 204-219.

Roger, VL, Go, AS, Lloyd-Jones, DM, Benjamin, EJ, Berry, JD, Borden, WB, Bravata, DM, Dai, S,
Ford, ES, Fox, CS, Fullerton, HJ, Gillespie, C, Hailpern, SM, Heit, JA, Howard, VJ, Kissela,
BM, Kittner, SJ, Lackland, DT, Lichtman, JH, Lisabeth, LD, Makuc, DM, Marcus, GM,
Marelli, A, Matchar, DB, Moy, CS, Mozaffarian, D, Mussolino, ME, Nichol, G, Paynter, NP,
Soliman, EZ, Sorlie, PD, Sotoodehnia, N, Turan, TN, Virani, SS, Wong, ND, Woo, D,

Turner, MB, American Heart Association Statistics, C and Stroke Statistics, S (2012).

Heart disease and stroke statistics-2012 update: a report from the American Heart
Association. Circulation 125(1): e2-e220.

Rosamond, W, Broda, G, Kawalec, E, Rywik, S, Pajak, A, Cooper, L and Chambless, L (1999).

Comparison of medical care and survival of hospitalized patients with acute myocardial
infarction in Poland and the United States. The American Journal of Cardiology 83(8):
1180-1185.

Rosenthal, FS, Carney, JP and Olinger, ML (2008). Out-of-hospital cardiac arrest and airborne

fine particulate matter: a case-crossover analysis of emergency medical services data in
Indianapolis, Indiana. Environ Health Perspect 116(5): 631-636.

Siegel, RL, Miller, KD and Jemal, A (2019). Cancer statistics, 2019. CA Cancer J Clin 69(1): 7-34.

Silverman, RA, Ito, K, Freese, J, Kaufman, BJ, De Claro, D, Braun, J and Prezant, DJ (2010).

Association of ambient fine particles with out-of-hospital cardiac arrests in New York
City. Am J Epidemiol 172(8): 917-923.

Soni, A (2008). Allergic Rhinitis: Trends in use and expenditures, 2000 and 2005. Medical
Expenditure Panel Survey, Agency for Healthcare Research and Quality.

Sparrow, D, O'Connor, GT, Rosner, B, Segal, MR and Weiss, ST (1991). The influence of age and
level of pulmonary function on nonspecific airway responsiveness. The Normative Aging
Study. Am Rev Respir Dis 143(5 Pt 1): 978-982.

Stavins, R. (2000). Letter from Robert Stevens, Chair, Science Advisory Board, and the Chair,

Environmental Economics Advisory Committee, to EPA Administrator Carol M. Browner,
Re: EPA's White Paper Valuing the Benefits of Fatal Cancer Risk Reduction. EPA-SAB-
EEAC-00-013. Office of the Administrator, Science Advisory Board U.S. EPA HQ,
Washington DC.

Sullivan, J, Sheppard, L, Schreuder, A, Ishikawa, N, Siscovick, D and Kaufman, J (2005). Relation
between short-term fine-particulate matter exposure and onset of myocardial
infarction. Epidemiology: 41-48.

146


-------
Talbott, EO, Rager, JR, Benson, S, Brink, LA, Bilonick, RA and Wu, C (2014). A case-crossover

analysis of the impact of PM(2.5) on cardiovascular disease hospitalizations for selected
CDC tracking states. Environ Res 134: 455-465.

Tetreault, LF, Doucet, M, Gamache, P, Fournier, M, Brand, A, Kosatsky, T and Smargiassi, A
(2016). Childhood Exposure to Ambient Air Pollutants and the Onset of Asthma: An
Administrative Cohort Study in Quebec. Environ Health Perspect 124(8): 1276-1282.

Tolley, GS, Babcock, L, Berger, M, Bilotti, A, Blomquist, G, Brien, M, Fabian, R, Fishelson, G, Kahn,
C and Kelly, A (1986). Valuation of reductions in human health symptoms and risks. U.S
EPA.

Trivedi, M and Denton, E (2019). Asthma in Children and Adults-What Are the Differences and
What Can They Tell us About Asthma? Front Pediatr 7: 256.

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. Am J Respir Crit Care Med 193(10): 1134-1142.

U.S. Census Bureau (1998). Population Profile of the United States: 1997. Economics and
Statistics Administration U.S. Department of Commerce, Bureau of the Census.
Washington, DC. Available at: https://www.census.gOv/prod/3/98pubs/p23-194.pdf.

U.S. Census Bureau (2012). Technical documentation - 2010 Census Summary File 1—Technical
Documentation/prepared by the U.S. Census Bureau, Revised 2012. Available at:
http://www.census.gov/prod/cen2010/doc/sfl.pdf. Employment Status from the 5-year
American Community Survey (ACS) data, 2010 U.S. Census American FactFinder.
Available at: http://factfinder2.census.gov/. Commuting times file from U.S. Census data
portal (http://dataferrett.census.gov/), Table P31, variables P031001-P031015.

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/recordisplay.cfm?deid=216546.

U.S. EPA (2011a). The Benefits and Costs of the Clean Air Act 1990 to 2020: Final Report. Office
of Air and Radiation. Washington, DC. U.S. EPA. Available at:

https://www.epa.gov/sites/production/files/2015-07/documents/fullreport_rev_a.pdf.

U.S. EPA (2011b). Regulatory Impact Analysis: National Emission Standards for Hazardous Air

Pollutants for Industral, Commercial, and Institutional Boilers and Process Heaters. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards, Health
and Environmental Impact Division. Research Triangle Park, NC. U.S EPA. Available at:
https://www3.epa.gov/ttn/ecas/regdata/RIAs/boilersriafinalll0221_psg.pdf.

U.S. EPA (2011c). Regulatory Impact Analysis for the Final Mercury and Air Toxics Standards. U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards, Health
and Environmental Impact Division. Research Triangle Park, NC. U.S. EPA. EPA-452/R-11-

147


-------
Oil. December 2011. Available at:

http://www.epa.gov/ttn/ecas/regdata/RIAs/matsriafinal.pdf.

U.S. EPA (2011d). Regulatory Impact Analysis for the Federal Implementation Plans to Reduce

Interstate Transport of Fine Particulate Matter and Ozone in 27 States; Correction of SIP
Approvals for 22 States. U.S. Environmental Protection Agency, Office of Air Quality
Planning and Standards, Health and Environmental Impact Division. Research Triangle
Park, NC. U.S. EPA. June 2011. Available at:

https://www3.epa.gov/ttn/ecas/docs/ria/transport_ria_final-csapr_2011-06.pdf.

U.S. EPA (2012a). Regulatory Impact Analysis for the Proposed Revisions to the National

Ambient Air Quality Standards for Particulate Matter. U.S. Environmental Protection
Agency, Office of Air Quality Planning and Standards, Health and Environmental Impact
Division. Research Triangle Park, NC. U.S. EPA. EPA-452/R-12-003. June 2012. Available
at:

https://www3.epa.gov/ttnecasl/regdata/RIAs/PMRIACombinedFile_Bookmarked.pdf.

U.S. EPA (2012b). 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, Health and Environmental Impact Division.
Research Triangle Park, NC. U.S. EPA. EPA-452/R-12-005. December 2012. Available at:
https://www3.epa.gov/ttnecasl/regdata/RIAs/finalria.pdf.

U.S. EPA (2013). Integrated Science Assessment of Ozone and Related Photochemical Oxidants
(Final Report). Office of Research and Development, National Center for Environmental
Assessment. Research Triangle Park, NC. U.S. EPA. EPA-600/R-10-076F. February 2013.
Available at: https://nepis.epa.gov/Exe/ZyPURL.cgi?Dockey=P100KETF.txt.

U.S. EPA (2014). Guidelines for Preparing Economic Analyses National Center for Environmental
Economics. US Environmental Protection Agency Washington, DC.

U.S. EPA (2015a). Regulatory Impact Analysis of the Final Revisions to the National Ambient Air
Quality Standards for Ground-Level Ozone. Office of Air Quality Planning and Standards,
Health and Environmental Impacts Division. Research Triangle Park, NC. U.S. EPA. EPA-
452/R-15-007. September 2015. Available at:
https://www3.epa.gov/ttnecasl/docs/20151001ria.pdf.

U.S. EPA (2015b). 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/recordisplay.cfm?deid=310244.

U.S. EPA (2017). Trichloroethylene (TCE); Regulation of Use in Vapor Degreasing Under TSCA
Section 6(a). 40 CFR 751. Washington, DC. Available at:
https://www.federalregister.gov/documents/2017/01/19/2017-

01229/trichloroethylene-tce-regulation-of-use-in-vapor-degreasing-under-tsca-section-
6a.

148


-------
U.S. EPA (2018). Environmental Benefits Mapping and Analysis Program - Community Edition

User's Manual. Office of Air Quality Planning and Standards. Research Triangle Park, NC.
U.S. EPA. Available at: https://www.epa.gov/sites/production/files/2015-
04/documents/benmap-ce_user_manual_march_2015.pdf.

U.S. EPA (2019a). Regulatory Impact Analysis for the Repeal of the Clean Power Plan, and the
Emission Guidelines for Greenhouse Gas Emissions from Existing Electric Utility
Generating Units. U.S. Environmental Protection Agency, Office of Air Quality Planning
and Standards, Health and Environmental Impact Division. Research Triangle Park, NC.
U.S. EPA. EPA-452/R-19-003. June 2019. Available at:
https://www.epa.gov/sites/production/files/2019-
06/documents/utilities_ria_final_cpp_repeal_and_ace_2019-06.pdf.

U.S. EPA (2019b). Policy Assessment for the Review of the Ozone National Ambient Air Quality
Standards, External Review Draft. U.S. Environmental Protection Agency, Office of Air
Quality Planning and Standards, Heath and Environmental Impacts Division. Research
Triangle Park, NC. U.S. EPA. EPA-452/P-19-002. October 2019 Available at:
https://www.epa.gov/naaqs/ozone-o3-standards-policy-assessments-current-review.

U.S. EPA (2019c). 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 (2020a). Integrated Science Assessment for Ozone and Related Photochemical

Oxidants. U.S. Environmental Protection Agency. Washington, DC. Office of Research
and Development. EPA/600/R-20/012. Available at:

https://www.epa.gov/isa/integrated-science-assessment-isa-ozone-and-related-
photochemical-oxidants.

U.S. EPA (2020b). Regulatory Impact Analysis for the for the Proposed Revised Cross-State Air
Pollution Rule (CSAPR) Update for the 2008 Ozone NAAQS. U.S. Environmental
Protection Agency, Office of Air Quality Planning and Standards, Health and
Environmental Impact Division. Research Triangle Park, NC. U.S. EPA. EPA-452/P-20-003.
October 2020. Available at: https://www.epa.gov/sites/production/files/2020-
10/documents/revised_csapr_update_ria_proposal.pdf.

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

Standards for Particulate Matter. U.S. Environmental Protection Agency, Office of Air
Quality Planning and Standards, Heath and Environmental Impacts Division. Research
Triangle Park, NC. U.S. EPA. EPA-452/R-20-002. January 2020. Available at:
https://www.epa.gov/naaqs/particulate-matter-pm-standards-policy-assessments-
current-review-O.

U.S. EPA (2020d). Policy Assessment for the Review of the Ozone National Ambient Air Quality
Standards. U.S. Environmental Protection Agency, Office of Air Quality Planning and

149


-------
Standards, Heath and Environmental Impacts Division. Research Triangle Park, NC. U.S.
EPA. EPA-452/R-20-001. 2020 Available at: https://www.epa.gov/naaqs/ozone-o3-
standards-policy-assessments-current-review.

US Bureau of Labor Statistics (2015). Current Population Survey. 2015 Annual Social and
Economic Supplement. Available at: https://www.bls.gov/opub/reports/womens-
databook/2016/home.htm.

Villeneuve, PJ, Chen, L, Rowe, BH and Coates, F (2007). Outdoor air pollution and emergency
department visits for asthma among children and adults: a case-crossover study in
northern Alberta, Canada. Environ Health 6(1): 40.

Viscusi, WK (1992). Fatal tradeoffs : public and private responsibilities for risk. Oxford University
Press. New York. Publisher description

http://www.loc.gov/catdir/enhancements/fy0635/91036121-d.html.

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, Crouse, DL, Pinault, L, Godri-Pollitt, K, Lavigne, E, Evans, G, van Donkelaar, A,

Martin, RV and Burnett, RT (2016). Oxidative burden of fine particulate air pollution and
risk of cause-specific mortality in the Canadian Census Health and Environment Cohort
(CanCHEC). Environ Res 146: 92-99.

Winer, RA, Qin, X, Harrington, T, Moorman, J and Zahran, H (2012). Asthma incidence among

children and adults: findings from the Behavioral Risk Factor Surveillance system asthma
call-back survey—United States, 2006-2008. J Asthma 49(1): 16-22.

Woodruff, TJ, Darrow, LA and Parker, JD (2008). Air pollution and postneonatal infant mortality
in the United States, 1999-2002. Environ Health Perspect 116(1): 110-115.

Woodruff, TJ, Grillo, J and Schoendorf, KC (1997). The relationship between selected causes of
postneonatal infant mortality and particulate air pollution in the United States. Environ
Health Perspect 105(6): 608-612.

Woods & Poole (2015). Complete Demographic Database.

Yang, W, Hamilton, JL, Kopil, C, Beck, JC, Tanner, CM, Albin, RL, Ray Dorsey, E, Dahodwala, N,
Cintina, I, Hogan, P and Thompson, T (2020). Current and projected future economic
burden of Parkinson's disease in the U.S. NPJ Parkinsons Dis 6: 15.

Zanobetti, A, Franklin, M, Koutrakis, P and Schwartz, J (2009). Fine particulate air pollution and
its components in association with cause-specific emergency admissions. Environ Health
8(1): 58.

Zanobetti, A and Schwartz, J (2006). Air pollution and emergency admissions in Boston, MA.
Journal of Epidemiology Community Health 60(10): 890-895.

150


-------
Zanobetti, A and Schwartz, J (2008). Mortality displacement in the association of ozone with

mortality: an analysis of 48 cities in the United States. Am J Respir Crit Care Med 177(2):
184-189.

Zhang, Z, Whitsel, EA, Quibrera, PM, Smith, RL, Liao, D, Anderson, GLand Prineas, RJ (2009).

Ambient fine particulate matter exposure and myocardial ischemia in the Environmental
Epidemiology of Arrhythmogenesis in the Women's Health Initiative (EEAWHI) study.
Environ Health Perspect 117(5): 751-756.

151


-------