Estimating PM2.5- and Ozone-Attributable
Health Benefits: 2024 Update
U.S Environmental Protection Agency
Office of Air and Radiation
Research Triangle Park, North Carolina
June 2024
i
-------
Contents
Preface 1
1 Introduction 2
1.1 Benefits Analysis Background 2
1.2 The Relationship Between Identifying Health Endpoints for Valuation and WTP 3
1.3 Document Purpose and Overview 4
2 Approach to Identifying Studies and Risk Estimates 6
2.1 Study and Risk Estimate Identification Criteria 6
2.1.1 Minimum Criteria 6
2.1.2 Preferred Criteria Categories 7
2.2 Available Epidemiologic Literature 11
2.2.1 Identification of Exposure-Attributable Health Outcomes 11
2.2.2 Identifying Quantifiable Health Outcomes 23
2.2.3 Study Information Table 24
2.2.4 Methods for Presenting Health Benefits Estimates Using Multiple Risk Estimates for a
Single Endpoint 26
2.3 Systematically Identifying Epidemiologic Studies and Risk Estimates for Benefits Assessment 27
2.3.1 PM2.5 27
2.3.2 03 48
2.4 Identified Study and Risk Estimates for Benefits Assessments 58
2.4.1 Health Endpoints 59
2.4.2 Risk Estimates 60
3 Baseline Incidence and Prevalence Estimates 66
3.1 Mortality 68
3.1.1 Mortality Data for 2012-2014 69
3.1.2 Mortality Rate Projections 2015-2060 71
3.1.3 Race-and Ethnicity-Stratified Incidence Rates 72
3.2 Hospitalizations 73
3.3 Emergency Department Visits 75
3.4 Health Endpoint Onset/Occurrence 75
3.4.1 Acute Myocardial Infarctions (AMIs) 76
3.4.2 Asthma Onset and Symptoms 77
3.4.3 Allergic Rhinitis 78
ii
-------
3.4.4 Lung Cancer 78
3.4.5 Minor Restricted Activity Days (MRAD) 79
3.4.6 School Loss Days 79
3.4.7 Work Loss Days 80
3.4.8 Hypertension 80
4 Demographic Information 83
5 Health Endpoint Valuation 84
5.1 Mortality 89
5.1.1 Value of a Statistical Life (VSL) 89
5.2 Hospitalizations and Emergency Department Visits 89
5.3 Health Endpoint Onset/Occurrence 92
5.3.1 Acute Myocardial Infarctions (AMIs) 92
5.3.2 Allergic Rhinitis (Hay Fever) 93
5.3.3 Asthma Onset 93
5.3.4 Asthma Symptoms/Exacerbation 93
5.3.5 Cardiac Arrest 94
5.3.6 Lung Cancer 95
5.3.7 Minor Restricted Activity Days (MRADs) 97
5.3.8 School Loss Days 97
5.3.9 Stroke 99
5.3.10 Work Loss Days (WLDs) 99
5.4 Developing Income Growth Adjustment Factors for Health Endpoint Onset/Occurrence 99
6 Characterizing Uncertainty and Evaluating Sensitivity to Alternate Assumptions 103
6.1 Quantitative Characterization of PM2.5 Uncertainty and Evaluating Sensitivity to Alternate
PM2.5 Assumptions 103
6.1.1 Statistical Uncertainty Around the Risk Estimate (Monte-Carlo Assessment) 104
6.1.2 Adult All-Cause Mortality 104
6.1.3 Asthma Onset in Children 109
6.1.4 Cardiovascular Hospital Admissions Ill
6.1.5 Hypertension 114
6.1.6 Effect Modification of Health Effects in At-Risk Populations 116
6.2 Quantitative Characterization of 03 Uncertainties and Evaluating Sensitivity to Alternate 03
Assumptions 119
6.2.1 Statistical Uncertainty Around the Risk Estimate (Monte-Carlo Assessment) 119
iii
-------
6.2.2 Respiratory Mortality 119
6.2.3 All-Cause Mortality 120
6.2.4 Asthma Onset in Children 121
6.2.5 Effect Modification of Health Effects in At-Risk Populations 123
6.3 Quantitative Characterization of Baseline Incidence Rate Uncertainties 127
6.4 Quantitative Characterization of Economic Valuation Estimate Uncertainties 128
6.4.1 Mortality Cessation Lag 128
6.4.2 Lung Cancer Cessation Lag 129
6.4.3 Income Elasticity of Willingness to Pay 133
6.4.4 Statistical Estimates of VSL 134
6.4.5 Alzheimer's Disease and Parkinson's Disease Onset Lifetime Costs 134
6.5 Qualitative Characterization of Uncertainties 137
6.5.1 Applying Risk Estimates to Locations and Populations not Specified in the Epidemiologic
Study 137
6.5.2 Causality Determination 138
6.5.3 Estimating and Assigning Exposures in Epidemiology Studies 138
6.5.4 Modeling the Influence of Air Pollution on the Risk of Mortality Over Time 139
6.5.5 Differential Toxicity of PM2.5 According to Chemical Composition 139
6.5.6 Different Long-Term Exposure Windows 140
6.5.7 Discounting Future Benefit Estimates 140
6.5.8 Statistical Estimates of WTP 140
6.5.9 Confounding by Individual Risk Factors 140
6.5.10 Confounding by Other Pollutants 141
6.5.11 Risk Attributable to Long-Term and Short-Term Exposures 141
6.5.12 Heterogeneity of Risk Estimates 141
6.5.13 03 Metrics 141
6.5.14 03 Season 143
6.5.15 Shape of the Concentration-Response Relationship 144
6.5.16 Short-Term Lag Structure 146
6.5.17 Statistical Technique/Model Used to Quantify Risks in Epidemiologic Study 147
6.5.18 Temperature and Weather 147
6.5.19 Unquantified Impacts 148
7 References 149
iv
-------
List of Tables
Table 1. Criteria for Identifying Studies and Risk Estimates to Use in a Benefits Assessment 7
Table 2. Rank-ordered priority 9
Table 3. Causality Determined for PM2.5-Related Health Effects 14
Table 4. Causality Determined for 03-Related Health Effects 14
Table 5. Study Information Tables 24
Table 6. PM2.5 Study and Risk Estimate Identification Diagram* 29
Table 7. 03 Study and Risk Estimate Identification Diagram 48
Table 8. Set of Health Endpoints for Main PM2.5 Benefits Assessments 59
Table 9. Set of Health Endpoints for Main 03 Benefits Assessments 60
Table 10. Set of Risk Estimates for Main PM2.5 Benefits Assessments 60
Table 11. Set of Risk Estimates for Main 03 Benefits Assessments 63
Table 12. Baseline Incidence Rates for Use in Impact Functions 67
Table 13. National Mortality Rates (per 100 people per year) by Health Endpoint and Age Group, 2012-
2014 70
Table 14. All-Cause Mortality Rate (per 100 people per year), by Source, Year, and Age Group 71
Table 15. Ratio of Future Year All-Cause Mortality Rate to 2013 Estimated All-Cause Mortality Rate, by
Age Group 72
Table 16. Asthma Prevalence Rates 77
Table 17. Weighted Average Asthma Prevalence by Study 77
Table 18. Lung Cancer Incidence Rates 79
Table 19. School Loss Day Rates (per student per year) 79
Table 20. Cost of Illness Economic Study Identification Consideration Factors 85
Table 21. Unit Values for Economic Valuation of Health Endpoints (2015$)1 87
Table 22. Central Unit Value for VSL based on 26-value-of-life studies 89
Table 23. Hospitalization and Emergency Department Cost Elements by Endpoint 90
Table 24. Medical Costs and Hospital Stay Data 91
Table 25. Medical Costs for AMIs (2015$) 92
Table 26. Total Valuation Estimates for AMIs (2015$) 93
Table 27. Valuation Estimate for Cardiac Arrests (2015$) 95
Table 28. Latency Periods Used in Lung Cancer Risk Assessment Papers 95
Table 29. Percent Lung and Bronchus Cancer Incidence by Age and Distribution of Risk Reduction by Age
for an Exposure Change at 55 96
Table 30. Income Elasticity Estimates for Minor Health Effects, Severe Health Effects, and Mortality... 100
Table 31. Income-Based WTP Adjustments by Health Effect and Year 102
Table 32. Low Concentration PM2.5 Exposures from the ACS CSP-II, Medicare, and CanCHEC Cohorts.. 106
Table 33. PM2.5-Attributable ACS CSP-II Mortality Risk Estimates per 10 ng/m3 from Different Exposure
Estimation Techniques 108
Table 34. Single- and Two-Pollutant (Including 03 as a Copollutant) PM2.5-Attributable Mortality Risk
Estimates per 10 ng/m3 108
v
-------
Table 35. Hazard Ratios per 10 ng/m3 Estimated Using Three Causal Inference Approaches in (Wu et al.,
2020) 109
Table 36. Complex and Basic Cox Proportional Hazard Model Estimates of PM25-Attributable Mortality
per 10 ng/m3 109
Table 37. Potential Sensitivity of Estimated Instances of Asthma Onset 110
Table 38. Beta Coefficients and Standard Errors (SE) from Studies of Examining Long-term PM25
Exposure and New Onset Asthma in Children Ill
Table 39. Potential Sensitivity of Estimated Cardiovascular Hospital Admissions 112
Table 40. PM25-Attributable Cardiovascular Hospital Admissions Beta Estimates 112
Table 41. Potential Respiratory Hospital Admission Sensitivity Insights 113
Table 42. PM25-Attributable Respiratory Hospital Admissions Beta Risk Estimates 114
Table 43. Comparison of the PM25-Attributable Respiratory Hospital Admissions Beta Risk Estimate to
the EHA Respiratory Estimate 114
Table 44. PM2 5 At-Risk Study Identification Criteria 117
Table 45. Identified PM2 5 At-Risk Beta Coefficients and Standard Errors 118
Table 46. Single- and Two-Pollutant (Including PM2 5 as a Copollutant) Long-Term 03-Attributable
Respiratory Mortality Risk Estimates per 10 ppb 119
Table 47. Single- and Two-Pollutant (Including PMi0 as a Copollutant) Short-Term 03 Exposure 03-
Attributable Excess Premature Respiratory Mortality Risk Estimates per 10 ppb 120
Table 48. Long-Term 03-Attributable Total Mortality Risk Estimates per 10 ppb 121
Table 49. Potential Sensitivity of Estimated Instances of Asthma Onset 122
Table 50. Long-Term 03-Attributable Asthma Beta Coefficients 122
Table 51. 03 At-Risk Study Identification Criteria 124
Table 52. Identified 03 At-Risk Beta Coefficients and Standard Errors 126
Table 53. Scaling Factors for Various Lung Cancer Lag Cessation Distribution Models 130
Table 54. Ranges of Elasticity Values Used to Account for Projected Real Income Growth3 133
Table 55. Ranges of Adjustment Factors Used to Account for Projected Real Income Growth3 133
Table 56. Sensitivity of Monetized Benefits to Alternative Income Elasticities3 133
Table 57. Annual Alzheimer's Disease Valuation Estimate Calculation 134
Table 58. Lifetime Alzheimer's Disease Valuation Estimate Calculation (2015$) 135
Table 59. Additional Lifetime Alzheimer's Disease Valuation Estimate Calculation with a 3% Discount
Rate (2015$) 135
Table 60. Annual Parkinson's Disease Valuation Estimate Calculation 136
Table 61. Lifetime Parkinson's Disease Valuation Estimate Calculation 137
vi
-------
List of Figures
Figure 1. Illustrative Diagram of Potential Biological Pathways of Health Effects Following Pollutant
Exposure 17
Figure 2. Potential Biological Pathways for Cardiovascular Effects Following Short-Term PM2.5 Exposure
18
Figure 3. Potential Biological Pathways for Cardiovascular Effects Following Long-Term PM2.5 Exposure 19
Figure 4. Potential Biological Pathways for Respiratory Effects Following Short-Term PM2.5 Exposure.... 19
Figure 5. Potential Biological Pathways for Respiratory Effects Following Long-Term PM2.5 Exposure 20
Figure 6. Potential Biological Pathways for Cancer Effects Following Long-Term PM2.5 Exposure 20
Figure 7. Potential Biological Pathways for Nervous System Effects Following Long-Term PM2.5 Exposure
21
Figure 8. Potential Biological Pathways for Respiratory Effects Following Short-Term 03 Exposure 22
Figure 9. Potential Biological Pathways for Respiratory Effects Following Long-Term 03 Exposure 23
Figure 10. Potential Biological Pathways for Metabolic Effects Following Short-Term 03 Exposure 23
Figure 11. Age-Adjusted Trend in Hypertension Prevalence Among Adults Aged 18 and Over by Sex in the
U.S. Between 1999-2018 81
Figure 12. Cumulative Percentile of PM2.5 Cohort Exposure from the ACS CSP-II, Medicare, and CanCHEC
Cohorts 105
Figure 13. Modeled PM2.5 Exposure Distribution for NHIS Study Population, with Select Fitted Probability
Density Functions (from (Higbee et al., 2020), publisher permission pending) 107
Figure 14. Example County-Level (Distribution) and National-Level (Red Dot) Emergency Department
Visit and Hospital Admission Baseline Incidence Data 127
Figure 15. Lung Cancer Cases Cessation Lag Distribution by Model 131
Figure 16. Lung Cancer Cases Reduction Distribution 132
Figure 17. Correlation of MDA8 and DA8 03 Exposures Between 2000-2019 (R=0.986) 143
VII
-------
Preface
In March of 2021, EPA published a version of this Technical Support Document (TSD) titled "Estimating
PM2.5- and Ozone-Attributable Health Benefits" ((U.S. EPA 2021)). That original TSD drew upon evidence
published in the Integrated Science Assessments (ISA) for PM2.5 and ozone available at that point in time
((U.S. EPA 2019, U.S. EPA 2020)). EPA recently published a Supplement to the PM ISA ((U.S. EPA 2022)).
This updated version of the TSD evaluates this new evidence and identifies alternative epidemiologic
studies and risk estimates to support EPA benefits analyses.
EPA expects to revise this TSD as new health, demographic, and economic evidence becomes available.
The table below summarizes the: (1) version and publication date of each TSD; (2) reason for updating
the TSD; (3) key changes as compared to the prior version. Each version of the TSD may be found on the
EPA website at: www.epa.gov/benmap.
Version Date
Update Rationale
Key Changes
March 2021
Describe benefits methods
updated in response to the
2019 PM ISA and used to
support the Revised Cross-
State Air Pollution Update
Regulatory Impact Analysis
NA
December
2022
Describe benefits methods
updated in response to the
2022 Supplement to the PM
ISA and used to support the
Regulatory Impact Analysis for
the Proposed Reconsideration
of the PM NAAQS
• Revised risk estimates used to quantify PM2.5-
related mortality and non-fatal heart attacks
• Updated school loss day valuation estimate
• Updated discussion of uncertainty associated
with PM2.5-related premature mortality
June 2024
Updates in response to Circular
A-4 and SAB guidance.
• Added valuation estimates based on a 2%
Discount Rate
• Updated criteria for identifying studies and risk
estimates to include studies published since the
most recent ISA or equivalent and to include
studies published outside the U.S. and Canada
1
-------
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)). 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 ((Exec. Order No. 12898 1994), (OMB 2003)). 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 reducing an exposure to an environmental hazard, like PM2.5 or 03, depends on the
expected effect of those reductions on human health.
Estimating the health benefits of reducing PM2.5 and 03 exposure in a BCA begins with estimating the
population-level change in exposure and then estimating the population-level change in risk for those
health outcomes affected by exposure. The benefit of reducing these health risks is estimated using a
summary measure of the population-level WTP for the risk change.1 The greater the magnitude of the
risk reduction from a given change in concentration, the greater the 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.
This document describes all three steps for the purposes of estimating health benefits from changes in
ambient PM2.5 and 03 exposure.3
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.
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
2
-------
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, Revesz 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 ((Kivi and Shogren 2010, Honeycutt 2020)). 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 Regulatory Impact Analysis for the Proposed Reconsideration of the PM NAAQS,
the EPA quantifies and monetizes all health effects that the ISA draws conclusions regarding the causal
relationship between a pollutant and a given effect, noting whether the effect is "causal" or "likely to be
causal," following 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-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
scientific information (U.S. EPA (2014). Guidelines for Preparing Economic Analyses, National Center for
Environmental Economics. US Environmental Protection Agency Washington, DC.).
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 (Freeman III, A. M., J. A.
Herriges and C. L. Kling (2014). The measurement of environmental and resource values: theory and methods,
Routledge.). 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.
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.
3
-------
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 Proposed Particulate Matter National Ambient Air
Quality Standards (hereafter, PM NAAQS). This document is organized according to the three key steps
we follow to collect information the information needed to quantify and monetize benefits, as
presented in section 1.1:
1. Establish criteria for identifying studies and risk estimates most appropriate to inform a PM? s
and O^ 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 Establishing these criteria prior to selecting candidate health endpoints helps reduce
the chance of bias in the process for selecting 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). Each ISA identifies some number of
clinically significant air pollution-related outcomes (e.g., premature mortality and hospital
admissions) ((U.S. EPA 2019, U.S. EPA 2020)) that may be quantified and possibly also
monetized. Each selected endpoint and study must meet the criteria specified in section 2.1
above.
3. Collect baseline incidence and prevalence estimates (section 3) and demographic information
(section 4). EPA uses health impact functions to quantify counts of pollutant-attributable effects.
These functions uses population and rates of death and disease. EPA selected either daily or
annual baseline incidence and prevalence rates at the most geographically- and age-specific
levels feasible for each health endpoint assessed. Projected population counts are drawn from
the 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.
4. Develop economic unit values (section 5). To compare benefits estimates associated with a
rulemaking directly with the estimated cost, counts of each air pollution-attributable event must
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.
4
-------
be expressed as an economic value. This step requires us to estimate a single year or multi-year
stream of discounted values for each unique health endpoint.
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.
5
-------
2 Approach to Identifying Studies and Risk Estimates10
This section describes the criteria EPA developed when selecting among 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 2.3);
and, finally, present the identified health endpoints and risk estimates (section 2.4) 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. Clearly specifying criteria for identifying such studies helps
ensure EPA transparently specifies its scientific judgement. 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, or have been published since the latest externally reviewed ISA or equivalent
assessment and represent an improvement upon the scientific literature at the time of the latest
externally reviewed ISA or equivalent assessment.
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
6
-------
2. 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 (and an
accompanying estimated standard error) to provide the information needed to quantify mean
estimates of effects and 95% confidence intervals.
2.1.2 Preferred Criteria Categories
Studies meeting the minimum criteria are then evaluated based on various factors, which we call
preferred criteria, to identify risk estimates that best characterize air pollution 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.
Preferred criteria are developed to ensure study quality and suitability to a benefits analysis and
established prior to evaluating any individual study or risk estimates. Conversely, EPA does not consider
factors such as the magnitude or variance of the reported risk estimate; accounting for these factors
might inadvertently bias our choice of studies or risk estimates.14
Each preferred criteria (Table 1) is considered 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 a benefits analysis. In practice, no single study or risk estimate
will possess the ideal attributes for all criteria and thus we consider 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 to Use in a Benefits Assessment
Study
Attributes1
Key Factors to Consider when Evaluating Study Quality
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 are less prone
to exposure estimate bias and generate higher-resolution of estimates than exposure data
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 (2019). Integrated
Science Assessment (ISA) for Particulate Matter (Final Report). U. S. EPA. Research Triangle Park, NC, U.S.
Environmental Protection Agency, Office of Research and Development, Center for Public Health and
Environmental Assessment, U.S. EPA (2020). Integrated Science Assessment for Ozone and Related Photochemical
Oxidants (Final Report). O. o. R. a. Development. Washington, DC, U.S. Environmental Protection Agency.). These
figures illustrate the heterogeneity in the size of the reported effect among this subset of studies and risk
estimates.
7
-------
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
Studies conducted in either the U.S. or Canada and representing air quality conditions,
affected populations, and other underlying characteristics of the U.S are preferred.18 U.S. or
Canadian studies are preferred 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. Studies using data from
non-U.S. locations studies are considered when U.S study options are limited or less
informative. Similarity to the U.S. context is a priority for non-U.S. studies. 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
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.
18 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, E. a. C. C. (2016). Canada-
United States Transboundary Particulate Matter Science Assessment 2013, Environment and Climate Change
Canada, CBC (2016). Canada by the Numbers. CBC News. Online, U.S. EPA (2019). Policy Assessment for the Review
of the Ozone National Ambient Air Quality Standards, External Review Draft. U. S. EPA. Research Triangle Park, NC,
U.S. Environmental Protection Agency, Office of Air Quality Planning and Standards, Heath and Environmental
Impacts Division.).
8
-------
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
through a case crossover study design. Study sizes are subject to diminishing returns when
the number of subjects become very large (e.g., millions) and thus relatively larger study
sizes (i.e., not necessarily the largest study sizes) 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 2020)).
'Although preferred criteria categories are not hierarchical, not all criteria are weighted equally, and expert judgement is involved.
2.1.2.1 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. Rank-ordered priority
Study
Attributes
Rank-Ordered Criteria to Consider when Evaluating Study Quality
Study Period
1. Most recent years with overlapping air quality and health data
9
-------
Study
Attributes
Rank-Ordered Criteria to Consider when Evaluating Study Quality
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
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
10
-------
Study
Attributes
Rank-Ordered Criteria to Consider when Evaluating Study Quality
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.
2.2 Available Epidemiologic Literature
We follow a structured and transparent process to identify individual epidemiologic studies from the
large body of available epidemiologic literature (section 2.1). This process involves the identifying 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 evaluated using the criteria identified in Table 1 and Table 2.
2.2,1 Identification of Exposure-Attributable Health Outcomes
Our procedure 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 "candidate" health endpoint must satisfy the below conditions prior to being included in the main
benefits assessment:
• The endpoint category (e.g., respiratory effects) is causally related to exposure (section 2.2.1.1)
• The specific health endpoint (e.g., exacerbated asthma) is a biologically plausible health effect of
exposure (section 2.2.1.2)
The review of the National Ambient Air Quality Standards (NAAQS) follows a structured and transparent
process for evaluating scientific information reported in the ISAs when determining whether a given
endpoint is causally related to the pollutant. 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 outcomes19 (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).
19 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.
11
-------
Our use of the ISAs to select among endpoints is consistent with the advice provided by a 2002 National
Academy of Science review, which found that "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 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 2015, U.S. EPA 2019)).20 As such, EPA relies on the systematic and
Clean Air Scientific Advisory Committee-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, 2022 PM Supplement, and 2020 03 ISAs
((U.S. EPA 2019, U.S. EPA 2020, U.S. EPA 2022)). 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 when assessing the causal relationship between each pollutant and health
outcome. 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
20 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). Integrated Science Assessment for Particulate Matter (Final Report).
U. S. EPA. Research Triangle Park, NC, Office of Research and Development, National Center for Environmental
Assessment, U.S. EPA (2019). Integrated Science Assessment (ISA) for Particulate Matter (Final Report). U. S. EPA.
Research Triangle Park, NC, U.S. Environmental Protection Agency, Office of Research and Development, Center
for Public Health and Environmental Assessment., 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 Osexposure made in the 2019 PM ISA and 2020 03 ISAs. This process
ensures a thorough and transparent strategy for literature identification.
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.
12
-------
effects, respiratory effects, nervous system effects, metabolic effects, etc.) using a weight-of evidence
approach ((U.S. EPA 2015, U.S. EPA 2019)), 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
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, 2022 supplemental 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 2019, U.S. EPA 2020), (U.S.
EPA 2022)).23 There were no "causal" or "likely to be causal" relationships for PM10-2.5 or ultrafine
particles in the 2019 PM ISA or 2022 supplemental 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 2019, U.S. EPA
2020)), as discussed in more detail on section 2.3.
22 See Preamble to Integrated Science Assessments, EPA/600/R-15/067,
https://cfpub.epa.gov/ncea/isa/recordisplay.cfm7deich347534
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. The 2022 Supplemental PM ISA did not update causality determinations,
but instead updated the literature associated with cardiovascular and mortality outcomes (U.S. EPA (2022).
Supplement to the 2019 Integrated Science Assessment for Particulate Matter (Final Report). U. S. EPA. Research
Triangle Park, NC, U.S. Environmental Protection Agency, Office of Research and Development, Center for Public
Health and Environmental Assessment.).
24 Ultrafine particles are generally considered to have an aerodynamic diameter less than or equal to 0.1 nm.
13
-------
Table 3. Causality Determined for PM2.5-Related Health Effects
Exposure
Health Outcome
2009 ISA Determinations
2019 ISA Determinations23
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
Suggestive of, but not
Suggestive of, but not
Reproduction and Fertility
sufficient to infer
sufficient to infer
Pregnancy and Birth
Suggestive of, but not
Suggestive
Outcomes
sufficient to infer
of, but not sufficient to infer
Mortality1
Causal
Causal
Cardiovascular Effects
Causal
Causal
Short-
term
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 Determined 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
14
-------
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 draw conclusions regarding the causal nature of relationships between PM and health and non-
ecological welfare effects. These conclusions apply to 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 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 2015)). 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.
25 Information in the biological plausibility diagrams includes studies identified in previous ISAs and Air Quality
Criteria Documents (AQCDs).
15
-------
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
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.
16
-------
Note: For additional information, please refer to the original biological plausibility diagrams in the ISAs ((U.S. EPA 2019, U.S. EPA
2020)).
Figure 1. Illustrative Diagram of Potential Biological Pathways of Health Effects Following Pollutant
Exposure.
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.
17
-------
2.2.1.2.1.1 Cardiovascular Effects
The 2019 PM ISA diagram of biological pathways for cardiovascular effects following short-term PM2.5
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.
Emergency
Department
Visits/
Hospital
Admissions
and/or Mortality
v /
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).
18
-------
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).
19
-------
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 2019)).
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
20
-------
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 2019)). 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
Cerebral Cortex
Hippocampus
Substantia Nigra
Hypothalamus
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 estimated 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.
21
-------
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
22
-------
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.
—
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 Identifying Quantifiable Health Outcomes27
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
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). Guidelines for Preparing Economic Analyses, National
Center for Environmental Economics. US Environmental Protection Agency Washington, DC
treated as separable from monetization given resource and data limitations (section 1.2).
Quantification is
23
-------
health endpoints (e.g., hospital admissions for cardiovascular ICD codes 390-459 or the
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.2S'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.
Title
Title of the study.
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.
24
-------
Abstract
Abstract of the study.
Publication Year
Year the article was published, according to PubMed.
Pollutant
PM2.5 or 03.
ISA Source
Integrated Science Assessment where study was initially referenced,
beginning with the 2019 PM ISA and 2020 Ozone ISA.
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 aged 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.
25
-------
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 account for both
the within-study variances and the between-study variance when weighting.
Multiple epidemiology studies might assess a common endpoint using very different input parameters
and analytical assumptions,31 making it difficult to pool this literature. 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).
Recently developed hybrid methods for estimating population exposure allow researchers to
characterize pollutant concentrations at more detailed temporal and spatial scales than those methods
using monitoring data alone. 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.
Conversely, while consistency between studies is generally desirable when pooling, there are some
instances when it could introduce uncertainty and/or bias. For example, pooling Hazard Ratios reported
in two or more studies of the same cohort, or Hazard Ratios reported for alternative cohorts, poses
special challenges; this procedure generally requires individual-level data not readily available for most
cohorts ((Burnett, Chen et al. 2018)). 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 by weighting each study according to the inverse of the reported variance, generating a
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.
26
-------
2.2.4.2 Individual Alternate Risk Estimates
In situations where multiple risk estimates should not or cannot be pooled33, we instead estimate
incidence using each individual risk estimate. 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 many 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.
2.3 Systematically Identifying 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.
2.3.1 PM2.5
The following sections of the 2019 PM ISA and 2022 PM ISA Supplement correspond to health endpoints
judged as having a "causal" or "likely to be causal" relationship with PM2.5 exposure, respectively:34
• 5.1 Short-Term PM2.5 Exposure and Respiratory Effects,
• 5.2 Long-Term PM2.5 Exposure and Respiratory Effects,
• 6.1 and 3.1 Short-Term PM2.5 Exposure and Cardiovascular Effects,
• 6.2 and 3.1 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 and 3.2 Long-Term PM2.5 Exposure and Total Mortality
Following the approach to identifying available epidemiologic literature (section 2.2), we began with the
2,882 studies cited by the 2019 PM ISA or 2022 PM ISA Supplement ((U.S. EPA 2019, U.S. EPA 2022)). Of
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, R., H. Chen, M. Szyszkowicz, N. Fann, B. Hubbell, C. A. Pope, 3rd, J. S. Apte, M. Brauer, A. Cohen, S.
Weichenthal, J. Coggins, Q. Di, B. Brunekreef, J. Frostad, S. S. Lim, H. Kan, K. D. Walker, G. D. Thurston, R. B. Hayes,
C. C. Lim, M. C. Turner, M. Jerrett, D. Krewski, S. M. Gapstur, W. R. Diver, B. Ostro, D. Goldberg, D. L. Crouse, R. V.
Martin, P. Peters, L. Pinault, M. Tjepkema, A. van Donkelaar, P. J. Villeneuve, A. B. Miller, P. Yin, M. Zhou, L. Wang,
N. A. H. Janssen, M. Marra, R. W. Atkinson, H. Tsang, T. Quoc Thach, J. B. Cannon, R. T. Allen, J. E. Hart, F. Laden, G.
Cesaroni, F. Forastiere, G. Weinmayr, A. Jaensch, G. Nagel, H. Concin and J. V. Spadaro (2018). "Global estimates of
mortality associated with long-term exposure to outdoor fine particulate matter." Proc Natl Acad Sci U S A 115(38):
9592-9597.).
34 Due to the cutoff date of the 2022 Supplement to the PM ISA (March 2021) and the recent and the evolving
nature of the pandemic, we are unable to assess associations between PM2.sand COVID-19.
27
-------
these, 559 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).35 Of these, 150 studies met the minimum required criteria (section 2.1.1).36
We preferentially selected epidemiologic studies specifying ICD-9 or ICD-10 codes that reflected a broad
array of adverse effects; the ISA reports that the evidence for these broader effects is stronger than it is
for specific outcomes, helps avoid double-counting of health benefits across categories and is consistent
recommendations from the EPA Science Advisory Board.37 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).
35 Mortality studies were treated slightly differently. More information is available in section 2.3.1.1.1.
36 This number may not equal the sum of available studies in Table 6 as individual studies may present risk
estimates for multiple health endpoints.
37 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, B. C., 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. M. Leavitt. U.S. EPA HQ, Washington DC,
Office of the Administrator, Science Advisory Board.). 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."
28
-------
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
All-Cause Mortality
(LT)
2
1
Infants
1
1
35
3
Adults and older adults
115
1
Older adults
70
1
Cardiovascular
Hospital Admissions
(ST)
15
1
Children, adults, and older
adults
28
7
Cardiovascular
Emergency
Department Visits (ST)
2
1
Children, adults, and older
adults
3
1
Acute Myocardial
Infarction (ST)
1
1
Adults and older adults
4
1
Stroke (LT)
4
1
Older adults
1
1
Cardiac Arrest (ST)
3
3
Adults and older adults
12
1
12
1
7
1
Respiratory Hospital
Admissions (ST)
12
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
PM NAAQS RIA will continue to be utilized.
*See associated Study Information Table for specific study details.
2.3.1.1 All-Cause Mortality
The 2022 supplement to the PM ISA re-affirmed that a "causal" relationship exists between both long-
and short-term PM2.5 exposure and all-cause mortality ((U.S. EPA 2022)). Specifically, the 2022
supplement to the ISA states that:
29
-------
Recent epidemiologic studies published since the 2019 PM ISA support and extend the
evidence that contributed to the conclusion of a causal relationship between long-term
PM2.5 exposure and cardiovascular effects. Numerous U.S. and Canadian cohort studies
conducted in locations where the long-term PM2.5 concentration are less than 13 /jg/m
add to the strong evidence base that was characterized in the 2019 PM ISA describing
the relationship between long-term PM2.sand cardiovascular mortality, and specifically
IHD- and stroke-related mortality. Overall, these recent cardiovascular mortality studies
reported positive associations at varying spatial scales and across different exposure
assessment and statistical methods. The associations between long-term PM2.5 exposure
and cardiovascular mortality generally persisted in models that were adjusted for ozone,
NO2, PM 10-2.5, or SO2, and most analyses of the C-R function supported a linear, no-
threshold relationship for cardiovascular mortality, especially at lower ambient
concentrations ofPM2.s
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) and so we assume that effects observed in studies of long-
term exposures may also reflect the influence 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 PIVh.s-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" ((IEc 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)).
2.3.1.1.1 Available Epidemiologic Literature
Whereas for all other health endpoints we began by identifying North American epidemiologic studies
from the relevant ISA ((U.S. EPA 2019), Figures 11-17 and 11-18), available literature for this health
endpoint had been further reviewed by EPA in the 2022 PM Policy Assessment (PA) ((U.S. EPA 2022)
sections 3.3 and Figure 3-4). 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 35 epidemiologic multi-city cohort studies identified in the 2022 PM PA, which all met
the minimum criteria and accounts for studies identified in the supplement to the PM ISA ((U.S. EPA
2022), section 2.1.1).
We separately evaluated the more limited literature available regarding PM2.5-attibutable infant
mortality (ages 0-12 months) cited in the 2009 ISA, as no more recent studies of PM2.5-attributable all-
cause infant mortality were available in either the 2019 ISA or the 2022 supplement to the PM ISA. Full
study information can be found in the Study Information Table (section 2.2.3).
30
-------
2.3.1.1.2 Identifying Suitable Studies for Use in Benefits Assessments
The systematic identification criteria (section 0) was applied to the 35 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.
The 35 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.
2.3.1.1.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
We select multiple cohort risk estimates to estimate counts of PM-related premature death, following
an approach employed in previous RIAs (e.g., (U.S. EPA 2011, U.S. EPA 2011, U.S. EPA 2011, U.S. EPA
2012, U.S. EPA 2012, U.S. EPA 2015, U.S. EPA 2019)). Quantifying effects using risk estimates reported in
alternative cohorts helps account for uncertainty in the estimated number of PM-related premature
deaths.
The systematic approach led us to identify three studies best characterizing risk across the U.S. ((Turner,
Jerrett et al. 2016, Pope III, Lefler et al. 2019, Wu, Braun et al. 2020)).38,39 These three studies used data
from three cohorts, an analysis of Medicare beneficiaries (Medicare), the American Cancer Society (ACS)
and the National Health Interview Survey (NHIS). The supplement to the PM ISA concluded that the NHIS
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 2019, U.S. EPA 2022)). We
discuss uncertainty and sensitivity considerations related to the identified mortality risk estimates in
sections 6.1.2 and 6.5.40
38 The 2020 and 2022 PM PAs 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 (2019). Integrated Science
Assessment (ISA) for Particulate Matter (Final Report). U. S. EPA. Research Triangle Park, NC, U.S. Environmental
Protection Agency, Office of Research and Development, Center for Public Health and Environmental Assessment,
U.S. EPA (2022). Policy Assessment for the Reconsideration of the National Ambient Air Quality Standards for
Particulate Matter (Final Report). U. S. EPA. Research Triangle Park, NC, U.S. Environmental Protection Agency,
Office of Air Quality Planning and Standards, Heath and Environmental Impacts Division.).
39 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 (2012). Regulatory
Impact Analysis for the Final Revisions to the National Ambient Air Quality Standards for Particulate Matter. U. S.
EPA. Research Triangle Park, NC, U.S. Environmental Protection Agency, Office of Air Quality Planning and
Standards, Health and Environmental Impact Division.).
40 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 (2019). Integrated Science Assessment (ISA) for Particulate Matter (Final Report). U. S. EPA.
Research Triangle Park, NC, U.S. Environmental Protection Agency, Office of Research and Development, Center
31
-------
2.3.1.1.3.1 Adult Mortality
2.3.1.1.3.1.1 NHIS
(Pope III, Lefler et al. 2019) constructed a cohort using a survey of 1,599,329 U.S. adults (aged 18-84
years) who were interviewed in the National Health Interview Surveys (NHIS) between 1986 and 2014
and linked to the National Death Index (NDI) through 2015. The authors then performed survival
modeling in this cohort to examine the relationship between long-term PM2.5 exposure and all-cause
mortality. The authors also constructed a subcohort of 635,539 adults from the full cohort for whom
body mass index (BMI) and smoking status data were available. The authors employed a hybrid
modeling technique to estimate annual-average PM2.5 concentrations derived from regulatory
monitoring data and constructed in a universal kriging framework using geographic variables including
land use, population, and satellite estimates. (Pope III, Lefler et al. 2019) assigned annual-average PM2.5
exposure from 1999-2015 to each individual by census tract and utilized complex (accounting for NHIS's
sample design) and simple Cox proportional hazards models for the full cohort and the subcohort. We
select the Hazard Ratio calculated using the complex model for the subcohort, which controls for
individual-level covariates including age, sex, race-ethnicity, inflation-adjusted income, education level,
marital status, rural versus urban, region, survey year, BMI, and smoking status. In a single-pollutant
model, the coefficient and standard error for PM2.5 are estimated from the hazard ratio (1.12) and 95%
confidence interval (1.08-1.15) associated with a change in annual mean PM2.5 exposure of 10 ug/m3
(Pope et al. 2019, Table 2, Subcohort).
2.3.1.1.3.1.2 Medicare
(Wu, Braun et al. 2020) evaluated the relationship between long-term PM2.5 exposure and all-cause
mortality in more than 68.5 million Medicare enrollees (over the age of 64), using Medicare claims data
from 2000-2016 representing over 573 million person-years of follow-up and over 27 million deaths.
This cohort included over 20% of the U.S. population and was, at the time of publishing, the largest air
pollution study cohort to date. The authors modeled PM2.5 exposure at a 1-km2 grid resolution using a
hybrid ensemble-based prediction model that combined three machine learning models and relied on
satellite data, land-use information, weather variables, chemical transport model simulation outputs,
and monitor data. (Wu, Braun et al. 2020) fit five different statistical models: a Cox proportional hazards
model, a Poisson regression model, and three causal inference approaches (GPS estimation, GPS
matching, and GPS weighting). All five statistical approaches provided consistent results; we report the
results of the Cox proportional hazards model here. The authors adjusted for numerous individual-level
and community-level confounders, and sensitivity analyses suggest that the results are robust to
unmeasured confounding bias. In a single-pollutant model, the coefficient and standard error for PM2.5
are estimated from the hazard ratio (1.066) and 95% confidence interval (1.058-1.074) associated with a
change in annual mean PM2.5 exposure of 10.0 ug/m3 ((Wu, Braun et al. 2020), Table S3, Main analysis,
2000-2016 Cohort, Cox PH).
2.3.1.1.3.1.3 American Cancer Society
Two independent studies evaluated the same years of data from the large, nationwide ACS CSP-II cohort
of those > 29 years old ((Turner, Jerrett et al. 2016, Pope III, Lefler et al. 2019)). These studies extended
for Public Health and Environmental Assessment., we assume that all fine particles are equally potent in causing
mortality, and 3) following conclusions of the ibid., we assume that the health impact function for fine particles is
linear within the range of ambient concentrations affected by these standards.
32
-------
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.
In addition to adjusting for individual-level and ecological covariates, (Turner, Jerrett 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 2022)), we identified it as the most suitable
hazard ratio when estimating health benefit impacts. Thus, the total mortality risk estimate is based on
the random-effects Cox proportional hazard model that incorporates multiple individual and ecological
covariates (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, Jerrett 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.
2.3.1.1.3.1.4 Summary
Based on the 2019 PM ISA, EPA has used two estimates of mortality: one from the American Cancer
Society cohort and one from the Medicare cohort ((Turner, Jerrett et al. 2016) and (Di, Wang et al.
2017), respectively). We use a risk estimate from (Pope III, Lefler et al. 2019) study in place of the risk
estimate from the ((Turner, Jerrett et al. 2016) analysis, as it: (1) includes a longer follow-up period that
includes more recent (and lower) PM2.5 concentrations; (2) the NHIS cohort is more representative of
the U.S. population than is the ACS cohort with respect to the distribution of individuals by race,
ethnicity, income and education.
Based on the 2020 Supplement to the PM ISA, EPA substituted a risk estimate from (Wu, Braun et al.
2020) in place of a risk estimate from (Di, Wang et al. 2017). These two epidemiologic studies share
many attributes, including the cohort and model used to characterize population exposure to PM2.5- As
compared to (Di, Wang et al. 2017), (Wu, Braun et al. 2020) includes a longer follow-up period and
reflects more recent PM2.5 concentrations.
2.3.1.1.3.2 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)).41 The 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. The supplement to the PM ISA
did not identify any new studies characterizing PM exposure and mortality among children ((U.S. EPA
2022)).
Although the ISA evidence supports a relationship between PM2.5 exposure and death in children, only
studies of infant mortality met the minimum criteria (section 2.1.1). Previously, EPA has included
estimates of post neonatal infant mortality from (Woodruff, Grillo et al. 1997) ((U.S. EPA 2012, U.S. EPA
41 For the purposes of this analysis, we only calculate benefits for infants aged 0-12 months, not all children under
5 years old.
33
-------
2019)). As such, compared to avoided deaths estimated for adult mortality, avoided deaths for infants
are significantly smaller due to the size of the population and the smaller risk estimates associated with
exposure to PM from epidemiology studies on infant mortality.
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,
Darrow 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.42 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.
2.3.1.2 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 2019), 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).
2.3.1.2.1 Available Epidemiologic Literature
Fifteen North American epidemiologic studies of cardiovascular hospital admissions43 were identified in
section 6.1 of the PM ISA ((U.S. EPA 2019)) and in the Supplement to the PM ISA ((U.S. EPA 2019, U.S.
EPA 2022)). 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.
2.3.1.2.2 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, Son et al. 2015)
evaluated the most recent study period and included the most nationally representative study locations.
42 Odds ratios are a subtype of risk estimates.
43 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).
34
-------
2.3.1.2.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Bell, Son 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, Son 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.
2.3.1.3 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 2019) 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).
2.3.1.3.1 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 2019)).
2.3.1.3.2 Study and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Ostro, Malig 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, Malig 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 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.
2.3.1.4 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 2019), section 6.1). The ISA also called out conductance abnormalities as a key clinical
effect associated with both short-and long-term PM2.5 exposures (section 2.2.1.2.1.1).
35
-------
This endpoint, like several others (e.g., lung cancer incidence, section 2.3.1.14) has a very high rate of
fatality. As mortality due to any cause is captured separately (section 2.3.1.1), we focus on impacts
following cardiac arrest, in the population that survive the initial event when considering this health
endpoint.44
2.3.1.4.1 Available Epidemiologic Literature
The 2019 PM ISA included three epidemiologic studies of out-of-hospital cardiac arrest that met our
minimum identification criteria ((U.S. EPA 2019), section 2.1.1).
2.3.1.4.2 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 ((Rosenthal, Carney et al. 2008, Silverman, Ito et al. 2010, Ensor, Raun et al. 2013)). 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.45
2.3.1.4.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Ensor, Raun et al. 2013) studied the association between short-term ambient air pollution (PM2.5 and
03) exposure and out-of-hospital cardiac arrest. (Ensor, Raun 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).
(Silverman, Ito et al. 2010) investigated the link between short-term ambient air pollution exposure
(PM2.5, N02, 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, Ito et al.
2010) found that in a single-pollutant case crossover model, each 10 ng/m3 increase in ambient PM2.5
44 Similarly, as any emergency department visits or hospital admissions resulting from cardiac arrest would be
included in other endpoints (sections 2.3.1.2 and 2.3.1.3), monetized benefits of this health endpoint would not
include and emergency department visits or hospital admissions costs.
45 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.
36
-------
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, Carney 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, Carney 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, Raun et al. 2013) and (Silverman, Ito 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, Carney 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.
2.3.1.5 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 ((U.S. EPA 2019)). In addition, the 2022 Supplement to
the 2019 PM ISA stated:
Studies examining short-term PM2.5 exposure report consistent positive associations for
cardiovascular-related emergency department (ED) visits and hospital admissions, specifically
for...myocardial infarction (Ml).... The epidemiologic studies reviewed in the 2019 PM ISA
strengthened the evidence characterized in the previous ISA (U.S. EPA, 2009). Most of the
evidence for IHD and Ml in the 2009 PM ISA was from multicity epidemiologic studies of ED visits
and hospital admissions [i.e., the U.S. Medicare Cohort Air Pollution Study (MCAPS) (Dominici et
al., 2006), a four-city study in Australia (Barnett et al., 2006), and a study among older adults in
several French cities (Host et al., 2008)]. The positive associations reported in these studies were
an important line of evidence in the 2009 PM ISA concluding a causal relationship between short-
term PM2.5 exposure and cardiovascular effects. Uncertainties noted in the 2009 PM ISA with
respect to exposure measurement error for those not living near a PM2.5 monitor were reduced in
the 2019 PM ISA with the consideration of studies that applied hybrid exposure assessment
techniques that combine land use regression data with satellite aerosol optical depth (AOD)
measurements and PM2.5 concentrations measured at fixed-site monitors to estimate PM2.5
concentrations. Further, compared with the 2009 PM ISA, the evidence in the 2019 PM ISA was
expanded to include studies examining the association of short-term PM2.5 exposure ST segment
depression in addition to ED visits and hospital admissions for Ml.... Overall, recent studies
37
-------
support and extend the findings of the 2019 PM ISA with additional studies reporting positive
associations between short-term PM2.5 exposure and...Ml hospital admissions and ED visits ((U.S.
EPA 2022)).
2.3.1.5.1 Available Epidemiologic Literature
The 2019 PM ISA identified epidemiologic studies associating AMIs with short-term PM2.5 exposures,
though the studies passing the initial screening stage were not more suitable than those currently used
to estimate benefits ((U.S. EPA 2019)). The Supplement to the ISA identified a single suitable short-term
exposure study that passed the screening stage ((U.S. EPA 2022)).46
2.3.1.5.2 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
The supplement to the PM ISA identified a single new epidemiologic study examining the relationship
between short-term PM exposure and AMI ((U.S. EPA 2022)). (Wei, Wang et al. 2019) evaluated the
relationship between short-term PM2.5 exposure and hospital admissions for 214 mutually exclusive
disease groups, including acute myocardial infarction, in a time-stratified, case-crossover analysis of over
95 million Medicare inpatient hospital claims from 2000-2012. The authors estimated daily PM2.5 levels
at a 1-km2 grid cell level using a satellite based, neural network model that was calibrated using monitor
data and assigned 0-1 day lagged PM2.5 exposure to each participant by zip code of residence. For each
disease group, (Wei, Wang et al. 2019) created a case crossover dataset that controlled for individual
level and zip code level variables, day of the week, seasonality, and long-term time trends. The authors
used conditional logistic regression models to estimate associations between PM2.5 exposure and risk of
hospital admission and found positive associations for numerous rarely studied and numerous well-
studied disease groups. In a single-pollutant model, the coefficient and standard error are estimated
from a reported relative increase in risk (0.11%) and 95% confidence interval (0.07%-0.16%) associated
with a 1 ug/m3 increase in 0-1 day lagged PM2.5 exposure ((Wei, Wang et al. 2019), Figure 3, CCS 100
AMI). We thus based our estimates of AMI incidence on this study, assuming that all AMI were
associated with a hospital admission.
2.3.1.6 Stroke
2.3.1.6.1 Available Epidemiologic Literature
The 2019 PM ISA included three epidemiologic studies of stroke that met the minimum identification
criteria ((U.S. EPA 2019), section 2.1.1).
2.3.1.6.2 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, Coull et al. 2012)).
46 Five long-term exposure studies evaluated associations between PM2.5 and AMI, however the Supplement to the
PM ISA stated that "evidence informing the relationship between long-term exposure to PM2.5 and IHD, including
the recent studies of Ml,... do not all report positive associations; however, the strongest evidence of a
relationship continues to be for those with preexisting diseases or patient populations that are followed after a
cardiac event or procedure such as catheterization." Due to the strength of the evidence, AMI associated with
long-term PM2.5 exposures was not included as a health endpoint.
38
-------
2.3.1.6.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Kloog, Coull 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.
2.3.1.7 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 2019), 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." 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).
2.3.1.7.1 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.
2.3.1.7.2 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.
39
-------
2.3.1.7.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Bell, Son 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, Son
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, Roth 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, Roth
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 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, Roth 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 ng/m3 increase in daily mean PM2.5 concentrations, lagged by 3 days, came from a single-pollutant
model.
2.3.1.8 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 2019), 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).
40
-------
2.3.1.8.1 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.
2.3.1.8.2 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.
2.3.1.8.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Krai I, Anderson 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, Anderson 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, Anderson 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 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.
2.3.1.9 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 2019)). 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 2019)).
2.3.1.9.1 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 2019)). 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.
41
-------
Five North American epidemiologic studies of asthma onset in children were identified in section 5.2 of
the 2019 PM ISA ((U.S. EPA 2019)). 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.
2.3.1.9.2 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.47 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.
2.3.1.9.3 Study Identified as Most Suitable for Use in Benefits Assessments
(Tetreault, Doucet 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-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, Doucet et al. 2016) showed that childhood asthma onset may be associated with exposure to
PM2.5, NO2, and O3.
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 stratified estimates using exposure estimates at birth.
The study identified as best characterizing risk across the U.S. was (Tetreault, Doucet 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, Doucet 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.
47 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).
42
-------
As the physiology and etiology of lung development in children is similar in children 6-17, we apply the
4-12 year age-stratified effect estimate from (Tetreault, Doucet et al. 2016) to children ages 4-17
((Sparrow, O'Connor et al. 1991, Guerra, Wright et al. 2004, Ochs, Nyengaard et al. 2004, Baena-
Cagnani, Rossi et al. 2007, Trivedi and Denton 2019)).
2.3.1.10 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" ((U.S. EPA 2019)).
2.3.1.10.1 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.
2.3.1.10.2 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.48
2.3.1.10.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Rabinovitch, Strand 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 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, Strand 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. As the
48 https://www.mayoclinic.org/diseases-conditions/asthma/in-depth/asthma-medications/art-20045557
43
-------
physiology and etiology of lung development in children is similar in children 6-17, we apply the 6-13-
year age-stratified effect estimate to children ages 6 to 17 ((Sparrow, O'Connor et al. 1991, Guerra,
Wright et al. 2004, Ochs, Nyengaard et al. 2004, Baena-Cagnani, Rossi et al. 2007, Trivedi and Denton
2019)).
2.3.1.11 Allergic Rhinitis (Hay Fever/Respiratory Allergies)
The 2019 PM 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 2019), 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 2019), 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.
2.3.1.11.1 Available Epidemiologic Literature
The 2019 PM ISA identified one epidemiologic study of long-term 2019 PM2.5 exposure and allergic
rhinitis ((U.S. EPA 2019)).
2.3.1.11.2 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Parker, Akinbami 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, Akinbami et al. 2009) placed all study
participants reporting symptoms of respiratory allergies or hay fever into a combined rhinitis group.
(Parker, Akinbami 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 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.
2.3.1.12 Minor Restricted Activity Days
No new epidemiologic studies of minor restricted activity days (MRADs) were identified in the 2019 PM
ISA ((U.S. EPA 2019)).
2.3.1.12.1 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
44
-------
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.
2.3.1.13 Work Loss Days
No new studies of work loss days (WLDs) were identified in the 2019 PM ISA ((U.S. EPA 2019)).
2.3.1.13.1 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.
2.3.1.14 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 2019)), 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.49 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
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.
2.3.1.14.1 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 2.3.1.1.3.1). This resulted in four study options.
2.3.1.14.2 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
49 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 2.3.1.1), this endpoint focuses on non-fatal
lung cancer incidence.
45
-------
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, Shavlik et al. 2017)).
2.3.1.14.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Gharibvand, Shavlik 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).
2.3.1.15 Alzheimer's Disease
Evidence connecting long-term PM2.5 exposure to nervous system effects led to the 2019 PM ISA
concluding a "likely to be causal" relationship exists ((U.S. EPA 2019)) and various clinically relevant
nervous system endpoints were called out in the biological plausibility section, including Alzheimer's
disease, Parkinson's disease, autism spectrum 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 2019), 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
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
2019), section 8.2.6).
2.3.1.15.1 Available Epidemiologic Literature
One epidemiologic study of Alzheimer's disease met our minimum required identification criteria
(section 2.1.1).
2.3.1.15.2 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Kioumourtzoglou, Schwartz 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
46
-------
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, Schwartz 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.
2.3.1.16 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 2019)) 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 2019), 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 2019), section
8.2.6).
2.3.1.16.1 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.
2.3.1.16.2 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, Jette et al. 2014)).
47
-------
2.3.1.16.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Kioumourtzoglou, Schwartz 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, Schwartz 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.
2.3.2 03
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 ((U.S. EPA 2020)). 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 number decreased to 27 when broad hospital admissions and emergency department
endpoints were identified.50 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
50 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
-------
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.
2.3.2.1 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.
2.3.2.1.1 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 2020)). 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
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" ((U.S. EPA 2020)). By
49
-------
comparison, the 2013 ISA identified this endpoint as "likely to be causal" ((U.S. EPA 2013)). 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.
2.3.2.1.1.1 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 ((U.S. EPA 2020)). 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.
2.3.2.1.1.2 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.
2.3.2.1.1.3 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 a 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 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, Samet 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
50
-------
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.
2.3.2.1.2 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)). 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
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.
2.3.2.1.2.1 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 ((U.S. EPA 2020)).51 All four studies evaluated
51The 2020 03 ISA identified five North American studies of long-term 03-attributable respiratory mortality, but as
Weichenthal, S., D. L Crouse, L. Pinault, K. Godri-Pollitt, E. Lavigne, G. Evans, A. van Donkelaar, R. V. Martin and R.
51
-------
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.
2.3.2.1.2.2 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.
2.3.2.1.2.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Turner, Jerrett et al. 2016) 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, Jerrett 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, Burnett et al. 2009)), but differs in many aspects
including study size, included study locations, and exposure estimation technique.
2.3.2.2 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
admissions and emergency department visits for combined respiratory diseases" ((U.S. EPA 2020),
section 3.1.8).
T. Burnett (2016). "Oxidative burden of fine particulate air pollution and risk of cause-specific mortality in the
Canadian Census Health and Environment Cohort (CanCHEC)." Environmental Research 146: 92-99. 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).
52
-------
2.3.2.2.1 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
2020)).
2.3.2.2.2 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.
2.3.2.2.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Katsouyanni, Samet 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.
2.3.2.3 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 2020), section 3.1.8).
2.3.2.3.1 Available Epidemiologic Literature
Seven U.S.-based studies of respiratory emergency department visits were identified in the 2020 03 ISA
((U.S. EPA 2020)). 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.
2.3.2.3.2 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.
53
-------
2.3.2.3.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Barry, Klein et al. 2019) 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, Klein
et al. 2019) 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.
2.3.2.4 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 2020),
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 2020), section 3.2.6).
2.3.2.4.1 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 2020)).
2.3.2.4.2 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.
2.3.2.4.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Tetreault, Doucet 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 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, Doucet et al.
2016) used Cox proportional hazard models to observe associations between long-term 03 exposure and
54
-------
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 ((Sparrow,
O'Connor et al. 1991, Guerra, Wright et al. 2004, Ochs, Nyengaard et al. 2004, Baena-Cagnani, Rossi et
al. 2007, Trivedi and Denton 2019)), we apply the 4-12 year age-stratified effect estimate from
(Tetreault, Doucet et al. 2016) to children ages 4-17.
2.3.2.5 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 2020), section 3.1.5.7).
2.3.2.5.1 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 2020)). 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).
2.3.2.5.2 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).
2.3.2.5.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Lewis, Robins 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,
Robins 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 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, Robins 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,
55
-------
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).52
2.3.2.6 Minor Restricted Activity Days
No new epidemiologic studies of minor restricted activity days (MRADs) were identified in the 2020 03
ISA ((U.S. EPA 2020)).
2.3.2.6.1 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.
2.3.2.7 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 2020), 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.
2.3.2.7.1 Available Epidemiologic Literature
The 2020 03 ISA identified one epidemiologic study of long-term 03 exposure and allergic rhinitis.
2.3.2.7.2 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Parker, Akinbami 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, Akinbami et al. 2009) placed all study
participants reporting symptoms of respiratory allergies or hay fever into a combined rhinitis group.
(Parker, Akinbami 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 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.
2.3.2.8 School Loss Days
No new studies of work loss days (WLDs) were identified in the 2020 03 ISA ((U.S. EPA 2020)).
52 Estimates were obtained from figures. Authors did not respond to requests for exact results.
56
-------
2.3.2.8.1 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Gilliland, Berhane 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, Berhane 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).
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
57
-------
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.
2.4 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 0) 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).
58
-------
2,4.1 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
consideration of recent ISA conclusions along with the availability of clinically relevant epidemiologic risk
estimates (Table 8 and Table 9).
2.4.1.1 PM2.5
Table 8. Set of Health Endpoints for Main PM2.5 Benefits Assessments
Endpoint Group
Endpoint Type
Endpoint
Exposure
Ages
Mortality
Mortality
All cause
LT
Adults and older adults
(18-99 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
Older adults (65-99
years)
Stroke3
LT
Older adults (65-99
years)
Cardiac Arrest3
ST
Children, 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.
59
-------
2.4.1.2 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.
2.4.2 Risk Estimates
This section presents the risk estimates identified for the main PM2.5 (section 2.4.2.1) and 03 (section
2.4.2.2) benefits assessments. These lists reflect the application of the available epidemiologic literature
(section 2.2) to the identification criteria (section 2.1).
2.4.2.1 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
(Wu, Braun et al.
2020)
Older adults
(65-99 years)
LT
|B = 0.0064 (0.0003)
(Pope III, Lefler et al.
2019)
Adults (18-99
years)
LT
|B = 0.0113 (0.0016)
(Turner, Jerrett et al.
2016)
Adults (30-99
years)
LT
|B = 0.0058 (0.0010)
(Woodruff, Darrow et
al. 2008)
Infants (1-12
months)
LT
P = 0.0056 (0.00454)
Hospital
Admissions,
Cardiovascular
(Bell, Son et al. 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
(Ostro, Malig et al.
2016)—ICD 390-459
Children,
adults, and
ST
P = 0.00061 (0.00042)
60
-------
Endpoint
Study Information
Ages
Exposure
Duration
Beta Coefficient
(SE)1
Visits,
Cardiovascular
older adults
(0-99 years)
Acute
Myocardial
Infarction
(Wei, Wang et al.
2019)
Adults and
older adults
(18-99 years)
ST
13 = 0.02412(0.00928)
Cardiac Arrest
(Ensor, Raun et al.
2013)
(Rosenthal, Carney et
al. 2008)
(Silverman, Ito 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, Coull et al.
2012)—ICD 430-436
Older adults
(65-99 years)
LT
13 = 0.00343 (0.00127)
Hospital
Admissions,
Respiratory
(Bell, Son et al.
2015)—ICD 490-492,
464-466, 480-487,
493
Older adults
(65-99 years)
ST
13 = 0.00025 (0.00012)
(Ostro, Roth et al.
2009)—ICD 460-519
Children (0-18
years)
ST
13 = 0.00275 (0.00077)
Emergency
Department
Visits,
Respiratory
(Krall, Anderson 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
P = 0.00055 (0.00027) (GA)
13 = 0.00097 (0.00035) (AL)
13 = 0.00083 (0.00033) (MO)
(3 = 0.00135 (0.00059) (TX)
Asthma Onset
(Tetreault, Doucet et
al. 2016)
Children (0-17
years)
LT
(3 = 0.04367 (0.00088)
Allergic Rhinitis
(Parker, Akinbami et
al. 2009)
Children (3-
17)
LT
13 = 0.02546 (0.00962)
Lung Cancer
(Gharibvand, Shavlik
et al. 2017)
Adults and
older adults
(>29 years)
LT
13 = 0.03784(0.01312)
Alzheimer's
Disease
(Kioumourtzoglou,
Schwartz et al.
2016)—ICD 331.0
Older adults
(>64 years)
LT
13 = 0.13976 (0.01775)
Parkinson's
Disease
(Kioumourtzoglou,
Schwartz et al.
2016)—ICD 332
Older adults
(>64 years)
LT
13 = 0.07696 (0.01891)
Asthma
Symptoms
(Rabinovitch, Strand
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)
61
-------
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.
62
-------
2.4.2.2 03
Table 11. Set of Risk Estimates for Main 03 Benefits Assessments
63
-------
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,
Samet et al.
2009)
Children,
adults, and
older adults
(0-99 years)
ST; April-September;
MDA1
|B = 0.00073 (0.00057)
(warm season)
(Turner, Jerrett
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,
Samet 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, Klein 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,
Doucet et al.
2016)
Children (0-
17 years)
LT; June-August;
MDA8
P = 0.02075 (0.00146)
(warm season)
Asthma
Symptoms
(Lewis, Robins
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,
Akinbami 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)
64
-------
School Loss Days
(Gilliland,
Berhane et al.
2001)
Children (5-
17 years)
ST; January-June;
DA8
|B = 0.0078 (0.0044)
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.
65
-------
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.53 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.54
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.
53 CDC WONDER mortality data; https://www.cdc.gov/nchs/fastats/deaths.htm.
54 Turner, M. C., M. Jerrett, A. Pope, III, D. Krewski, S. M. Gapstur, W. R. Diver, B. S. Beckerman, J. D. Marshall, J. Su,
D. L Crouse and R. T. Burnett (2016). "Long-term ozone exposure and mortality in a large prospective study."
American Journal of Respiratory and Critical Care Medicine 193(10): 1134-1142. and Tetreault, L. F., M. Doucet, P.
Gamache, M. Fournier, A. Brand, T. Kosatsky and A. Smargiassi (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. 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.
66
-------
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/cou nty-,
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, Lipsett et al.
2001)
(Rabinovitch, Strand et
al. 2006)
Asthma Onset
Annual incidence
0-4
5 -11
12 -17
0.0234
0.0111
0.0044
(Winer, Qin 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
67
-------
Endpoint
Parameter
Rates
Value
Source
Allergic Rhinitis
Respondents aged 3-17
experiencing allergic
rhinitis/hay fever
symptoms within the year
prior to the survey
0.192
(Parker, Akinbami et
al. 2009)
Cardiac Arrest
Daily nonfatal incidence
rates
(Rosenthal, Carney et
al. 2008, Silverman, Ito
0 -17
0.00000002
et al. 2010, Ensor,
18-39
0.00000009
Raun et al. 2013)
40-64
0.00000056
65-99
0.00000133
Lung Cancer
Annual nonfatal incidence
(SEER 2015)and
25-34
0.000001746
(Gharibvand, Shavlik et
35-44
0.000014919
al. 2017)
45-54
0.000067463
55-64
0.000208053
65-74
0.000052370
75-84
0.000576950
95-99
0.000557130
Stroke
Annual nonfatal incidence
in ages 65-99
0.00446
(Kloog, Coull et al.
2012)
Work Loss Days
Daily incidence rate per
person(18-64)
(Adams, Hendershot et
al. 1999), Table 41;
Aged 18-24
0.00540
(U.S. Census Bureau
Aged 25-44
0.00678
2000)
Aged 45-64
0.00492
School Loss Days
Rate per person per year,
assuming 180 school days
per year
9.9
(Adams, Hendershot et
al. 1999), Table 47
Minor Restricted-
Daily MRAD incidence
0.02137
(Ostro and Rothschild
Activity Days
rate 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, Jerrett 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.
68
-------
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.55
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:
= Du,fc(2012)+Dufc(2013)+Dufc(2014)
i,j,k Pl,k (2012) +Puk (2013) +Puk (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)).
55 http://wonder.cdc.gov
69
-------
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.
70
-------
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.56 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
56 All-cause mortality projections are applied to each cause-specific mortality rate.
71
-------
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- and Ethnicity-Stratified Incidence Rates
To estimate race-stratified and age-stratified all-cause and respiratory mortality incidence rates at the
county level, data from 2007 to 2016 was downloaded from the CDC WONDER mortality database with
and without the 'Hispanic Origin' disaggregation for two age groups: >25 and <25.57 The average race
rates were calculated for the 10-year timespan before the death/incidence rate for race combinations
were back calculated. Correction ratios were calculated for the Hispanic and non-Hispanic rates of each
region/race (e.g., (Black - Hispanic) / (Black - All) = Hispanic correction factor and (Black - non-Hispanic) /
(Black - All) = non-Hispanic correction factor). The age-group Hispanic and Non-Hispanic correction
factors were then applied to the existing, county-level race-stratified baseline incidence rates for all-
cause mortality in BenMAP-CE.
3.1.3.1 Race-Stratified Incidence Rates
To estimate race-stratified and age-stratified all-cause and respiratory mortality incidence rates at the
county level, we downloaded 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. To address the frequent county-level data suppression for race-specific death counts, we
stratified the county-level data into two broad race categories, White and Non-White populations. In a
later step, we stratified the non-White incidence rates by race (Black, Asian, American Indian) using the
relative magnitudes of incidence values by race at the regional level, described in more detail below.
We followed the methods outlined in Section D.l.l of the BenMAP User Manual with one notable
difference in methodology; we included an intermediate spatial scale between county and state for
imputation purposes ((U.S. EPA 2018)). We designated urban and rural counties within each state using
CDC WONDER and, where possible, imputed missing data using the state-urban and state-rural
57 http://wonder.cdc.gov
58 http://wonder.cdc.gov
72
-------
classifications before relying on broader statewide data. We followed methods for dealing with
suppressed and unreliable data at each spatial scale as described in Section D.l.l.
A pooled non-White incidence rate masks important differences in mortality risks by race. To estimate
county-level mortality rates by individual race (Black, Asian, American Indian), we applied regional race-
specific incidence relationships to the county-level pooled non-White incidence rates. We calculated a
weighted average of race-specific incidence rates using regional incidence rates for each
region/age/race group normalized to one reference population (the Asian race group) and county
population proportions based on race-specific county populations from CDC WONDER where available.
In cases of population suppression across two or more races per county, we replaced all three race-
specific population proportions derived from CDC WONDER with population proportions derived from
2010 Census data in BenMAP-CE (e.g., 50 percent Black, 30 percent Asian, 20 percent American Indian).
3.1.3.2 Ethnicity-Stratified Incidence Rates
To estimate ethnicity-stratified and age-stratified all-cause and respiratory mortality incidence rates at
the county level, we downloaded data from 2007 to 2016 from the CDC WONDER mortality database.59
Ethnicity-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 county-level data by Hispanic origin (Hispanic and non-Hispanic). We
followed the methods outlined in Section D.l.l to deal with suppressed and unreliable data ((U.S. EPA
2018)). We also included an intermediate spatial scale between county and state designating urban and
rural counties for imputation purposes, described in detail in Section D.1.3 of the BenMAP User Manual.
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,
do not provide data to HCUP. For these states, regional statistics from HCUPnet60 were used to estimate
baseline hospitalization rates.
59 https://wonder.cdc.gov/
60 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.
73
-------
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.61 For each ICD code combination, unique baseline incidence rates are developed.
In this TSD update, adjustments were made to three existing hospital admission sensitivity endpoints:
Myocardial Infarction (ICD Codes 410.X0 and 410.X1)
61 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
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.
74
-------
Using data from the Healthcare Cost and Utilization Project (HCUPnet), Hospital Inpatient
National Statistics, we downloaded the 2014 national estimates of discharge rates per 100,000
people for the set of ICD-9 codes associated with myocardial infarction in BenMAP (410) and the
set of ICD-9 codes associated with myocardial infarction in (Danesh Yazdi, Wang et al. 2021)
(410.X0, 410.X1). We applied the ratio of (national discharge ratesiCD-9 = 4io.xo,4io.xi: national
discharge ratesiCD-9 = 4io), 0.995, to the existing, county-level baseline incidence rates for Acute
Myocardial Infarction, Nonfatal in BenMAP.
Ischemic Stroke (ICD Codes 433.X1, 434.X1, and 436)
Using data from the Healthcare Cost and Utilization Project (HCUPnet), Hospital Inpatient
National Statistics, we downloaded the 2014 national estimates of discharge rates per 100,000
people for the set of ICD-9 codes associated with stroke in BenMAP (431-437) and the set of
ICD-9 codes associated with ischemic stroke in (Danesh Yazdi, Wang et al. 2021) (433.XI, 434.X1,
436). We applied the ratio of (national discharge ratesiCD-9 = 433.xi,434.xi,436: national discharge
ratesicD-9 = 431-437), 0.564, to the existing, county-level baseline incidence rates for HA, Stroke in
BenMAP.
Atrial Fibrillation and Flutter (ICD Codes 427.3)
Using data from the Healthcare Cost and Utilization Project (HCUPnet), Hospital Inpatient
National Statistics, we downloaded the 2014 national estimates of discharge rates per 100,000
people for the set of ICD-9 codes associated with dysrhythmia in BenMAP (427) and the set of
ICD-9 codes associated with atrial fibrillation and flutter in (Danesh Yazdi, Wang et al. 2021)
(427.3). We applied the ratio of (national discharge ratesiCD-9 = 427.3: national discharge ratesiCD-9 =
427), 0.666, to the existing, county-level baseline incidence rates for HA, Dysrhythmia in BenMAP.
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).
75
-------
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, Dockery et al. 2001) and supplemented with evidence
found in more recent single and multi-city studies ((Sullivan, Sheppard et al. 2005, Pope III, Muhlestein
et al. 2006)). 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, Go 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
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, Broda et al.
1999).
76
-------
3,4.2 Asthma Onset and Symptoms
3.4.2.1 Asthma Onset
Baseline incidence rates for new asthma onset are estimated from (Winer, Qin et al. 2012). (Winer, Qin
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).62 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.
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, Doucet et al. 2016)
PM2.5
0.0380
5 -17
(Tetreault, Doucet et al. 2016)
PM2.5
0.0893
0 -17
(Tetreault, Doucet et al. 2016)
03
0.0750
Asthma symptoms, albuterol
use
6 -13
(Rabinovitch, Strand 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.
62 https://www.cdc.gov/asthma/nhis/2018/data.htm and
https://www.cdc.gov/asthma/most_recent_national_asthma_data.htm
77
-------
3.4.2.2 Albuterol Use
We develop incidence rates for albuterol use from the rates presented in (Rabinovitch, Strand 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, Strand 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, Robins 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, Robins 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. '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, Akinbami 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, Akinbami
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 (SEER 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 ((SEER 2015)). Table 18 presents the
78
-------
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.
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) ((NCES 1996, Adams, Hendershot et al.
1999), 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, Jennison et
al. 2000) and (Ransom and Pope 1992), which ranged from 4.5% to 5.1%.
79
-------
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, Hendershot 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, Hendershot 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, Berhane 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, Hendershot 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, Hendershot et al. 1999), Table 41). They reported 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 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.
3.4.8 Hypertension
The state-level baseline incidence rates for hypertension among women ages 50-99 were calculated
using the following relationship between incidence and prevalence:
Prevalence
= Incidence * Average Duration of Disease.
1 -Prevalence °
This relationship assumes that hypertension prevalence is at a steady state in the population of interest.
Data from the National Health and Nutrition Examination Survey supports this assumption, finding no
significant change in age-adjusted hypertension prevalence among U.S. women from 1999-2000 to
2017-2018.
80
-------
2011-
2012
2013-
2014
2015- 2017-
2016 2018
Survey period
'Significant quadratic trend from 1999 through 2018.
NOTES: Hypertension Is defined as systolic blood pressure greater than or equal to 130 mmHg or diastolic blood pressure grealer than or equal lo 80 mmHg, of
currently taking medication to lower blood pressure. All estimates are age ad|usted by the direct method to the U.S. Census 2000 population using age groups
18-39,40-59, and 60 and over. Access data table for Figure 4 at: Mips.//www cdc gov/nchs/data/dalabhefs/db364-tables-50B pdfiM
SOURCE: NCHS, National Health and Nutrition Examination Survey, 1999-2018
Figure 11. Age-Adjusted Trend in Hypertension Prevalence Among Adults Aged 18 and Over by Sex in the
U.S. Between 1999-2018
We calculated state-level hypertension prevalence rates using data from the Behavioral Risk Factor
Surveillance System (BRFSS), a system of health-related telephone surveys administered by the CDC that
collects data from all fifty states. We calculated national-level duration of disease using data from the
CDC, the Canadian Chronic Disease Surveillance System (CCDSS), and Leukine et al. (2010). In this
calculation, we assumed that hypertension, once diagnosed, lasts until death (even if controlled by
medication or diet). The definition of hypertension employed by BRFSS (see below) supports this
assumption.
The procedure for calculating state-level hypertension prevalence rates is summarized as follows:
• We obtained individual-level records from the 2019 BRFSS Data file for all states except New
Jersey, which was not included in the 2019 data file.
• For consistency with the relevant C-R function, we filtered the individual-level 2019 BRFSS
dataset to include responses from female participants aged 50 and older.
81
-------
• We aggregated the resulting sample by state, using raking-derived weights63 assigned to each
respondent by BRFSS. State-level prevalence was calculated as the sum of weights associated
with respondents who responded "Yes" to the question, "Have you ever been told by a doctor,
nurse, or other health professional that you have high blood pressure?" divided by the sum of
weights of all respondents in the state.
• For New Jersey, we repeated the above steps using the 2017 BRFSS Data File, the most recent
year available for the state.
The procedure for calculating the national-level duration of disease is summarized as follows:
• For each five-year age group (50-54 years, 55-59 years, 60-64 years, 65-69 years, 70-74 years,
75-79 years, and 80+ years), we estimated duration of disease using the following equation:
Duration of disease — LE aii u.s. females — (LE All Canadian females LE Canadian females with hypertension)
• We obtained the life expectancy (LE) values from the following sources:
o We obtained LE An u.s. females from CDC Vital Statistics Rapid Release and applied the life
expectancy of the lower bound of each age group (i.e., we applied the life expectancy at
age 50 to the 50-54 years age group).
O We calculated the LE All Canadian females by taking the weighted average of LE Canadian females with
hypertension and LE Canadian females without hypertension from Loukine et al. (2010).
¦ Weights were derived from hypertension prevalence estimates among Canadian
females (aged 50-64, 65-79, and 80+) from CCDSS.
¦ We are not aware of any U.S. studies that estimate the change in life expectancy
associated with hypertension at 5-year intervals.
o We obtained LE Canadian females with hypertension from Loukine et al. (2010).
o In all three instances, LE for the 80+ age group was calculated as the average of LE for
the 80-84 and 85+ age groups for consistency with age groups provided by BRFSS.
• We calculated the overall duration of disease for females ages 50-99 by taking the weighted
average of the duration of disease values for the five-year age groups.
o Weights were calculated from the raking-derived weights provided by BRFSS as the
fraction of the female hypertensive population aged 50-99 that falls within the given age
class.
63 "Raking," or iterative proportional fitting, is the process that BRFSS uses to adjust for demographic differences
between the sample surveyed and the U.S. population. Through this process, BRFSS assigns a final, "raking-derived
weight" to each respondent that can be used to generate a nationally-representative dataset.
82
-------
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, Mulder 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- and tract-level
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.
Population projections stratified by race/ethnicity, age, and sex are based on economic forecasting
models developed by Woods and Poole ((Woods & Poole 2015)). The Woods and Poole database
contains county-level projections of population by age, sex, and race out to 2050, relative to a baseline
using the 2010 Census data. Population projections for each county are determined simultaneously with
every other county in the U.S to consider patterns of economic growth and migration. County-level
estimates of population percentages within the poverty status and educational attainment groups were
derived from 2015-2019 5-year average ACS estimates. Additional information can be found in Appendix
J of the BenMAP-CE User's Manual (https://www.epa.gov/benmap/benmap-ce-manual-and-
appendices).
83
-------
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.64 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 four primary components of the value to society of an individual's avoidance of a non-fatal
illness: 1) medical costs, 2) lost productivity, 3) cost of caregiving, and 4) impacts on quality of life (i.e.,
"pain and suffering"). Some health injuries may incur other costs. Estimates of individual WTP are
conventionally thought to reflect the total cost to society and are the preferred welfare valuation
measure.65 However, WTP values are available for a very limited subset of health endpoints, such as
mortality.66
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,
Blomquist 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, medications, the cost of caregiving, and the cost of subsequent health
injuries due to the stroke (sequella).
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
64 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).
65 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).
66 Economic theory also argues that WTP for most goods (such as environmental protection) will increase if real
income increases.
84
-------
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
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).
b(Onukwugha, McRae 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
85
-------
provided below. When possible, costs are seperated into components which can be separately adjusted
for growth in healthcare costs, income, and other costs. Costs occuring over multiple years are
discounted with a constant 2% discount rate.
86
-------
Table 21. Unit Values for Economic Valuation of Health Endpoints (2015$)1
Health Endpoint
Type
Central Estimate of
Healthcare
Labor
Care
Source
Value Per Statistical
Costs
Costs
and
Incidence (2015$)
Other
Costs
Mortality
Value of Statistical
Life (VSL)
Undiscounted:
$8,705,114
2:
$8,132,666
Weibull distribution fitted to 26
published VSL estimates (5
contingent valuation and 21 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 costs
Varies by ICD codes,
HCUP data (details available in
Department Visits
ranging between
$600 and $1,200
section 3.3)
Nonfatal
3-year medical
$49,108
$49,108
Multiple
(O'Sullivan, Rubin et al. 2011),
Myocardial
costs (excluding ED
of annual
Cropper and Krupnick, 1990
Infarction (AMI)3
and
hospitalization)
plus lost labor
earnings
Asthma Symptom-
Medical costs
$0.35 per albuterol
Average prescription costs derived
Albuterol Usec
inhaler puff
from Epocrates.com and Goodrx.com
accessed March 19, 2020
87
-------
Asthma Symptom-
WTP for 1
$219
$219
(Dickie and Messman 2004)
Chest Tightness,
symptom day
Cough, Shortness of
Breath, or Wheeze
Asthma Onsetc
Lifetime medical
costs and lost
productivity
$182,681
$93,820
$85,800
$3,136
(Maniloff and Fann 2023)
Allergic Rhinitis0
1-year medical
costs
$600
$600
(Soni 2008)
Cardiac Arrestc
3-year medical
costs
3%: $36,000
7%: $35,000
(O'Sullivan, Rubin et al. 2011)
Lung Cancerc
5-year medical
costs
$34,000
(Kaye, Min et al. 2018)
Strokec
Lifetime medical
costs, care costs,
and lost
productivity
$159,067
$57,330
$90,039
$12,044
(Maniloff and Fann 2023)
Work Loss Days
Median daily wage
U.S. mean: $298
$298
EPA calculation(IEc 1993)
School Loss Days
Lost productivity
of parent and lost
learning
$1186
$1186
(Liu, Lee et al. 2021), (US Bureau of
Labor Statistics 2021)Z_(US Bureau of
Labor Statistics 2021), (Chetty,
Friedman et al. 2014), (U.S. Census
Bureau 2010), (U.S. EPA 2020), (U.S.
Census Bureau 2021)
Minor Restricted-
Median WTP
$70
$70
(lEc 1993)
Activity Days
88
-------
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 $8.7 million (2015$). 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 2011)). 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 2011)). For mortality estimates using the 20-year lagged mortality, the present
value of the lagged mortality is $8.1 million (2015$) when discounting at 2%.
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 $12.3 million range." Five of the 26 studies are contingent valuation studies, which directly solicit
WTP information from subjects; the rest are wage-risk studies, which base WTP estimates on estimates
of the additional compensation demanded in the labor market for riskier jobs. Because this VSL-based
unit value does not distinguish among people based on the age at their death or the quality of their
lives, it can be applied to all deaths. Table 22 presents the central unit value from the 26 value of
statistical life studies and their underlying distribution.
Table 22. Central Unit Value for VSL based on 26-value-of-life studies
Basis for Estimate
Age Range at
Death
Unit Value
(VSL) (2015$)
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
89
-------
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.67 We then identify all discharges in the HCUP datasets with ICD-
10 codes that match to a study's ICD-9 code(s).68 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
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.
67 General Equivalence Mapping Files, FY 2016 release of ICD-10-CM. https://www.cdc.gov/nchs/icd/icdlOcm.htm.
68 For emergency hospitalizations, we further restrict the sample to (1) hospitalizations designated as "emergency"
and (2) emergency department visits that result in hospitalization.
90
-------
Table 24. Medical Costs and Hospital Stay Data
Endpoint
Studies
ICD Codes
Age Range
Mean
Hospital
Charge
(2015$)
Mean
Length
of Stay
(days)
Total Cost
of Illness
(Unit
Value in
2015$)*
Start
End
HA, All Cardiac
Outcomes
(Talbott, Rager et
al. 2014)
390-459
0
99
$16,045
5.05
$16,918
HA, All
Respiratory
(Ostro, Roth et al.
2009)
460-519
0
18
$9,075
3.49
$9,678
HA,
Alzheimer's
Disease
(Kioumourtzoglou,
Schwartz et al.
2016)
331.0**
65
99
$10,696
7.95
$12,070
HA, Cardio-,
Cerebro- and
Peripheral
Vascular
Disease
(Bell, Son et al.
2015)
410- 414,
429, 426-
427, 428,
430-438,
440-449
65
99
$14,665
4.82
$15,498
HA,
Respiratory-1
(Jones, Diez-Roux et
al. 2015)
491, 492,
493, 496
0
99
$7,676
3.86
$8,343
HA,
Respiratory-2
(Bell, Son et al.
2015)
464-466,
480-487,
490-492,
493
65
99
$9,003
4.66
$9,808
HA, Ischemic
Stroke (ICD9
433.XI, 434.XI,
436)
(Danesh Yazdi,
Wang et al. 2021)
433.XI,
434.XI,
436**
65
99
$12,212
4.29
$12,954
HA, Atrial
Fibrillation and
Flutter (ICD9
427.3)
(Danesh Yazdi,
Wang et al. 2021)
427.3**
65
99
$10,656
3.66
$11,288
HA, Parkinson's
Disease
(Kioumourtzoglou,
Schwartz et al.
2016)
332
18
99
$12,190
3.83
$12,852
* The opportunity cost of a day spent in the hospital was estimated, for the above exhibit, at the median daily
wage of all workers, regardless of age. The median daily wage was calculated by dividing the median weekly
wage ($864 in 2015$) by 5. The median weekly wages for 2015 were obtained from the U.S. Census Bureau's
2015 American Community Survey, "Selected Economic Characteristics: 2015 American Community Survey 1-
Year Estimates."
**Although the health impact function and incidence data were developed using the 331.0 ICD-9 code for
Alzheimer's Disease, the COI was derived using the three-digit ICD-9 code of 331. Therefore, the COI is
representative of all health endpoints captured by the 331 ICD-9 code.
91
-------
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, Rubin 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 $49,108 using a 2% discount
rate. (Table 25). Although this study analyzed costs associated with individuals ages 35 and older, we
apply the total medical costs to all ages from zero to 99 since only a small portion (<10%) of annual AMI
incidence occurs in the age range below 35.
Table 25. Medical Costs for AMIs (2015$)
Costs
Cumulative Costs
Annual Costs
Undiscounted
2% Discount Rate
Month of Event*
$43,523
$43,523
$43,523
Year 1
$70,629
$27,106
$27,106
Year 2
$82,591
$11,962
$11,727
Year 3
$93,281
$10,690
$10,275
Years 1-3
$93,281
$49,758
$49,108
We supplement AMI medical costs with estimates o
lost earnings using age-specific estimates from
(Cropper and Krupnick 1990). Using a 2% discount rate, we estimated the following present discounted
values in lost earnings over 5 years due to a heart attack: 0.229 times annual earnings for someone
between the ages of 25 and 44, 0.371 times annual earnings for someone between the ages of 45 and
54, and 1.306 times annual earnings for someone between the ages of 55 and 65. (Cropper and Krupnick
1990) does not provide lost earnings estimates for populations under 25 or over 65. As such we do not
include lost earnings in the cost estimates for these age groups. These costs, along with the total
valuation estimates for AMIs, are presented in Table 26.
92
-------
Table 26. Total Valuation Estimates for AMIs (2015$)
Age F
Min
lange
Max
Medical Cost
Lost Earnings Multiplier
Total Cost
0
24
$49,108
0
$49,108
25
44
$49,108
0.219
$49,108+ 0.219*earnings
45
54
$49,108
3.534
$49,108 + 3.534*earnings
55
65
$49,108
1.245
$49,108 + 1.245*earnings
66
99
$49,108
0
$49,108
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
Maniloff and Fann (2023) estimated the lifetime costs of asthma using data from the 2017 to 2019
Medical Expenditure Panel Survey (MEPS). The paper identifies individuals with asthma based on
individual surveys. Average annual health care costs, lost wages, and caregiving costs are estimated
using a regression approach. These were estimated to be $3267, 2988, and 109, respectively, for a total
cost of $6361. The present value is calculated by summing the discounted average annual costs over the
average lifespan. Health care costs are updated based on the health care cost index, while wage and
caregiving costs are updated based on wage indices. The MEPS primarily includes respondents 18 years
or older.
5.3.4 Asthma Symptoms/Exacerbation
5.3.4.1 Albuterol Use
As albuterol use is a new measure of PM2.5-attributable asthma symptoms, we developed a method for
valuing this health endpoint. We estimate the economic value for albuterol use associated with asthma
symptoms using prescription prices for albuterol inhalers. Epocrates and GoodRx provide cost and
actuation information for four common types of albuterol inhalers in 2020 dollars.70,71 Both online
70 https://online.epocrates.com/drugs searched March 19th, 2020.
71 https://www.goodrx.com/albuterol searched March 19th, 2020.
93
-------
resources utilize published price lists, purchases, claim records, and pharmaceutical data to provide
clinical statistics. Epocrates and the FDA provide cost and actuation information for one additional, less
common, albuterol inhaler.72 We divide the cost of inhalers by the actuations per inhaler to calculate an
average cost per actuation across all inhaler types. We then adjust the values to 2015$ 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.
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, Rubin 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 $35,880 using
a 2% discount rate (Table 27 and Table 21).
72 https://www.accessdata.fda.gov/drugsatfda_docs/label/2016/205636s006lbl.pdf
94
-------
Table 27. Valuation Estimate for Cardiac Arrests (2015$)
Costs
Cumulative Costs
Annual Costs
Undiscounted
2% Discount Rate
Month of Event*
$43,904
$43,904
$43,904
Year 1
$71,901
$27,997
$27,997
Year 2
$74,701
$2,800
$2,745
Year 3
$80,046
$5,345
$5,137
Years 1-3
$80,046
$36,142
$35,880
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, Min 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., $3,269 for males). We calculate costs over a five-year span,
beginning with the initial onset which is occurs with a delay after exposure. The specific lag periods
between exposure and onset are discussed in Section 6.4.2. The initial year's treatment cost is summed
with four years of continuing annual costs discounted at 2%.
Furthermore, (Kaye, Min et al. 2018) provides separate treatment cost estimates for men and women.
The distribution of new lung cancer cases by sex in the United States from (Siegel, Miller et al. 2019) is
approximately 51% male and 49% female. This distribution of new lung cancer cases was used to weight
the sex-specific cost estimates from (Kaye, Min et al. 2018) to obtain a combined five-year cost estimate
for both sexes. In order to adjust the cost estimate to 2015$ using a medical cost index, we assume that
costs presented by (Kaye, Min 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 $34,155 using a 2% 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 28. 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 28. Latency Periods Used in Lung Cancer Risk Assessment Papers
95
-------
Study
Latency Period (years)
Location
(Gogna, Narain et al. 2019)
5
Canada
(Bai, Shin et al. 2020)
4; 10
Canada
(Kulhanova, Morelli et al. 2018)
10
France
(Coleman, Burnett et al. 2020)
10; 15
US
(Harrison, Smith 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 29 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 occur 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.
Table 29. 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.
96
-------
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 ((IEc 1993)).73 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
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 IEc 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 IEc
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 IEc estimate of $69.58 (2015$) (Table 21).
5.3.8 School Loss Days
We include two costs of school loss days: caregiver costs and loss of learning. We calculate each
separately and then sum them. Caregiver costs are valued at their employers' average cost for employed
caregivers. For unemployed caregivers, the opportunity cost of their time is calculated as the average
take-home pay. The loss of learning is calculated based on the impact of absences on learning multiplied
73 IEc (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. derived this estimate of WTP to
avoid a MRRAD using WTP estimates from Tolley, G. S., L. Babcock, M. Berger, A. Bilotti, G. Blomquist,
M. Brien, R. Fabian, G. Fishelson, C. Kahn and A. Kelly (1986). Valuation of reductions in human health
symptoms and risks. Prepared for US Environmental Protection Agency, U.S EPA. for avoiding a three-
symptom combination of coughing, throat congestion, and sinusitis. This estimate of WTP to avoid a
MRRAD, so defined, is $69.58 in 2015$.
97
-------
by the impact of school learning on adult earnings. The loss of learning estimate is currently only
available for middle and high school students. The two costs are summed.
The caregiver costs assumes that an adult caregiver stays home with the child and loses any wage
income they would have earned that day. For working caregivers, we follow U.S. EPA guidance and value
their time at the average wage including fringe benefits and overhead costs (U.S. EPA, 2020a). We apply
a multiplier of 1.46 for fringe benefits based on the U.S. Bureau of Labor Statistics' 2023 National
Compensation Survey and 1.2 for overhead from U.S. EPA's National Center for Environmental
Economics (NCEE) to county-specific mean wage estimates from the 2021 5-year American Community
Survey (ACS). For nonworking caregivers, we assume that the opportunity cost of time is the average
after tax earnings. We estimate the income tax rate for a median household to be 7% and apply to 2021
ACS county-specific mean wage estimates. The income tax rate of 7% is the percentage difference in
median post-tax income and median income from (U.S. Census Bureau, 2021). We then apply county-
specific employment-population ratios from the 2021 ACS to yield county-specific estimates of caregiver
costs per school loss day. The average of this county-specific measure across all counties is
approximately $235 (2015$).
To measure the loss of learning, we update the Liu et al., 2021 estimate of the impact of a school
absence on learning as measured by an end-of-course test score. Liu et al., 2021 provide an estimate
that a school absence leads to a $1,200 reduction in lifetime earnings, which is based on the Chetty et
al., 2014 estimate of mean lifetime earnings ($522,000 in 2010$). We first use 2015 Current Population
Survey data from the U.S. Census to calculate expected lifetime earnings of $1,137,732 (discounting at
2%). We then multiply the Liu et al., 2021 estimate of $1,200 by (1,137,732/522,000) and convert from
2010 dollars to 2015 dollars based on the Consumer Price Index for All Urban Consumers. This approach
yields an estimated learning loss of $2,843 per school absence (discounted at 2%).
For BenMAP application, we create a valuation function that uses caregiver costs for preschool and
elementary school children and the sum of caregiver costs and loss of learning for middle school and
high school students. We calculate that 31% of children under 18 are middle school and high school ages
13-18, distributed equally across the two bins); thus, we estimate county-specific averages of the
combined effect by summing the county-specific estimates of caregiver costs and ($2,843*0.31) with 2%
discounting. The average of this county-specific measure across all counties is approximately $1,116
(discounted at 2%) per school day for ages 0 to 17 (2015$).
A unit value based on the approach described above is likely to understate the value of a school loss day
in at least four ways:
1) It omits WTP to avoid the symptoms/illness which resulted in the school absence
2) The approach omits other aspects of school attendance such as social and emotional
development or meals
3) It does not account for deleterious effects on student learning which are not captured by a
single end-of-grade test score. These other effects might include lost learning in other subjects
such as science, language arts, and history, as well as social and emotional development.
98
-------
5.3.9 Stroke
Maniloff and Fann (2023) developed COI estimates of direct and delayed health care costs, loss of wage
income, and home paid caregiving costs associated with a non-fatal first stroke. The authors conduct a
regression analysis based on 2017 - 2019 medical expenditure data from the Agency for Healthcare
Research and Quality's (AHRQ's) Medical Expenditure Panel Survey (MEPS) to estimate the difference in
total medical expenditures between individuals who report having been diagnosed with new onset
stroke and otherwise similar individuals who do report a stroke diagnosis. This approach allows for
inclusion of direct medical costs due to the stroke itself and medical costs due subsequent health effects
that may be related to having experienced a stroke (sequelae). The MEPS data includes actual (not
billed) insurer payments, government payments and individual payments, and includes agency and non-
agency paid caregivingl5 in its caregiving cost estimates. The authors report the annual costs and
present value of lifetime costs of a first stroke for those ages 18 and older, broken into three cost
categories: health care, lost wage income, and home caregiving. The authors estimated lifetime costs
using data on expected remaining life expectancy at age of first stroke; the resulting present value
lifetime estimates are the values used in the BenMAP valuation function. The study estimated the
average present value lifetime COI for a new onset stroke to be $159,067 (2015$) using a 2% discount
rate.
5.3.10 Work Loss Days (WLDs)
Work loss days are valued at a day's wage. BenMAP calculates county-specific mean daily wages from
county-specific annual wages by dividing by (52*5), then applies a multiplier of 1.46 for fringe benefits
estimates based on the U.S. Bureau of Labor Statistics' 2023 National Compensation Survey and an
overhead multiplier of 1.2 from U.S. EPA's National Center for Environmental Economics (NCEE) as
described in U.S. EPA (2014) for workers as defined by county-specific employment rate data from the 5-
year 2021 American Community Survey. The average of this county-specific daily wage measure across
all counties is approximately $298 for ages 18 to 65 (2015$) (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
99
-------
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
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 30 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 30. 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 ten-year projections of US Gross Domestic Product (GDP) are used to estimate changes in future
income ((CBO 2016)). 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 30 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:
100
-------
1)
AWTP
~WTP~ (yVTP2 - WTP,) X (J2 + /,)
£=—T7— = —: — ^
(h ~ h) x (WTP2 + WTPJ
I
2) eI2WTP2 + e12WTP1 - s/, WTP2 - eI1WTP1 = I2WTP2 + I1WTP2 - I2WTP1 - 11WTP1
Table 31 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
101
-------
I
13 10
1 Q
2* 11
2" 12
2014
2015
2016
201?
20 to
20 i9
-"*n20
222 j
2024
2025
2028
Income-Based WTP Adjustments by Health Effect and Year
Miner Health Enripoint
Severe Health Endpoint
Mortality
Lovt
Mid
Upper
Low
Miff
Upper
Low
Mid
0 9994 «14
1 00U27d *5
i noon:- -ia
1 9Q1'C9926
1 0i2:^676
1 20^5'=".? '52
1 0>Q48_,3718
1 006 05334
1 307473026
1 ;o8g:,?73j
1 0Q962*5'33
1 00896_274
1 CO9722471
i oi:dj,n:4
1 01i770725
1 0I24306;8
1 Q1275'46
1 012262J44
1 01072917
1 21 1459589
1 011781-222
1 212354255
1 312712717
1 013344'99
1 214022827
t C142749_0
1 014827723
1 013322024
I 015639'96
1 21 390 95.<9
1 019285399
1 31c661671
1 21708*:029
1 01748:324
1 0178T9009
1 0192L33579
0 2'D7dfi7 * t)4
i 00P463
CO'i 35-124
1 0072*3^B12
I 009509491
' 0'3_,6994f
1 3'9235313
' C2^0635t?
* 329 32 5492
t 032765627
I 012733301
1 a>402°05H
' 03':95'3Q92
' 041232561"
I 04486*= 582
' 247425629
' 249671241"
1 04077246
' 241 1074 l<3
I 04 "'GG1 .55
i 244923405
t 047129916
t 349520923
050"t7b 57
C5361^623
354002981
t n5 "761265
' 259697701
• C5;93c809
1 263991403
i CC24G1257
I 0i340 249 25
t 063614223
f067t&933c
f 069730 9C
373252 >39
0 295779,359
C023935-4
1 2062133541
2 4559414
' D*9;Q7d1c«
" 020903 37,.,
1 33660972"'
' 346717Q49
337473063
266622734
' J365h2'7.;
' 369)233.344
' 075297475
0d42012l7
! 091792464
097 54975
299772427
' 29576490.
383943'49
* 389271665
* 091917515
' 096532583
• 0994547b1
10461q7o3
TO 93172
I 1227t07
1 6838455
120945096
• 1235^4972
12581968
126943761
132264641
135679722
139053007
142365217
145622227
1
0 *96431 * 8
1 001744^9
02523102-1
012117267
, <; '=,0:17^7
0224167-1
030580044
038779835
047635954
055212379
f 35516243
0C7U 3937
032 , j6 5y1
O^' I678h64
075912952
090312252
i 38245947
079196704
0294' 4207
0736437*6
076015592
079821798
0821 "*7606
02642a881
0'"09 32 3 -i2
0*26920 K
0'.b4266U6
09979*671
1C " 929545
19 .73 1792
10bi 156 3
IC9035254
f 1 * 150224
114547729
11724662o
119835339
2 993674994
1 00314-939
1 0U94 .5654
1 02 9174c2
1 0287: y 176
1 04071509rt
1 05572361j
1 070935659
1 0974'567i59
1 i0*6j3635
1 0"532698
1 (03690138
1 1151131'9
1 129017531
1 '40922546
' 49.1'0299
1 15^4836r6
1 .47202173
" 126309 - &
1 136984232
1 '*4-119123
1 '48379693
' 1329a 593
1 i>" 142945
1 SIj972539
1 73_:9n:A
1 13053 102-i
1 1970740C5
1 i5*27 544
1 ' 34r239 39
1 '3369T845
1 205213C 66
1 210635345
1 21609G745
1 22 42863 3
1 226682792
1
0 99157W ,
1 00413- * ft -47
0*2"00* 6
I 02^330373
I 059594113
1 034657698
1 074937633
1 095668674
1 (19347049
1 157636993
1 137707233
1 434330-4
1 f Ctb4?6 5^2
1 175h0*i7fl|
1 9241 "95 i
1 2042. 024
1 213017h9 i
1 2012Q62C7
1 i75235033
1 166694939
1 192690849
1 202856193
1 209..183J9
1 223796929
1 233217505
I *.376965; 2
1 24 P1752*" 3
1 22.7459432
1 26 "426423
1 269532872
1 27567 3.,99
1 283306029
1 291138l3i
1 299679623
1 30C55"45G
1 314110994
1
0 9jfi°72r yd
1 OOQc57u99
1 021970 "*"*7
1 0u .°6Ifbfi
1 035359956
1 007 1306
1 00968492
1 01224c339
1 31500969
1 017342448
1 317327U7
1 018005252
1 019539933
1 021781206
1 02 3 e 80 "*25
1 0250 6785
1 0256*66475
1 024"75365
1 0217'5641
1 023.051262
1 0237 2156
1 024°62051
1 025586155
1 026867747
I 22P24379
1 0-87551dd
1 029970947
1 U30P8247O
1 03 1524897
1 032072306
1 032834172
I 3.,364408
1 034466924
1 035281181
1 0 30280539
1 Q"686_i9U4
1
994 375765
022792476
008 3827'17
01*34564'3
02'356.52
036109447
049.81137
0927761*5
077371955
099927657
039745045
0"3089
96->d99Sb
102
-------
6 Characterizing Uncertainty and Evaluating Sensitivity to Alternate
Assumptions
Complex analyses such as the one presented in the Regulatory Impact Analysis for the Proposed
Reconsideration of the PM NAAQS 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 ((National
Resources Council 2004)) 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, respectively. 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 impact the magnitude of bias most strongly.
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
103
-------
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 0).
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
Three studies of all-cause, long-term PM2.5 exposure and mortality were identified as best characterizing
U.S. risk in adults, (Turner, Jerrett et al. 2016, Pope III, Lefler et al. 2019, Wu, Braun et al. 2020).
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
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.
77 As estimates of infant mortality incidence are relatively small, we do not perform quantitative uncertainty
analyses for that health endpoint.
104
-------
pollutant concentrations over time). To address the potential mismatch between projected air quality
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 three long-term
exposure epidemiologic cohort studies examining mortality, ACS CSP-II and Medicare are presented in
Figure 12 ((Turner, Jerrett et al. 2016, Di, Wang et ai. 2017, Wu, Braun et al. 2020)); the distribution of
PM exposure incorporated in the (Pope III, Lefler et al. 2019) study is reflected in the histogram reported
below (Figure 13). 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, Peters et al. 2015)). Points reflect cohort specific PM2.5 concentration data, with
connecting lines estimating missing data.
100%-
90%-
o
1.
g 80%-
Q.
X
LU
t 70%-
0
Jc
° 60%-
©
01
£ 50%-
CD
U
a! 40%-
o
>
u 20%-
10%-
H 30%- _. _
3 / ¦ Medicare (Wu 2019)
i ™ / 1/ ¦ ACS CSP-II
Medicare (Di 2017)
CanCHEC
10 15 20 25 30 35 40
PM2.5 Concentration (ng/mA3)
Figure 12. Cumulative Percentile of PM2.5 Cohort Exposure from the ACS CSP-II, Medicare, and CanCHEC
Cohorts
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.
105
-------
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
provided in Table 32. For comparison, the lowest reported PM2.5 concentrations from previous studies
((Krewski, Jerrett et al. 2009, Lepeule, Laden 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 ((Turner, Jerrett et al. 2016, Di, Wang et al. 2017, Pope III,
Lefler et al. 2019, Wu, Braun et al. 2020)). Pope and co-authors separately reported the exposure
distribution ((Higbee, Lefler et al. 2020)); we report the histogram from this manuscript below.
Table 32. 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
<
O
O
3.2
3.5
3.8
4.1
4.3
4.5
4.7
4.9
5.0
6.1
CanCHEC
0.0
3.2
3.5
3.6
4.0
4.7
A Lowest modeled value is 0.0078 ng/m3
106
-------
Figure 13. Modeled PM2 5 Exposure Distribution for NHIS Study Population, with Select Fitted Probability
Density Functions (from (Higbee, Lefler et al. 2020), publisher permission pending)
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 2022 PM PA ((U.S. EPA 2022)).
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"
((U.S. EPA 2019)).
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
107
-------
using monitor data as the only exposure information have increasing exposure uncertainty the farther
people live from the monitor site.
(Turner, Jerrett et al. 2016) and (Pope, Turner et al. 2015) analyzed the same ACS CSP-II population over
the same time period but used different hybrid exposure estimation techniques. (Turner, Jerrett 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, Turner 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 33, including
the estimate identified for the main benefits assessment in italics.
Table 33. PIVh.s-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" ((U.S. EPA 2019)).
Both (Turner, Jerrett et al. 2016) and (Di, Dai et al. 2017) 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 ((U.S.
EPA 2019)). Differences in the magnitude of risk estimates including or excluding 03 as a copollutant are
provided in Table 34. Italicized risk estimates were identified for use in the main benefits assessment.
Table 34. Single- and Two-Pollutant (Including 03 as a Copollutant) PM2.5-Attributable Mortality Risk
Estimates per 10 ng/m3
(Turner, Jerrett et al. 2016)
(Di, Dai etal. 2017)
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)
6.1.2.4 Statistical Technique
Wu, 2020 reported hazard ratios estimated using causal inference approaches, wherein the authors
"design the study creating a pseudo-population which mimics a randomized experiment..." Wu and co-
authors applied three variants of the Generalized Propensity Score (GPS) technique: (1) matching by
79 Modeling more than two correlated pollutants can be problematic due to collinearity issues.
108
-------
GPS; (2) weighting by GPS; and (3) adjustment by GPS. Each of the three techniques provide similar
Hazard Ratios, providing evidence that the results were robust to confounding (Table 35)
Table 35. Hazard Ratios per 10 ng/m3 Estimated Using Three Causal Inference Approaches in (Wu, Braun
etal. 2020)
Causal Inference Technique
Hazard Ratio
and 95% Confidence Interval
Weighting
1.076 (1.065—1.088)
Matching
1.068 (1.054—1.083)
Adjustment
1.072 (1.061—1.082)
(Pope III, Lefler et al. 2019) also generated Cox proportional hazard estimates using both complex and
basic models, of which the basic model did not account for complex survey design (using PHREG
procedure in SAS version 9.3). The two estimates for all-cause mortality from long-term PM2.5 exposures
were nearly identical (Table 36).
Table 36. Complex and Basic Cox Proportional Hazard Model Estimates of PIVh.s-Attributable Mortality
per 10 ng/m3
(Pope III, Lefler et al. 2019)
Complex
1.13 (1.11-1.16)
Basic
1.13 (1.11-1.15)
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, 0, and 6.2.4.
The study identified as best characterizing risk for this health endpoint took place in Canada ((Tetreault,
Doucet et al. 2016)). Even though comparatively (Tetreault, Doucet 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, Doucet et al. 2016) and the alternative risk estimates and confidence intervals from
(McConnell, Islam et al. 2010) and (Nishimura, Galanter et al. 2013) can be found in Table 37. Details
about the two studies providing alternate risk estimates is below.
109
-------
Table 37. 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, Doucet et al. 2016) included only Canadians whereas
(Nishimura, Galanter et al. 2013) included five U.S. urban areas and
(McConnell, Islam et al. 2010) was restricted to southern CA
Study Size
(Tetreault, Doucet et al. 2016) included the largest study size,
approximately twenty-five times the size of either (Nishimura, Galanter
et al. 2013) or (McConnell, Islam et al. 2010)
Study Period
(Tetreault, Doucet et al. 2016) evaluated the most recent health study
period (1996-2011) compared to 2002-2006 for (McConnell, Islam et al.
2010) and 1986-2005 for (Nishimura, Galanter et al. 2013)
Exposure Estimate
(Tetreault, Doucet et al. 2016) used hybrid exposure estimates whereas
sever states, whereas (Nishimura, Galanter et al. 2013) and (McConnell,
Islam et al. 2010) used monitor-based estimates
Statistical Technique
(Tetreault, Doucet et al. 2016) and (McConnell, Islam et al. 2010) use
time-varying and multilevel Cox proportional hazard models,
respectively, whereas (Nishimura, Galanter et al. 2013) uses logistical
regression models
Two of the five ISA-identified studies of asthma onset took place in the U.S. ((McConnell, Islam et al.
2010, Nishimura, Galanter et al. 2013)). (McConnell, Islam 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, Islam 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, Galanter 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
110
-------
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 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 38.
Table 38. 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, Doucet et al. 2016)
0-17
0.044 (0.0009)
(McConnell, Islam et al. 2010)
4-17
0.029 (0.017)
(Nishimura, Galanter et al. 2013)
7-21
0.030 (0.069)
6.1.4 Cardiovascular Hospital Admissions
(Bell, Son 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, Rager 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 39). Therefore, we include a risk estimate of
cardiovascular hospital admission impacts from long-term PM2.5 exposure from (Talbott, Rager et al.
2014) as a sensitivity analysis of this health endpoint (Table 40). Please note that (Talbott, Rager et al.
2014) provides individual risk estimates for each state, which will be pooled into a single estimate to
compare with (Bell, Son et al. 2015).
(Talbott, Rager 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, Rager 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 CMAQ
to 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
111
-------
period were identified over single-day 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, Rager 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, Rager et al. 2014), Table 3). Beta effect coefficients from the main (italicized) and
sensitivity analyses are available in Table 40.
Table 39. 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, Rager et al. 2014) included all ages whereas (Bell, Son
et al. 2015) was restricted to ages >64
Confounding by Individual Risk
Factors (Location)
(Talbott, Rager et al. 2014) was restricted to seven states, (Bell,
Son et al. 2015) included all states
Confounding by Other Pollutants
(Talbott, Rager et al. 2014) included the copollutant 03
Exposure Estimate
(Talbott, Rager et al. 2014) used hybrid exposure estimates
whereas sever states, (Bell, Son et al. 2015) used monitor-based
estimates
Table 40. PM2.5-Attributable Cardiovascular Hospital Admissions Beta Estimates
Study
Beta Coefficient (SE)
(Bell, Son et al.
2015)
0.00065 (0.00009)
(Talbott, Rager 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)
A study from the 2020 PM ISA Supplement also evaluated cardiovascular hospital admissions. (Danesh
Yazdi, Wang et al. 2021) examined the relationship between long-term ambient air pollution (N02, 03,
PM2.5) exposure and hospital admissions for four cardiovascular and respiratory outcomes among
63,006,793 Medicare beneficiaries over the age of 64 in the contiguous U.S. The authors retrieved data
from the Medicare denominator file and the Medicare Provider Analysis and review file, including all
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.
112
-------
hospital admissions from 2000 through 2016 with primary discharge codes for myocardial infarction,
ischemic stroke, atrial fibrillation and flutter, and pneumonia. They modeled PM2.5 exposure at a 1-km2
scale using high-resolution spatiotemporal models that combined three machine learning algorithms
using land use terms, chemical transport model predictors, meteorologic variables, and satellite
measurements and validating against monitor data. (Danesh Yazdi, Wang et al. 2021) employed doubly
robust additive models (DRAM), which account for confounding through inverse probability weights of
exposure and adjustment in the outcome regression model. They examined the relationship between
hospital admissions for myocardial infarction, ischemic stroke, atrial fibrillation and flutter, or
pneumonia and each of three ambient air pollutants, adjusting each model for individual,
socioeconomic, and behavioral covariates as well as the other two pollutants. Because we were unable
to obtain validated Hazard Ratios from this study, we elected not to include it in our primary analysis.
Similar to cardiovascular hospital admissions, there was an estimate of PIVh.s-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 41 and Table 42). As compared to PIVh.s-attributable mortality and cardiovascular
hospital admission impact estimates, there may be greater uncertainty associated with estimates of
PM2.5-attributable respiratory hospital admissions (Table 42).
(Jones, Diez-Roux 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 CMAQ and 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 41. 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, Diez-Roux et al. 2015) included all ages whereas (Bell,
Son et al. 2015) was restricted to ages >64
Confounding by Individual Risk
Factors (Location)
(Jones, Diez-Roux et al. 2015) was restricted to a single state,
(Bell, Son et al. 2015) included all states
Exposure Estimate
(Jones, Diez-Roux et al. 2015) used hybrid exposure estimates,
whereas (Bell, Son et al. 2015) used monitor-based estimates
113
-------
Table 42. PM2.5-Attributable Respiratory Hospital Admissions Beta Risk Estimates
Study
Age Range
Beta Coefficient (SE)
(Bell, Son et al. 2015)
65-99
0.00025 (0.0001)
(Ostro, Malig et al. 2016)
0-18
0.00275 (0.0008)
(Jones, Diez-Roux et al. 2015)
0-99
0.00080 (0.0002)
One study analyzed the effects of long-term PM2.5 exposure on cardiovascular hospital admissions with
PM2.5 exposures. (Kloog, Coull et al. 2012) 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. However, as the study area was restricted to New England (Connecticut, Maine,
Massachusetts, New Hampshire, Rhode Island, and Vermont) from 2000 to 2006, this study serves only
as an exposure duration sensitivity analysis for the main benefits assessment endpoints.
6.1.4.1 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, Franklin 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).
Table 43. Comparison of the PM2.5-Attributable Respiratory Hospital Admissions Beta Risk Estimate to
the EHA Respiratory Estimate
Study
Beta Coefficient (SE)
(Bell, Son et al. 2015)
0.00025 (0.0001)
(Zanobetti, Franklin et al. 2009)
0.00204 (0.0004)
6.1.5 Hypertension
The 2022 PM ISA Supplement states that "[r]ecent studies add to the evidence providing support for
positive associations among post-menopausal women and referenced four epidemiologic studies
meeting the minimum criteria (section 2.1.1) of PM2.5-attributable hypertension ((U.S. EPA 2022)).
Additionally, we identified a recent publication estimating the lifetime COI for incidence hypertension
114
-------
(section 6.1.5.2). However, we include hypertension only as a sensitivity analysis here due to the limited
evidence available in the 2019 PM ISA and 2022 Supplement to the ISA ((U.S. EPA 2019, U.S. EPA 2022)).
Specifically:
The literature assessed in the 2019 PM ISA provided evidence of associations between long-term
PM2.5 exposure and hypertension. Recent studies are generally consistent with this assessment,
reporting positive associations among post-menopausal women enrolled in the WHI study and in
cardiac catheterization patients. However, no association between long-term PM2.5 exposure and
hypertension was observed among Black women enrolled in the JHS.
6.1.5.1 Identification of Epidemiologic Studies and Risk Estimates
6.1.5.1.1 Available Epidemiologic Literature
The 2022 supplement to the PM ISA identified three epidemiologic studies of hypertension that met the
minimum identified criteria above ((U.S. EPA 2022), section 2.1.1).
6.1.5.1.2 Identifying Suitable Studies for Use in Benefits Assessments
Of the three available studies, one was more representative of the U.S. with respect to the population
observed and the geographic scope ((Honda, Eliot et al. 2017)).
6.1.5.1.3 Studies and Risk Estimates Identified as Most Suitable for Use in Benefits Assessments
(Honda, Eliot et al. 2017) studied the association between long-term PM2.5 exposure and incident
hypertension in a cohort of 44,255 post-menopausal (aged 50-79 years at baseline) women enrolled in
the Women's Health Initiative (WHI) clinical trials nationwide who were free of hypertension at baseline.
Authors defined incident hypertension as systolic blood pressure > 140 mmHg, diastolic blood pressure
> 90 mmHg, or self-reported anti-hypertensive medication use, and participants were followed from
enrollment until 2010 for a total of 298,383 person-years of follow-up. (Honda, Eliot et al. 2017)
employed a hybrid modeling technique using daily PM2.5 measurements and geographic covariates to
calculate annual moving average estimates of PM2.5, PM10, and PM10-2.5 exposure for each participant.
They modeled the association between hypertension incidence and particulate matter exposure using
single-pollutant and two-pollutant (PM2.5 and PM10-2.5) Cox proportional hazards models that adjusted
for numerous individual-level covariates, clinical trial treatment arm, and WHI study clinical site.
6.1.5.2 Identification of Valuation Estimates
The lifetime cost-of-illness (COI) for incidence hypertension is derived from (Kirkland, Heincelman et al.
2018), who estimate the annual out-of-pocket expenditures associated with high blood pressure among
adults ages 45 and older in the United States. The authors utilize data from the Medical Expenditures
Panel Survey (MEPS) from 2003-2014 and identify patients with hypertension using the ICD-9 code
401.xx. Cost elements include inpatient, outpatient, and emergency room services as well as
prescription medicine expenses. The authors adjust annual expenses such that they reflect the
incremental expenditures for individuals with hypertension versus without hypertension. They calculate
an annual cost of $1,850 ($2015) associated with hypertension. To calculate the lifetime costs of
hypertension we calculate the costs incurred over 20 years, reflective of the rounded age-weighted life-
expectancy of women diagnosed with hypertension in Canada ((Loukine, Waters et al. 2011)). The
continuing annual costs are discounted by three or seven percent. Over 20 years, the total medical costs
are calculated as $28,348 using a 3 percent discount rate and $20,970 using a 7 percent discount rate.
115
-------
6.1,6 Effect Modification of Health Effects 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 2019)).
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 stratified 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
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 2019)). 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.
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.
116
-------
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 PM ISA Chapter 12 for each at-risk factor listed above,
resulting in a set of 123 studies for at-risk populations ((U.S. EPA 2019)). 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 44.
Table 44. PM2.5 At-Risk Study Identification Criteria
Criteria
Description
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
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
117
-------
2.3.1.1): (Di, Wang et al. 2017), (Kioumourtzoglou, Schwartz et al. 2016), (Parker, Kravets et al. 2018),
and (Wang, Shi 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, Wang et al. 2017)
provided sufficient information to apply risk models quantifying increased risks to nonwhite groups,
including non-Hispanic white, Black, Asian, American Indian, and Hispanic-white populations. Additional
detail on the study can be found in section 2.3.1.1.3.1.2 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, Chang 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, Wang et
al. 2017) and (Alhanti, Chang et al. 2016), summarized in Table 45.
Table 45. Identified PM2.5 At-Risk Beta Coefficients and Standard Errors
At-Risk Factor
Endpoint
Study
Population Demographic
Beta Coefficient (SE)
Race, nonwhite
populations
Morality,
All Cause
(Di, Wang et al.
2017)
White
0.0061 (0.0001)
Hispanic
0.0110 (0.0008)
Black
0.0189 (0.0004)
Asian
0.0092 (0.0010)
American Indian
0.0095 (0.0019)
Race, nonwhite
populations
Morality,
All Cause
(Pope III, Lefler et
al. 2019)
Non-Hispanic White
0.001806 (0.00184)
Hispanic
0.004084 (0.00403)
Non-Hispanic Black
0.005064 (0.00485)
Other/unknown
0.007732 (0.00788)
Race, nonwhite
populations
Emergency
Room
Visits,
Asthma
(Alhanti, Chang
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)
118
-------
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 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" ((U.S. EPA 2020)).
(Turner, Jerrett 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 46.
Table 46. 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, Jerrett et al. 2016)
1.14 (1.10-1.18)
1.12 (1.08-1.16)
119
-------
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 ((U.S. EPA 2020)). The 03 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 2020)). Clinical
evidence provides little indication that adverse effects occur at extremely low levels in most individuals.
Epidemiologic evidence is qualitatively discussed further in section 6.5.15.2.
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, Samet 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 47. 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 47.
Table 47. 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, Samet 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
120
-------
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 ((U.S. EPA 2020)). 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 PIVh.s-attributable mortality risk (Table 48). Please note these
long-term, all-cause risk estimates include respiratory mortality estimates and should not be added to
the respiratory mortality estimates.
Table 48. Long-Term 03-Attributable Total Mortality Risk Estimates per 10 ppb
Study
Risk Estimate (per 10
ppb increase in 03)
Risk Estimate Details
(Turner, Jerrett et
al. 2016)
1.02 (1.01-1.03)
Fully adjusted HBM multipollutant estimate from Table
E9, ages 35-99
(Di, Wang et al.
2017)
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, Doucet et al. 2016)). Even though comparatively (Tetreault, Doucet 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, Doucet et al. 2016) and the alternative risk estimates from (Garcia, Berhane et
al. 2019) can be found in Table 49. Details about the study providing an alternate risk estimate is below.
121
-------
Table 49. 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, Doucet et al. 2016) included only Canadians whereas
(Garcia, Berhane et al. 2019) was restricted to southern CA
Study Size
(Tetreault, Doucet et al. 2016) included the largest study size,
approximately twenty-five times the size of (Garcia, Berhane et al.
2019)
Study Period
(Garcia, Berhane et al. 2019) evaluated a more recent and longer
health study period (1993-2014) compared to 2002-2011 for
(Tetreault, Doucet et al. 2016)
Exposure Estimate
(Tetreault, Doucet et al. 2016) used hybrid exposure estimates
whereas sever states, whereas (Garcia, Berhane et al. 2019) used
monitor-based estimates
Statistical Technique
(Tetreault, Doucet et al. 2016) used time-varying Cox proportional
hazard models, whereas (Garcia, Berhane 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, Berhane et al. 2019)).
(Garcia, Berhane 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, Berhane 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 50.
Table 50. Long-Term 03-Attributable Asthma Beta Coefficients
Study
Beta Coefficient
Age Range
(Tetreault, Doucet et al. 2016)
0.020754
0-17
(Garcia, Berhane et al. 2019)
0.01695
9-18
122
-------
6,2.5 Effect Modification of Health Effects 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 method to that described for PM2.5 to compile and assess studies
cited in support of the Agency's determinations ((U.S. EPA 2020), 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 ((U.S. EPA 2020)). 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 51.
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.
123
-------
Table 51. 03 At-Risk Study Identification Criteria
Criteria
Description
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, Samet 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 2020)). 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, Burnett 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 aged 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
85 Calculations required to apply risk model information from Medina-Ramon, M. and J. Schwartz (2008). "Who is
more vulnerable to die from ozone air pollution?" Epidemiology: 672-679. are described in the following
paragraph.
124
-------
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 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:
_ E^Male(P°PMale) ^^Female(P°PFemale)
k "Total — D„„
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:
ERFemale = ERMaie + 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
ERMale = 0.36 %
and
ERFemale = 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, Chen 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, Dales et al. 2006), and (Lin, Stieb 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 52.
125
-------
Table 52. 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, Samet 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, Burnett 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, Chen et al. 2007)
Age 2-4
0.0032
(0.0033)
126
-------
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, Dales et al. 2006)
Female
0.0013
(0.0004)
Male
0.0017
(0.0003)
Sex
Hospital Admissions,
Lower Respiratory
Infection
(Lin, Stieb 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 (Distribution) and National-Level (Red Dot) Emergency Department
Visit and Hospital Admission Baseline Incidence Data
Emergency Caidiovascular
Department Vis ts Re«pii atcr>
Hospital Admissions Cardio- Cerehro- and Peripheral Vascular Disease
Respirator !l ne
-------
• 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 2012)). 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.
Newer evidence suggests the lag between exposure and the change in the risk of PM-attributable
mortality may be shorter than the 20 years EPA now assumes. A long-term exposure cohort study
performed by Crouse and colleagues of the CanCHEC evaluated the estimated hazard of non-accidental
mortality according to three temporal moving averages (1, 3 and 8 years) ((Crouse, Erickson et al.
2020)). The authors found that "...longer moving averages resulted in stronger associations between
PM2.5 and mortality." An analysis of the Harvard Six Cities cohort observed that PM2.5 concentrations
observed in the year prior to mortality "...were the best fit exposure window for all-cause mortality"
((Lepeule, Laden et al. 2012)).
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.
128
-------
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"
(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, Shavlik et al. 2017) for non-fatal lung cancer incidence ((U.S. EPA 2019)). 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 28), we
129
-------
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.
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
53). 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 53. 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.
130
-------
Figure 15. Lung Cancer Cases Cessation Lag Distribution by Model
150-
|
Cases
o
o
L
\
Distribution
— Trianglar
Adjusted 20-Year Distributed Lag
— SEER Age of Diagnosis
50
I
0
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.
131
-------
re 16. Lung Cancer Cases Reduction Distribution
Age at Exposure Change: 33 Age at Exposure Change: 40
0.5
0.4
0 3-
Distribution
— SEER Age of Diagnosis "§
£ 0.2-1 £ 0.2
0 25 SO 75 100
Age
Age at Exposure Change: 50
J? 0.3
0 25 50 75 100
Age
Age at Exposure Change: 70
; o.3-
0 25 50 75 100
Age
Age at Exposure Change: 93
i113
o
£ 0.2
Age
0.0
Distribution
— SEER Age of Diagnosis
0 25 50 75 100
Age
Age at Exposure Change: 60
Distribution 3
S
— SEER Age of Diagnosis "§
€L<
Distribution
— SEER Age of Diagnosis
0 25 50 75 100
Age
Age at Exposure Change: 80
Distribution
— SEER Age of Diagnosis
Distribution
- SEER Age of Diagnosis
L
0 25 50 75 100
Age
Distribution
SEER Age of Diagnosis
132
-------
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 2012)). 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).
We previously examined how sensitive the estimate of total benefits is to alternative estimates of the
income elasticities. Table 54 lists the ranges of elasticity values used to calculate the income adjustment
factors, while Table 55 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 56.
Table 54. 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 55. 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 54, U.S. Census population projections, and projections of real GDP
per capita.
Table 56. 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, Burnett et al. 2000) and 3% discount rate. Results using
(Laden, Schwartz 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.
133
-------
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
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 2020) report and (Jutkowitz, Kane et al. 2017). Using (Alzheimer's 2020), we first developed
an estimate of incremental annual medical expenses for Medicare beneficiaries living with Alzheimer's
Disease (Table 57). Then, using the estimated life expectancy duration of 5 year from (Jutkowitz, Kane et
al. 2017), 3% and 7% discounted costs were extrapolated (Table 58). 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 2020).
Table 57. 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)
$33,266
87 Baseline incidence and prevalence data would need to be updated to estimate impacts of disease onset.
134
-------
Table 58. Lifetime Alzheimer's Disease Valuation Estimate Calculation (2015$)
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
$145,946
(Jutkowitz, Kane et a . 2017) provided information needed to separately develop a lifetime Alzheimer's
Disease cost estimate with a 3% discount rate, but not with a 7% discount rate (Table 59. Additional
Lifetime Alzheimer's Disease Valuation Estimate Calculation with a 3% Discount Rate (2015$)). As the 3%
discount rate estimate of $156,920 from (Alzheimer's 2020) is fairly similar to the lifetime 3% discount
rate estimate of $184,500 from (Jutkowitz, Kane et al. 2017), we have additional confidence in the
validity of the (Alzheimer's 2020) estimates (Table 21).
Table 59. Additional Lifetime Alzheimer's Disease Valuation Estimate Calculation with a 3% Discount
Rate(2015$)
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, Hamilton et al. 2020) provided estimates of lifetime costs, including direct, indirect, and non-
medical costs. Using (Yang, Hamilton et al. 2020), we first developed an annual estimate of excess costs
associated with living with Parkinson's Disease for one year (Table 60). Then, using the estimated life
expectancy duration of 14.6 years from (De Pablo-Fernandez, Tur et al. 2017), the 3% and 7% discounted
costs were extrapolated (Table 61). 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).
135
-------
Table 60. 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
136
-------
Table 61. 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).
137
-------
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 ((IEc 2006, Hammitt 2008)). 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 2019)). 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 pollutant concentration
estimates. 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" ((U.S. EPA 2013, U.S.
EPA 2019)).
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.
138
-------
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 2019)). 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 2020)).
6.5.4 Modeling the Influence of Air Pollution on the Risk of Mortality Over Time
Air pollution benefits assessments commonly use a "pulse" approach, wherein counts of premature
deaths and illnesses are attributed to air quality changes in a single year. To the extent that simulated
changes in air quality persist over time—that is, concentrations are reduced over a multi-year period-
then the pulse method may under-estimate the cumulative impact of air pollution on health ((Roman,
Neumann et al. 2022)). Partly for this reason, researchers have employed a "dynamic" approach to
estimating multi-year changes in air pollution risk using a life table (Miller and Hurley 2003). A life table
estimates the air pollution-attributable risk for each individual in a specified cohort on a year-to-year
basis. When estimating the risk of premature death, the probability of dying in each year is conditional
upon having lived to that year. Noting the advantages of this approach compared to the pulse method,
the U.S. EPA Science Advisory Board noted that life tables "...provide the most realistic available
modeling how, over time, changes in population risk led to changes in the size and age distribution of
the population, with consequent implications for estimated mortality impacts" ((Hammitt and Bailar
2010)).
EPA's PopSim tool, which builds upon the World Health Organization (WHO) LIFET tool, estimates
changes in life expectancy at birth, life years and counts of attributable deaths ((Miller and Hurley
2003)). In contrast with the BenMAP-CE tool, PopSim quantifies these effects at a national scale and
thus (in its current form) is unable to estimate spatially resolved effects; it is also not designed to
quantify morbidity effects and does not yet include baseline death rates stratified by race or ethnicity.
Hence, the tool is a complement, rather than a substitute, for BenMAP-CE.
6.5.5 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 2019 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 2019)).
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
139
-------
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 2019)).
6.5.6 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 2019)). 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.
6.5.7 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
estimates and therefore introduce uncertainty. Additional detail on discounting can be found in the EPA
Guidelines for Preparing Economic Analyses ((U.S. EPA 2014)).
6.5.8 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.
6.5.9 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.
88 Common covariates include education level, marital status, body mass index, cigarette smoking, diet,
occupational exposures, income, percentage of minority, unemployment, and poverty.
140
-------
6.5.10 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.
6.5.11 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 durations. 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.
6.5.12 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.
6.5.13 O3 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
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.
141
-------
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.
6.5.13.1 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.
6.5.13.1.1 DA24 to MDA8
Currently, air quality projections using the DA24 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.
6.5.13.1.2 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 denotes 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.
142
-------
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)
6.5.13.1.3 MDA1 to MDA8
Due to time and resource limitations, air quality projections using the MDA1 metric are also unavailable
for the Proposed Reconsideration of the PM NAAQS 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.
6.5.14 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 2020)). Recently, there are an
143
-------
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 ((Lewis, Robins et al. 2013, Barry,
Klein et al. 2019)). 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.
6.5,15 Shape of the Concentration-Response Relationship
6.5.15.1 PM25
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
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 2019)).
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 (2020). Integrated Science Assessment for Ozone and Related
Photochemical Oxidants (Final Report). O. o. R. a. Development. Washington, DC, U.S. Environmental Protection
Agency.).
144
-------
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
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
145
-------
6.5.15.2 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 2020)). 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 2020), 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 2020), 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,
Jerrett 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.
6.5.16 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
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.
146
-------
biasing results toward the null, we are currently unable perform quantitative uncertainty analyses
regarding this source of uncertainty.
6.5.16.1 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" ((U.S. EPA 2019)). 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)."
6.5.16.2 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)" ((U.S. EPA 2020)).
6.5.17 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.
6.5.17.1 PM2.5
The 2019 PM ISA compared the use of various statistical techniques, spatial random effects, and fixed93
effect models ((U.S. EPA 2019)). 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."
6.5.17.2 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 2020)).
6.5.18 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.
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.
147
-------
6.5.18.1 PM2.s
The PM ISA included a number of studies that assessed whether statistical models adequately account
for temporal trends and weather covariates ((U.S. EPA 2019)). 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 2019), section 11.1.12)
6.5.18.2 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 ((U.S. EPA 2020)). 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.
6.5.19 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.
148
-------
7 References
Adams, P. F., G. E. Hendershot and M. A. Marano (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.
Alhanti, B. A., H. H. Chang, A. Winquist, J. A. Mulholland, L. A. Darrow and S. E. Sarnat (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, A. (2020). "2020 Alzheimer's disease facts and figures." Alzheimers Dement.
Anderson, G. B. and M. L. Bell (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., G. A. Rossi and G. W. Canonica (2007). "Airway remodelling in children: when does it
start?" Curr Opin Allergy Clin Immunol 7(2): 196-200.
Bai, L., S. Shin, R. T. Burnett, J. C. Kwong, P. Hystad, A. van Donkelaar, M. S. Goldberg, E. Lavigne, S.
Weichenthal, R. V. Martin, R. Copes, A. Kopp and H. Chen (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., M. Klein, A. Winquist, H. H. Chang, J. A. Mulholland, E. O. Talbott, J. R. Rager, P. E. Tolbert and S.
E. Sarnat (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, M. L., J. Y. Son, R. D. Peng, Y. Wang and F. Dominici (2015). "Ambient PM2.5 and Risk of Hospital
Admissions: Do Risks Differ for Men and Women?" EpidemioloRV 26(4): 575-579.
Berger, M. C., G. C. Blomquist, D. Kenkel and G. S. Tolley (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 LarynRoscope 121(9): 1830-1833.
Burnett, R., H. Chen, M. Szyszkowicz, N. Fann, B. Hubbell, C. A. Pope, 3rd, J. S. Apte, M. Brauer, A. Cohen,
S. Weichenthal, J. Coggins, Q. Di, B. Brunekreef, J. Frostad, S. S. Lim, H. Kan, K. D. Walker, G. D. Thurston,
R. B. Hayes, C. C. Lim, M. C. Turner, M. Jerrett, D. Krewski, S. M. Gapstur, W. R. Diver, B. Ostro, D.
Goldberg, D. L. Crouse, R. V. Martin, P. Peters, L. Pinault, M. Tjepkema, A. van Donkelaar, P. J. Villeneuve,
A. B. Miller, P. Yin, M. Zhou, L. Wang, N. A. H. Janssen, M. Marra, R. W. Atkinson, H. Tsang, T. Quoc Thach,
J. B. Cannon, R. T. Allen, J. E. Hart, F. Laden, G. Cesaroni, F. Forastiere, G. Weinmayr, A. Jaensch, G. Nagel,
H. Concin and J. V. Spadaro (2018). "Global estimates of mortality associated with long-term exposure to
outdoor fine particulate matter." Proc Natl Acad Sci U S A 115(38): 9592-9597.
Cakmak, S., R. E. Dales and S. Judek (2006). "Respiratory health effects of air pollution gases: modification
by education and income." Archives of environmental & occupational health 61(1): 5-10.
Cameron, T. A. (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. C. Whitman. U.S. EPA HQ, Washington
DC, Office of the Administrator, Science Advisory Board.
Cameron, T. A. (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. M. O. Leavitt. U.S. EPA HQ, Washington DC, Office of the
Administrator, Science Advisory Board.
149
-------
Canada, E. a. C. C. (2016). Canada-United States Transboundary Particulate Matter Science Assessment
2013, Environment and Climate Change Canada.
CBC (2016). Canada by the Numbers. CBC News. Online.
CBO. (2016). "Budget and Economic Data: 10-Year Economic Projections." 2022, from
https://www.cbo.gOv/data/budget-economic-data#3.
Chen, L., B. L. Jennison, W. Yang and S. T. Omaye (2000). "Elementary school absenteeism and air
pollution." Inhal Toxicol 12(11): 997-1016.
Chetty, R., J. N. Friedman and J. E. Rockoff (2014). "Measuring the impacts of teachers II: Teacher value-
added and student outcomes in adulthood." American economic review 104(9): 2633-2679.
Coleman, N. C., R. T. Burnett, M. Ezzati, J. D. Marshall, A. L. Robinson and C. A. Pope, 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, M. L. and A. J. Krupnick (1990). Social costs of chronic heart and lung disease, Quality of the
Environment Division, Resources for the Future.
Crouse, D. L., A. C. Erickson, T. Christidis, L. Pinault, A. van Donkelaar, C. Li, J. Meng, R. V. Martin, M.
Tjepkema, P. Hystad, R. Burnett, A. Pappin, M. Brauer and S. Weichenthal (2020). "Evaluating the
Sensitivity of PM2.5-Mortality Associations to the Spatial and Temporal Scale of Exposure Assessment."
Epidemiology 31(2): 168-176.
Crouse, D. L., P. A. Peters, P. Hystad, J. R. Brook, A. van Donkelaar, R. V. Martin, P. J. Villeneuve, M. Jerrett,
M. S. Goldberg, C. A. Pope, M. Brauer, R. D. Brook, A. Robichaud, R. Menard and R. T. Burnett (2015).
"Ambient PM 2.5, 0 3, and NO 2 exposures and associations with mortality over 16 years of follow-up in
the Canadian Census Health and Environment Cohort (CanCHEC)." Environmental Health Perspectives
123(11): 1180-1186.
Danesh Yazdi, M., Y. Wang, Q. Di, Y. Wei, W. J. Requia, L. Shi, M. B. Sabath, F. Dominici, B. A. Coull and J. S.
Evans (2021). "Long-term association of air pollution and hospital admissions among Medicare
participants using a doubly robust additive model." Circulation 143(16): 1584-1596.
De Pablo-Fernandez, E., C. Tur, T. Revesz, A. J. Lees, J. L. Holton and T. T. Warner (2017). "Association of
Autonomic Dysfunction With Disease Progression and Survival in Parkinson Disease." JAMA Neurol 74(8):
970-976.
Di, Q., L. Dai, Y. Wang, A. Zanobetti, C. Choirat, J. D. Schwartz and F. Dominici (2017). "Association of
short-term exposure to air pollution with mortality in older adults." JAMA: Journal of the American
Medical Association 318(24): 2446-2456.
Di, Q., Y. Wang, A. Zanobetti, Y. Wang, P. Koutrakis, C. Choirat, F. Dominici and J. D. Schwartz (2017). "Air
pollution and mortality in the Medicare population." New England Journal of Medicine 376(26): 2513-
2522.
Dickie, M. and V. L. Messman (2004). "Parental altruism and the value of avoiding acute illness: are kids
worth more than parents?" Journal of Environmental Economics and Management 48(3): 1146-1174.
Ensor, K. B., L. H. Raun and D. Persse (2013). "A case-crossover analysis of out-of-hospital cardiac arrest
and air pollution." Circulation 127(11): 1192-1199.
Exec. Order No. 12898 (1994). Federal Actions To Address Environmental Justice in Minority Populations
and Low-Income Populations. 59.
Freeman III, A. M., J. A. Herriges and C. L. Kling (2014). The measurement of environmental and resource
values: theory and methods, Routledge.
Garcia, E., K. T. Berhane, T. Islam, R. McConnell, R. Urman, Z. Chen and F. D. Gilliland (2019). "Association
of Changes in Air Quality With Incident Asthma in Children in California, 1993-2014." JAMA 321(19): 1906-
1915.
150
-------
Gharibvand, L., D. Shavlik, M. Ghamsary, W. L. Beeson, S. Soret, R. Knutsen and S. F. Knutsen (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, F. D., K. Berhane, E. B. Rappaport, D. C. Thomas, E. Avol, W. J. Gauderman, S. J. London, H. G.
Margolis, R. McConnell and K. T. Islam (2001). "The effects of ambient air pollution on school absenteeism
due to respiratory illnesses." Epidemiology: 43-54.
Gogna, P., T. A. Narain, D. E. O'Sullivan, P. J. Villeneuve, P. A. Demers, P. Hystad, D. R. Brenner, C. M.
Friedenreich, W. D. King and P. S. T. Com (2019). "Estimates of the current and future burden of lung
cancer attributable to PM2.5 in Canada." Prev Med 122: 91-99.
Guerra, S., A. L. Wright, W. J. Morgan, D. L. Sherrill, C. J. Holberg and F. D. Martinez (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, J. K. (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. S. L. Johnson. U.S. EPA HQ, Washington DC, Office of the Administrator, Science
Advisory Board.
Hammitt, J. K. and J. Bailar (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). L. Jackson. U.S. EPA HQ, Washington DC, Office of the Adminitrator, Science Advisory
Board.
Harrington, W. and P. R. Portney (1987). "Valuing the benefits of health and safety regulation." Journal of
Urban Economics 22(1): 101-112.
Harrison, R. M., D. Smith and A. Kibble (2004). "What is responsible for the carcinogenicity of PM2. 5?"
Occupational and environmental medicine 61(10): 799-805.
Higbee, J. D., J. S. Lefler, R. T. Burnett, M. Ezzati, J. D. Marshall, S.-Y. Kim, M. Bechle, A. L. Robinson and C.
A. I. Pope (2020). "Estimating long-term pollution exposure effects through inverse probability weighting
methods with Cox proportional hazards models." Environmental Epidemiology 4(2): e085.
Hollmann, F., T. Mulder and J. J. W. Kalian, 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.
Honda, T., M. N. Eliot, C. B. Eaton, E. Whitsel, J. D. Stewart, L. Mu, H. Suh, A. Szpiro, J. D. Kaufman and S.
Vedal (2017). "Long-term exposure to residential ambient fine and coarse particulate matter and incident
hypertension in post-menopausal women." Environment international 105: 79-85.
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.". L. Jackson. Washington, DC, Office of the Adminstrator, Science
Advisory Board.
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.
lEc (2006). Expanded expert judgment assessment of the concentration - response relationship between
PM2. 5 exposure and mortality.
Jerrett, M., R. T. Burnett, C. A. Pope, 3rd, K. Ito, G. Thurston, D. Krewski, Y. Shi, E. Calie and M. Thun
(2009). "Long-term ozone exposure and mortality." N Engl J Med 360(11): 1085-1095.
151
-------
Jones, M. R., A. V. Diez-Roux, M. S. O'Neill, E. Guallar, A. R. Sharrett, W. Post, J. D. Kaufman and A. Navas-
Acien (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., R. L. Kane, J. E. Gaugler, R. F. MacLehose, B. Dowd and K. M. Kuntz (2017). "Societal and
Family Lifetime Cost of Dementia: Implications for Policy." J Am Geriatr Soc 65(10): 2169-2175.
Katsouyanni, K., J. M. Samet, H. R. Anderson, R. Atkinson, A. Le Tertre, S. Medina, E. Samoli, G. Touloumi,
R. T. Burnett, D. Krewski, T. Ramsay, F. Dominici, R. D. Peng, J. Schwartz, A. Zanobetti and H. E. I. H. R.
Committee (2009). "Air pollution and health: a European and North American approach (APHENA)." Res
Rep Health Eff Inst( 142): 5-90.
Kaye, D. R., H. S. Min, L. A. Herrel, J. M. Dupree, C. E11 imootti I and D. C. Miller (2018). "Costs of Cancer
Care Across the Disease Continuum." Oncologist 23(7): 798-805.
Kioumourtzoglou, M. A., J. Schwartz, P. James, F. Dominici and A. Zanobetti (2016). "PM2.5 and mortality in
207 us cities: Modification by temperature and city characteristics." Epidemiology 27(2): 221-227.
Kirkland, E. B., M. Heincelman, K. G. Bishu, S. O. Schumann, A. Schreiner, R. N. Axon, P. D. Mauldin and W.
P. Moran (2018). "Trends in healthcare expenditures among US adults with hypertension: national
estimates, 2003-2014." Journal of the American Heart Association 7(11): e008731.
Kivi, P. A. and J. F. Shogren (2010). "Second-order ambiguity in very low probability risks: food safety
valuation." Journal of ARricultural and Resource Economics: 443-456.
Kleckner, N. and J. Neumann (1999). "Recommended Approach to Adjusting WTP Estimates to Reflect
Changes in Real Income." Memorandum to Jim Democker, US EPA/OPAR, June 3.
Kloog, I., B. A. Coull, A. Zanobetti, P. Koutrakis and J. D. Schwartz (2012). "Acute and chronic effects of
particles on hospital admissions in New-England." PLoS ONE 7(4): e34664.
Krall, J. R., G. B. Anderson, F. Dominici, M. L. Bell and R. D. Peng (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., R. Burnett, M. Goldberg, K. Hoover, J. Siemiatycki, M. Jerrett, M. Abrahamowich and W.
White (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., M. Jerrett, R. T. Burnett, R. Ma, E. Hughes, Y. Shi, M. C. Turner, C. A. Pope III, G. Thurston and
E. E. Calie (2009). Extended follow-up and spatial analysis of the American Cancer Society study linking
particulate air pollution and mortality, Health Effects Institute Boston, MA.
Krupnick, A. J. and M. L. Cropper (1992). "The effect of information on health risk valuations." Journal of
Risk and Uncertainty 5(1): 29-48.
Kulhanova, I., X. M ore 11 i, A. Le Tertre, D. Loomis, B. Charbotel, S. Medina, J. N. Ormsby, J. Lepeule, R.
Slama and I. Soerjomataram (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.
Laden, F., J. Schwartz, F. E. Speizer and D. W. Dockery (2006). "Reduction in fine particulate air pollution
and mortality: extended follow-up of the Harvard Six Cities study." American Journal of Respiratory
Critical Care Medicine 173(6): 667-672.
Lepeule, J., F. Laden, D. Dockery and J.J. E. h. p. Schwartz (2012). "Chronic exposure to fine particles and
mortality: an extended follow-up of the Harvard Six Cities study from 1974 to 2009." Environmental
Health Perspectives 120(7): 965-970.
Lewis, T. C., T. G. Robins, G. B. Mentz, X. Zhang, B. Mukherjee, X. Lin, G. J. Keeler, J. T. Dvonch, F. Y. Yip, M.
S. O'Neill, E. A. Parker, B. A. Israel, P. T. Max, A. Reyes and C. Community Action Against Asthma Steering
(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.
152
-------
Lin, M., D. M. Stieb and Y. Chen (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.
Liu, J., M. Lee and S. Gershenson (2021). "The short-and long-run impacts of secondary school absences."
Journal of Public Economics 199: 104441.
Loukine, L., C. Waters, B. C. Choi and J. Ellison (2011). "Health-adjusted life expectancy among Canadian
adults with and without hypertension." Cardiology Research and Practice 2011.
Maniloff, P. and N. Fann (2023). "Estimates of the cost of illness of myocardial infarction, asthma and
stroke." Applied Economics: 1-11.
Mar, T. F. and J. Q. Koenig (2009). "Relationship between visits to emergency departments for asthma
and ozone exposure in greater Seattle, Washington." Annals of Allergy, Asthma & Immunology 103(6):
474-479.
McConnell, R., T. Islam, K. Shankardass, M. Jerrett, F. Lurmann, F. Gilliland, J. Gauderman, E. Avol, N.
Kunzli, L. Yao, J. Peters and K. Berhane (2010). "Childhood incident asthma and traffic-related air pollution
at home and school." Environ Health Perspect 118(7): 1021-1026.
McGartland, A., R. Revesz, D. A. Axelrad, C. Dockins, P. Sutton and T. J. Woodruff (2017). "Estimating the
health benefits of environmental regulations." Science 357(6350): 457-458.
Medina-Ramon, M. and J. Schwartz (2008). "Who is more vulnerable to die from ozone air pollution?"
EpidemioloRv: 672-679.
Miller, B. G. and J. F. Hurley (2003). "Life table methods for quantitative impact assessments in chronic
mortality." Journal of EpidemioloRv and Community Health 57: 200 - 206.
National Resources Council (2004). Air Quality Management in the United States. NRC. Washington, DC.,
The National Academies Press.
NCES (1996). The Condition of Education 1996, Indicator 42: Student Absenteeism and Tardiness. U. S. D.
o. Education. Washington, DC, Office of Educational Research and Improvement.
Nishimura, K. K., J. M. Galanter, L. A. Roth, S. S. Oh, N. Thakur, E. A. Nguyen, S. Thyne, H. J. Farber, D.
Serebrisky, R. Kumar, E. Brigino-Buenaventura, A. Davis, M. A. LeNoir, K. Meade, W. Rodriguez-Cintron, P.
C. Avila, L. N. Borrell, K. Bibbins-Domingo, J. R. Rodriguez-Santana, S. Sen, F. Lurmann, J. R. Balmes and E.
G. Burchard (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. N. R. Council,
National Academies Press.
O'Sullivan, A. K., J. Rubin, J. Nyambose, A. Kuznik, D. J. Cohen and D. Thompson (2011). "Cost estimation
of cardiovascular disease events in the US." Pharmacoeconomics 29(8): 693-704.
Ochs, M., J. R. Nyengaard, A. Jung, L. Knudsen, M. Voigt, T. Wahlers, J. Richter and H. J. Gundersen (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. O. f. E. C.-o. a. Development.
OMB (2003). Circular A-4. Washinton, DC, Office of Management and Budget: 58366.
Onukwugha, E., J. McRae, A. Kravetz, S. Varga, R. Khairnar and C. D. Mullins (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." Journal of Environmental
Economics and ManaRement 14(1): 87-98.
Ostro, B., M. Lipsett, J. Mann, H. Braxton-Owens and M. White (2001). "Air pollution and exacerbation of
asthma in African-American children in Los Angeles." Journal of Epidemiology 12(2): 200-208.
Ostro, B., B. Malig, S. Hasheminassab, K. Berger, E. Chang and C. Sioutas (2016). "Associations of Source-
Specific Fine Particulate Matter With Emergency Department Visits in California." Am J Epidemiol 184(6):
450-459.
153
-------
Ostro, B., L. Roth, B. Maligand M. Marty (2009). "The effects of fine particle components on respiratory
hospital admissions in children." Environ Health Perspect 117(3): 475-480.
Ostro, B.C., 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. M. Leavitt. U.S. EPA HQ,
Washington DC, Office of the Administrator, Science Advisory Board.
Ostro, B. D. and S. Rothschild (1989). "Air pollution and acute respiratory morbidity: an observational
study of multiple pollutants." Journal of Environmental Research 50(2): 238-247.
Parker, J. D., L. J. Akinbami and T. J. Woodruff (2009). "Air pollution and childhood respiratory allergies in
the United States." Environ Health Perspect 117(1): 140-147.
Parker, J. D., N. Kravets and A. Vaidyanathan (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 A. E. Smith (2008). "Tracking associations between ambient ozone and asthma-related
emergency department visits using case-crossover analysis." Journal of Public Health ManaRement and
Practice 14(6): 581-591.
Peters, A., D. W. Dockery, J. E. Muller and M. A. Mittleman (2001). "Increased particulate air pollution and
the triggering of myocardial infarction." Circulation 103(23): 2810-2815.
Pope, C. A., M. C. Turner, R. Burnett, M. Jerrett, S. M. Gapstur, W. R. Diver, D. Krewski and R. D. Brook
(2015). "Relationships between fine particulate air pollution, cardiometabolic disorders and
cardiovascular mortality." Circulation Research 116(1): 108-U258.
Pope III, C., J. B. Muhlestein, H. T. May, D. G. Renlund, J. L. Anderson and B. D. Home (2006). "Ischemic
heart disease events triggered by short-term exposure to fine particulate air pollution." Circulation
114(23): 2443-2448.
Pope III, C. A., J. S. Lefler, M. Ezzati, J. D. Higbee, J. D. Marshall, S.-Y. Kim, M. Bechle, K. S. Gilliat, S. E.
Vernon and A. L. Robinson (2019). "Mortality risk and fine particulate air pollution in a large,
representative cohort of US adults." Environmental health perspectives 127(7): 077007.
Pringsheim, T., N. Jette, A. Frolkis and T. D. Steeves (2014). "The prevalence of Parkinson's disease: a
systematic review and meta-analysis." Mov Disord 29(13): 1583-1590.
Rabinovitch, N., M. Strand and E. W. Gelfand (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, M. R. and C. A. Pope, 3rd (1992). "Elementary school absences and PM10 pollution in Utah
Valley." Environ Res 58(2): 204-219.
Roger, V. L., A. S. Go, D. M. Lloyd-Jones, E. J. Benjamin, J. D. Berry, W. B. Borden, D. M. Bravata, S. Dai, E.
S. Ford, C. S. Fox, H. J. Fullerton, C. Gillespie, S. M. Hailpern, J. A. Heit, V. J. Howard, B. M. Kissela, S. J.
Kittner, D. T. Lackland, J. H. Lichtman, L. D. Lisabeth, D. M. Makuc, G. M. Marcus, A. Marelli, D. B. Matchar,
C. S. Moy, D. Mozaffarian, M. E. Mussolino, G. Nichol, N. P. Paynter, E. Z. Soliman, P. D. Sorlie, N.
Sotoodehnia, T. N. Turan, S. S. Virani, N. D. Wong, D. Woo, M. B. Turner, C. American Heart Association
Statistics and S. Stroke Statistics (2012). "Heart disease and stroke statistics-2012 update: a report from
the American Heart Association." Circulation 125(1): e2-e220.
Roman, H., J. E. Neumann, S. Penn, A. White and N. Fann (2022). "Dynamic Versus Static Modeling of
Mortality-Related Benefits of PM2. 5 Reductions in the USA and Chile: 1990 to 2050." Journal of Benefit-
Cost Analysis: 1-26.
Rosamond, W., G. Broda, E. Kawalec, S. Rywik, A. Pajak, L. Cooper and L. Chambless (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.
154
-------
Rosenthal, F. S., J. P. Carney and M. L. Olinger (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.
SEER (2015). SEER: Surveillance, EpidemioloRV, and End Results ProRram. Bethesda, MD, National
Institutes of Health, National Cancer Institute.
Siegel, R. L., K. D. Miller and A. Jemal (2019). "Cancer statistics, 2019." CA Cancer J Clin 69(1): 7-34.
Silverman, R. A., K. Ito, J. Freese, B. J. Kaufman, D. De Claro, J. Braun and D. J. Prezant (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., G. T. O'Connor, B. Rosner, M. R. Segal and S. T. Weiss (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. U.S. EPA HQ, Washington DC, Office of the
Administrator, Science Advisory Board.
Sullivan, J., L. Sheppard, A. Schreuder, N. Ishikawa, D. Siscovick and J. Kaufman (2005). "Relation between
short-term fine-particulate matter exposure and onset of myocardial infarction." EpidemioloRV: 41-48.
Talbott, E. O., J. R. Rager, S. Benson, L. A. Brink, R. A. Bilonick and C. Wu (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, L. F., M. Doucet, P. Gamache, M. Fournier, A. Brand, T. Kosatsky and A. Smargiassi (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, G. S., L. Babcock, M. Berger, A. Bilotti, G. Blomquist, M. Brien, R. Fabian, G. Fishelson, C. Kahn and
A. Kelly (1986). Valuation of reductions in human health symptoms and risks. Prepared for US
Environmental Protection Agency, U.S EPA.
Trivedi, M. and E. Denton (2019). "Asthma in Children and Adults-What Are the Differences and What Can
They Tell us About Asthma?" Front Pediatr 7: 256.
Turner, M. C., M. Jerrett, A. Pope, III, D. Krewski, S. M. Gapstur, W. R. Diver, B. S. Beckerman, J. D.
Marshall, J. Su, D. L. Crouse and R. T. Burnett (2016). "Long-term ozone exposure and mortality in a large
prospective study." American Journal of Respiratory and Critical Care Medicine 193(10): 1134-1142.
U.S. Census Bureau (1998). Population Profile of the United States: 1997. Washington, DC.
U.S. Census Bureau (2000). American Community Survey. Washington, DC.
U.S. Census Bureau (2010). ARe and Sex Composition in the United States: 2010, Government Printing
Office.
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.Rov/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.Rov/. Commuting times file from U.S. Census data portal
(http://dataferrett.census.Rov/), Table P31, variables P031001-P031015.
U.S. Census Bureau (2021). Income and Poverty in the United States: 2020, Government Printing Office.
U.S. EPA (2009). Integrated Science Assessment for Particulate Matter (Final Report). U. S. EPA. Research
Triangle Park, NC, Office of Research and Development, National Center for Environmental Assessment.
155
-------
U.S. EPA (2011). The Benefits and Costs of the Clean Air Act 1990 to 2020: Final Report. U. S. EPA.
Washington, DC, Office of Air and Radiation.
U.S. EPA (2011). 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. EPA. Research Triangle Park, NC, U.S. Environmental Protection Agency, Office of Air Quality Planning
and Standards, Health and Environmental Impact Division.
U.S. EPA (2011). Regulatory Impact Analysis for the Final Mercury and Air Toxics Standards. U. S. EPA.
Research Triangle Park, NC, U.S. Environmental Protection Agency, Office of Air Quality Planning and
Standards, Health and Environmental Impact Division.
U.S. EPA (2011). Regulatory Impact Analysis: National Emission Standards for Hazardous Air Pollutants for
Industral, Commercial, and Institutional Boilers and Process Heaters. U. S. EPA. Research Triangle Park,
NC, U.S. Environmental Protection Agency, Office of Air Quality Planning and Standards, Health and
Environmental Impact Division.
U.S. EPA (2012). Regulatory Impact Analysis for the Final Revisions to the National Ambient Air Quality
Standards for Particulate Matter. U. S. EPA. Research Triangle Park, NC, U.S. Environmental Protection
Agency, Office of Air Quality Planning and Standards, Health and Environmental Impact Division.
U.S. EPA (2012). Regulatory Impact Analysis for the Proposed Revisions to the National Ambient Air
Quality Standards for Particulate Matter. U. S. EPA. Research Triangle Park, NC, U.S. Environmental
Protection Agency, Office of Air Quality Planning and Standards, Health and Environmental Impact
Division.
U.S. EPA (2013). Integrated Science Assessment of Ozone and Related Photochemical Oxidants (Final
Report). U. S. EPA. Research Triangle Park, NC, Office of Research and Development, National Center for
Environmental Assessment.
U.S. EPA (2014). Guidelines for Preparing Economic Analyses, National Center for Environmental
Economics. US Environmental Protection Agency Washington, DC.
U.S. EPA (2015). Preamble to the Integrated Science Assessments. U. S. EPA. Research Triangle Park, NC,
U.S. Environmental Protection Agency, Office of Research and Development, National Center for
Environmental Assessment, RTP Division.
U.S. EPA (2015). Regulatory Impact Analysis of the Final Revisions to the National Ambient Air Quality
Standards for Ground-Level Ozone. U. S. EPA. Research Triangle Park, NC, Office of Air Quality Planning
and Standards, Health and Environmental Impacts Division.
U.S. EPA (2017). Trichloroethylene (TCE); Regulation of Use in Vapor Degreasing Under TSCA Section 6(a).
Washington, DC.
U.S. EPA (2018). Environmental Benefits Mapping and Analysis Program - Community Edition User's
Manual. U. S. EPA. Research Triangle Park, NC, Office of Air Quality Planning and Standards. Updated for
BenMAP-CE Version 1.4.8.
U.S. EPA (2019). Integrated Science Assessment (ISA) for Particulate Matter (Final Report). U. S. EPA.
Research Triangle Park, NC, U.S. Environmental Protection Agency, Office of Research and Development,
Center for Public Health and Environmental Assessment.
U.S. EPA (2019). Policy Assessment for the Review of the Ozone National Ambient Air Quality Standards,
External Review Draft. U. S. EPA. Research Triangle Park, NC, U.S. Environmental Protection Agency, Office
of Air Quality Planning and Standards, Heath and Environmental Impacts Division.
U.S. EPA (2019). 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. EPA.
Research Triangle Park, NC, U.S. Environmental Protection Agency, Office of Air Quality Planning and
Standards, Health and Environmental Impact Division.
U.S. EPA (2020). Handbook on Valuing Changes in Time Use Induced by Regulatory Requirements and
Other EPA Actions. O. o. R. a. Development. Washington, DC, U.S. Environmental Protection Agency.
156
-------
U.S. EPA (2020). Integrated Science Assessment for Ozone and Related Photochemical Oxidants (Final
Report). 0. o. R. a. Development. Washington, DC, U.S. Environmental Protection Agency.
U.S. EPA (2020). Policy Assessment for the Review of the National Ambient Air Quality Standards for
Particulate Matter. U. S. EPA. Research Triangle Park, NC, U.S. Environmental Protection Agency, Office of
Air Quality Planning and Standards, Heath and Environmental Impacts Division.
U.S. EPA (2020). Policy Assessment for the Review of the Ozone National Ambient Air Quality Standards.
U. S. EPA. Research Triangle Park, NC, U.S. Environmental Protection Agency, Office of Air Quality
Planning and Standards, Heath and Environmental Impacts Division.
U.S. EPA (2020). Regulatory Impact Analysis for the for the Proposed Revised Cross-State Air Pollution
Rule (CSAPR) Update for the 2008 Ozone NAAQS. U. S. EPA. Research Triangle Park, NC, U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards, Health and Environmental
Impact Division.
U.S. EPA (2021). Technical Support Document (TSD) for the Final Revised Cross-State Air Pollution Rule
Update for the 2008 Ozone Season NAAQS: Estimating PM2.5- and Ozone-Attributable Health Benefits.
Office of Air Quality Planning and Standards. Durham, NC, U.S. Environmental Protection Agency.
U.S. EPA (2022). Policy Assessment for the Reconsideration of the National Ambient Air Quality Standards
for Particulate Matter (Final Report). U. S. EPA. Research Triangle Park, NC, U.S. Environmental Protection
Agency, Office of Air Quality Planning and Standards, Heath and Environmental Impacts Division.
U.S. EPA (2022). Supplement to the 2019 Integrated Science Assessment for Particulate Matter (Final
Report). U. S. EPA. Research Triangle Park, NC, U.S. Environmental Protection Agency, Office of Research
and Development, Center for Public Health and Environmental Assessment.
US Bureau of Labor Statistics (2021). Current Employment Statistics (National), Employment and Earnings
Tables.
US Bureau of Labor Statistics (2021). Economic News Release: Table 5. Employment status of the
population by sex, marital status, and presence and age of own children under 18, 2020-2021 annual
averages, Employment and Earnings Tables.
Villeneuve, P. J., L. Chen, B. H. Rowe and F. Coates (2007). "Outdoor air pollution and emergency
department visits for asthma among children and adults: a case-crossover study in northern Alberta,
Canada." Environmental Health 6(1): 40.
Viscusi, W. K. (1992). Fatal tradeoffs : public and private responsibilities for risk. New York, Oxford
University Press.
Wang, Y., L. Shi, M. Lee, P. Liu, Q. Di, A. Zanobetti and J. D. Schwartz (2017). "Long-term exposure to PM2.5
and mortality among older adults in the Southeastern US." Epidemiology 28(2): 207-214.
Wei, Y., Y. Wang, Q. Di, C. Choirat, Y. Wang, P. Koutrakis, A. Zanobetti, F. Dominici and J. D. Schwartz
(2019). "Short term exposure to fine particulate matter and hospital admission risks and costs in the
Medicare population: time stratified, case crossover study." bmj 367.
Weichenthal, S., D. L. Crouse, L. Pinault, K. Godri-PolIitt, E. Lavigne, G. Evans, A. van Donkelaar, R. V.
Martin and R. T. Burnett (2016). "Oxidative burden of fine particulate air pollution and risk of cause-
specific mortality in the Canadian Census Health and Environment Cohort (CanCHEC)." Environmental
Research 146: 92-99.
Winer, R. A., X. Qin, T. Harrington, J. Moorman and H. Zahran (2012). "Asthma incidence among children
and adults: findings from the Behavioral Risk Factor Surveillance system asthma call-back survey—United
States, 2006-2008." Journal of Asthma 49(1): 16-22.
Woodruff, T. J., L. A. Darrow and J. D. Parker (2008). "Air pollution and postneonatal infant mortality in
the United States, 1999-2002." Environmental Health Perspectives 116(1): 110-115.
Woodruff, T. J., J. Grillo and K. C. Schoendorf (1997). "The relationship between selected causes of
postneonatal infant mortality and particulate air pollution in the United States." Environmental Health
Perspectives 105(6): 608-612.
157
-------
Woods & Poole (2015). "Complete Demographic Database."
Wu, X., D. Braun, J. Schwartz, M. Kioumourtzoglou and F. Dominici (2020). "Evaluating the impact of long-
term exposure to fine particulate matter on mortality among the elderly." Science Advances 6(29).
Yang, W., J. L. Hamilton, C. Kopil, J. C. Beck, C. M. Tanner, R. L. Albin, E. Ray Dorsey, N. Dahodwala, I.
Cintina, P. Hogan and T. Thompson (2020). "Current and projected future economic burden of Parkinson's
disease in the U.S." NPJ Parkinsons Pis 6: 15.
Zanobetti, A., M. Franklin, P. Koutrakis and J. Schwartz (2009). "Fine particulate air pollution and its
components in association with cause-specific emergency admissions." Environmental Health 8(1): 58.
Zanobetti, A. and J. Schwartz (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.
158
------- |