&EPA
United States
Envirofunmlal Protection
Agnncy
Health Risk and Exposure Assessment for
Ozone
Final Report
-------
This page left intentionally blank
-------
EPA-452/R-14-004a
August 2014
Health Risk and Exposure Assessment for Ozone
Final Report
U.S. Environmental Protection Agency
Office of Air and Radiation
Office of Air Quality Planning and Standards
Health and Environmental Impacts Division
Risk and Benefits Group
Research Triangle Park, North Carolina 27711
-------
DISCLAIMER
This final document has been prepared by staff from the Risk and Benefits Group, Health and
Environmental Impacts Division, Office of Air Quality Planning and Standards, U.S. Environmental
Protection Agency. Any findings and conclusions are those of the authors and do not necessarily reflect
the views of the Agency.
Questions related to this document should be addressed to Dr. Bryan Hubbell, U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards, C539-07, Research
Triangle Park, North Carolina 27711 (email: hubbell.bryan@epa.gov).
-------
TABLE OF CONTENTS
TABLE OF CONTENTS i
LIST OF FIGURES vii
LIST OF TABLES xvii
LIST OF ACRONYMS/ABBREVIATIONS xxiii
1 INTRODUCTION 1-1
1.1 HISTORY 1-3
1.2 CURRENT RISK AND EXPOSURE ASSESSMENT: GOALS AND PLANNED
APPROACH 1-6
1.3 ORGANIZATION OF DOCUMENT 1-7
1.4 REFERENCES 1-8
2 OVERVIEW OF EXPOSURE AND RISK ASSESSMENT DESIGN 2-1
2.1 POLICY-RELEVANT EXPOSURE AND RISK QUESTIONS 2-2
2.2 AIR QUALITY CHARACTERIZATION 2-4
2.2.1 Ozone Chemistry and Response to Changes in Precursor Emissions 2-5
2.2.2 Sources of Ozone and Ozone Precursors 2-6
2.2.3 Simulation of Just Meeting the Existing and Alternative Standards 2-7
2.3 CONSIDERATION OF HEALTH EVIDENCE 2-8
2.3.1 Exposures of Concern 2-9
2.3.2 Health Risk Endpoints 2-10
2.3.3 Exposure and Concentration-Response Functions for Health Risk Endpoints.. 2-13
2.3.4 At-Risk Populations 2-15
2.4 URBAN-SCALE MODELING OF INDIVIDUAL EXPOSURE 2-16
2.4.1 Microenvironmental Ozone Concentrations 2-17
2.4.2 Human Activity Patterns 2-18
2.4.3 Modeling of Exposures Associated with Simulating Just Meeting Ozone
Standards 2-19
2.4.4 Considerations in Selecting Urban Study Areas for the Exposure Analysis 2-20
2.5 URBAN-AND NATIONAL-SCALE RISK ASSESSMENT 2-20
2.5.1 Attributable Risk 2-21
2.5.2 Modeling of Health Risk Associated with Total Ozone Exposure 2-22
2.5.3 Distributions of Risk Across Ozone Concentrations 2-23
-------
2.5.4 Modeling Risk Associated with Simulating Just Meeting Ozone Standards 2-23
2.5.5 Considerations in Selecting Urban Study Areas for the Risk Assessment 2-24
2.6 RISK CHARACTERIZATION 2-24
2.7 REFERENCES 2-26
SCOPE 3-1
3.1 OVERVIEW OF THE EXPOSURE AND RISK ASSESSMENTS FROM THE
LAST REVIEW 3-2
3.1.1 Overview of Exposure Assessment from the Last Review 3-2
3.1.2 Overview of Risk Assessment from the Last Review 3-3
3.2 PLAN FOR THE CURRENT EXPOSURE AND RISK ASSESSMENTS 3-5
3.3 CHARACTTERIZATION OF UNCERTAINTY AND VARIABILITY IN THE
CONTEXT OF THE OZONE EXPOSURE AND RISK ASSESSMENT 3-7
3.4 CHARACTERIZATION OF AIR QUALITY 3-10
3.5 CHARACTERIZATION OF URBAN-SCALE HUMAN EXPOSURE 3-13
3.6 CHARACTERIZATION OF URBAN-SCALE HEALTH RISKS BASED ON
CONTROLLED HUMAN EXPOSURE STUDIES 3-15
3.7 CHARACTERIZATION OF URBAN-SCALE HEALTH RISK BASED ON
EPIDEMIOLOGICAL STUDIES 3-18
3.8 NATIONAL-SCALE MORTALITY RISK ASSESSMENT 3-21
3.9 PRESENTATION OF EXPOSURE AND RISK ESTIMATES TO INFORM THE
OZONE NAAQS POLICY ASSESSMENT 3-24
3.10 REFERENCES 3-26
AIR QUALITY CHARACTERIZATION 4-1
4.1 INTRODUCTION 4-1
4.2 OVERVIEW OF OZONE MONITORING AND AIR QUALITY DATA 4-1
4.3 OVERVIEW OF URBAN-SCALE AIR QUALITY INPUTS TO RISK AND
EXPOSURE ASSESSMENTS 4-4
4.3.1 Urban Study Areas 4-5
4.3.2 Recent Air Quality 4-8
4.3.3 Air Quality Adjustments for Just Meeting Existing and Potential Alternative
Ozone Standards 4-13
4.4 OVERVIEW OF NATIONAL-SCALE AIR QUALITY INPUTS 4-30
4.5 UNCERTAINTIES IN MODELING OF RESPONSES TO EMISSION
REDUCTIONS TO JUST MEET EXISTING AND POTENTIAL ALTERNATIVE
STANDARDS 4-37
4.6 REFERENCES 4-47
CHARACTERIZATION OF URBAN-SCALE HUMAN EXPOSURE 5-1
5.1 SYNOPSIS OF OZONE EXPOSURE AND EXPOSURE MODELING 5-2
5.1.1 Human Exposure 5-2
ii
-------
5.1.2 Estimating Ozone Exposure 5-3
5.1.3 Modeling Ozone Exposure Using APEX 5-4
5.2 SCOPE OF THE EXPOSURE ASSESSMENT 5-7
5.2.1 Urban Areas Selected 5-7
5.2.2 Time Periods Simulated 5-9
5.2.3 Ambient Concentrations Used 5-9
5.2.4 Meteorological Data Used 5-11
5.2.5 Populations Simulated 5-11
5.2.6 Key Physiological Processes and Personal Attributes Modeled 5-15
5.2.7 Microenvironments Modeled 5-17
5.2.8 Model Output 5-18
5.3 EXPOSURE ASSESSMENT RESULTS 5-24
5.3.1 Overview 5-24
5.3.2 Exposure Modeling Results for Base Air Quality 5-24
5.3.3 Exposure Modeling Results for Simulations of Just Meeting the Existing and
Alternative 8-hour Ozone Standards 5-26
5.4 TARGETED EVALUATION OF EXPOSURE MODEL INPUT AND OUTPUT
DATA 5-38
5.4.1 Analysis of Time-Location-Activity Data 5-38
5.4.2 Characterization of Factors Influencing High Ozone Exposures 5-47
5.4.3 Exposure Results for Additional At-Risk Populations and Lifestages, Exposure
Scenarios, and Alternative Air Quality Input Data 5-49
5.4.4 Limited Performance Evaluations 5-57
5.5 VARIABILITY AND UNCERTAINTY 5-70
5.5.1 Treatment of Variability 5-71
5.5.2 Characterization of Uncertainty 5-72
5.6 KEY OBSERVATIONS 5-80
5.7 REFERENCES 5-89
CHARACTERIZATION OF HEALTH RISKS BASED ON CONTROLLED
HUMAN EXPOSURE STUDIES 6-1
6.1 INTRODUCTION 6-1
6.1.1 Development of Approach for Current Risk Assessment 6-2
6.1.2 Comparison of Controlled Human Exposure- and Epidemiologic-based Risk
Assessments 6-3
6.2 SCOPE OF LUNG FUNCTION HEALTH RISK ASSESSMENT 6-4
6.2.1 Selection of Health Endpoints 6-4
6.2.2 Approach for Estimating Health Risk Based on Controlled Human Exposure
Studies 6-5
6.2.3 Controlled Human Exposure Studies 6-6
iii
-------
6.2.4 The McDonnell-Stewart-Smith (MSS) Model 6-9
6.2.5 The Exposure-Response Function Approach Used in Prior Reviews 6-17
6.3 OZONE RISK ESTIMATES 6-22
6.3.1 Lung Function Risk Estimates Based on the McDonnell-Stewart-Smith
Model 6-23
6.3.2 Lung Function Risk Estimates Based on the Exposure-Response Functions
Approach Used in Prior Reviews 6-29
6.3.3 Comparison of the MSS Model with the Exposure-Response Function
Approach 6-30
6.4 EVALUATION OF THE MSS MODEL 6-36
6.4.1 Summary of Published Evaluations 6-36
6.4.2 Children 6-36
6.4.3 Threshold vs. Non-Threshold Models 6-37
6.5 CHARACTERIZATION OF UNCERTAINTY 6-38
6.5.1 Statistical Model Form 6-39
6.5.2 Convergence of APEX Results 6-44
6.5.3 Application of Model for All Lifestages 6-45
6.5.4 Application of Model for Asthmatic Children 6-46
6.5.5 Interaction Between Ozone and Other Pollutants 6-46
6.5.6 2000 vs. 2010 Population Demographics 6-46
6.5.7 Qualitative Assessment of Uncertainty 6-48
6.6 KEY OBSERVATIONS 6-52
6.7 REFERENCES 6-56
7 CHARACTERIZATION OF HEALTH RISK BASED ON EPIDEMIOLOGICAL
STUDIES 7-1
7.1 GENERAL APPROACH 7-1
7.1.1 Basic Structure of the Risk Assessment 7-1
7.1.2 Calculating Ozone-Related Health Effects Incidence 7-11
7.2 AIR QUALITY CONSIDERATIONS 7-13
7.3 SELECTION OF MODEL INPUTS AND ASSUMPTIONS 7-15
7.3.1 Selection of Urban Study Areas 7-15
7.3.2 Selection of Epidemiological Studies and Specification of Concentration-
Response Functions 7-17
7.3.3 Baseline Health Effect Incidence and Prevalence Data 7-31
7.3.4 Population (demographic) Data 7-32
7.4 ADORES SING VARIABILITY AND UNCERTAINTY 7-32
7.4.1 Treatment of Key Sources of Variability 7-35
7.4.2 Qualitative Assessment of Uncertainty 7-40
7.4.3 Description of Core and Sensitivity Analyses 7-45
iv
-------
7.5 URBAN STUDY AREA RESULTS 7-48
7.5.1 Assessment of Health Risk after Adjusting Air Quality to Just Meet the Existing
Standard 7-65
7.5.2 Assessment of Health Risk after Adjusting Air Quality to Just Meet Alternative
Standards 7-67
7.5.3 Sensitivity Analyses Designed to Enhance Understanding of the Core Risk
Estimates 7-72
7.6 KEY OBSERVATIONS 7-85
7.7 REFERENCES 7-88
8 CHARACTERIZATION OF NATIONAL-SCALE MORTALITY RISK BASED
ON EPIDEMIOLOGICAL STUDIES AND AN URBAN-SCALE
REPRESENTATIVENESS ANALYSIS 8-1
8.1 NATIONAL-SCALE ASSESSMENT OF MORTALITY RELATED TO OZONE
EXPOSURE 8-1
8.1.1 Methods 8-2
8.1.2 Results 8-6
8.1.3 Sensitivity Analysis 8-15
8.1.4 Discussion 8-20
8.2 EVALUATING THE REPRESENTATIVENESS OF THE URBAN STUDY
AREAS IN A NATIONAL CONTEXT 8-21
8.2.1 Analysis Based on Consideration of National Distributions of Risk-Related
Attributes 8-22
8.2.2 Analysis Based on Consideration of National Distribution of Ozone-Related
Mortality Risk 8-39
8.2.3 Analysis Based on Consideration of National Responsiveness of Ozone
Concentrations to Emissions Changes 8-43
8.2.4 Discussion 8-65
8.3 REFERENCES 8-66
9 SUMMARY AND SYNTHESIS 9-1
9.1 INTRODUCTION 9-1
9.2 SUMMARY OF KEY RESULTS 9-2
9.2.1 Air Quality Considerations 9-5
9.2.2 Human Exposure Modeling 9-10
9.2.3 Health Risks Based on Controlled Human Exposure Studies 9-15
9.2.4 Health Risks Based on Epidemiological Studies (Chapters 7 and 8) 9-19
9.3 COMPARISON OF RESULTS ACROSS EXPOSURE, LUNG FUNCTION RISK,
AND EPIDEMIOLOGY-BASED MORTALITY AND MORBIDITY RISK
ANALYSES 9-26
9.3.1 Evaluation of Exposures and Risks after Just Meeting the Existing Standard.. 9-27
-------
9.3.2 Reductions in Exposure and Risk Metrics after Just Meeting Alternative
Standards 9-30
9.4 OVERALL ASSESSMENT OF REPRESENTATIVENESS OF EXPOSURE AND
RISK RESULTS 9-34
9.4.1 Representativeness of Selected Urban Study Areas in Reflecting Area across the
Nation with Elevated Risk 9-34
9.4.2 Representativeness of Selected Urban Study Areas in Reflecting Responsiveness
of Risk to Just Meeting Existing and Alternative Os Standards 9-35
9.5 OVERALL ASSESSMENT OF CONFIDENCE IN EXPOSURE AND RISK
RESULTS 9-37
9.5.1 Uncertainties in Modeling Ozone Responses to Meeting Standards 9-38
9.5.2 Uncertainties in Modeling Exposure and Lung Function Risk 9-39
9.5.3 Uncertainties in Modeling Epidemiological-Based Risk 9-39
9.6 OVERALL INTEGREATED CHARACTERIZATION OF RISK IN THE
CONTEXT OF KEY POLICY RELEVANT QUESTIONS 9-41
9.7 REFERENCES 9-46
VI
-------
LIST OF FIGURES
Figure 1-1. Nonattainment Area Classifications Based on the Existing Os NAAQS. (Source:
http://www.epa.gov/airquality/greenbook/) 1-2
Figure 2-1. Overview of Exposure and Risk Assessment Design 2-2
Figure 2-2. Causal Determinations for Os Health Effects 2-12
Figure 3-1. Conceptual Diagram for Air Quality Characterization in the HREA 3-10
Figure 3-2. Conceptual Diagram for Population Exposure Assessment 3-13
Figure 3-3. Conceptual Diagram of Os Lung Function Health Risk Assessment Based on
Controlled Human Exposure Studies 3-15
Figure 3-4. Conceptual Diagram of Urban-Scale Health Risk Assessment Based on Results of
Epidemiology Studies 3-18
Figure 3-5. Conceptual Diagram of National Os Mortality Risk Assessment Based on Results
of Epidemiology Studies 3-22
Figure 4-1. Map of Monitored 8-hr Os Design Values for the 2006-2008 Period 4-3
Figure 4-2. Map of Monitored 8-hr Cb Design Values for the 2008-2010 Period 4-3
Figure 4-3. Flowchart of Air Quality Data Processing for Different Parts of the Urban-scale
Risk and Exposure Assessments 4-4
Figure 4-4. Trends in Annual 4th Highest Daily Maximum 8-hr Average Os Concentrations in
ppb for the 15 Urban Study Areas for 2006-2010. Urban areas are grouped into 3
regions: Eastern (top), Central (middle), and Western (bottom) 4-6
Figure 4-5a. Maps of the 5 Eastern U.S. Urban Study Areas Including Os Monitor Locations.
4-10
Figure 4-6. Flowchart of HDDM Adjustment Methodology to Inform Risk and Exposure
Assessments 4-15
Figure 4-7. Distributions of Composite Monitor Daily Maximum 8-hr Average Os
Concentrations from Ambient Measurements (black), Quadratic Rollback (blue),
and the HDDM Adjustment Methodology (red) for Meeting the Existing
Standard. Values are based on the Zanobetti & Schwartz study areas for April-
October of 2006-2008 4-22
Figure 4-8. Distributions of Composite Monitor Daily Maximum 8-hr Average Os
Concentrations from Ambient Measurements (black), Quadratic Rollback (blue),
and the HDDM Adjustment Methodology (red) for Meeting the Existing
Standard. Values are based on the Zanobetti & Schwartz study areas for June-
August of 2006-2008 4-23
Figure 4-9. Distributions of Composite Monitor Daily Maximum 8-hr Average Values for the
12 Urban Study Areas in the Epidemiology-based Risk Assessment. Plots depict
values based on ambient measurements (base), and values obtained with the
HDDM adjustment methodology when just meeting the 75, 70, 65 and 60 ppb
standards. Values shown are based on CBSAs for April-October of 2007 4-24
vn
-------
Figure 4-10. Distributions of Composite Monitor Daily Maximum 8-hr Average Values for the
12 Urban Study Areas in the Epidemiology-based Risk Assessment. Plots depict
values based on ambient measurements (base), and values obtained with the
HDDM adjustment methodology when just meeting the 75, 70, 65 and 60 ppb
standards. Values shown are based on CBSAs for April-October of 2009 4-25
Figure 4-11. Maps Showing the 4th highest (top) and May-September Average (bottom) Daily
Maximum 8-hr Average Cb Concentrations in Atlanta based on 2006-2008
Ambient Measurements (left), HDDM Adjustment to Meet the Existing Standard
(center), and HDDM Adjustment to Meet AN Alternative Standard of 65 ppb
(right) 4-27
Figure 4-12. Maps Showing the 4th highest (top) and May-September Average (bottom) Daily
Maximum 8-hr Average Cb Concentrations in New York based on 2006-2008
Ambient Measurements (left), HDDM Adjustment to Meet the Existing Standard
(center), and HDDM Adjustment to Meet the Alternative Standard of 65 ppb
(right) 4-28
Figure 4-13. Maps of 4th Highest (top) and May-September Average (bottom) Daily Maximum
8-hr Average Os Concentrations in Houston for 2006-2008 Ambient
Measurements (left), HDDM Adjustment to Meet the Existing Standard (center),
and HDDM Adjustment to Meet the Alternative Standard of 65 ppb (right).. ..4-29
Figure 4-14. May-September Mean of the Daily Maximum 8-hr Average Os Concentrations in
ppb, based on a Downscaler Fusion of 2006-2008 Average Monitored Values
with a 12km 2007 CMAQ Model Simulation 4-31
Figure 4-15. June-August Average 8-hr Daily 10am-6pm Mean Os Concentrations in ppb,
based on a Downscaler Fusion of 2006-2008 Average Monitored Values with a
12km 2007 CMAQ Model Simulation 4-32
Figure 4-16. April-September Mean of the Daily Maximum 1-hr Cb Concentrations in ppb,
based on a Downscaler Fusion of 2006-2008 Average Monitored Values with a
12km 2007 CMAQ Model Simulation 4-33
Figure 4-17. Frequency and Cumulative Distributions of the Three Fused Seasonal Average Os
Surfaces Based on all CMAQ 12 km Grid Cells 4-34
Figure 4-18. 2006-2008 Os Design Values Versus 2006-2008 Fused Seasonal Average Os
Levels for the CMAQ 12km Grid Cells Containing Os Monitors 4-36
Figure 4-19. Propagated Standard Error in Daily Maximum 8-hr Average Os Concentrations
Due to Uncertainty in Linear Regression Central Tendency for Each Urban Study
Area 4-39
Figure 4-20. Propagated Standard Error in Daily Maximum 8-hr Average Os Concentrations
Due to Uncertainty in Linear Regression Central Tendency for Each Urban Study
Area. Note that this is the same as Figure 4-19 above except with Y-axis adjusted.
4-40
Figure 5-1. Conceptual Framework Used for Estimating Study Area Population Os Exposure
Concentrations 5-8
Figure 5-2. Percent of Asthmatic School-age Children in Atlanta with at least One Daily
Maximum 8-hr Average Os Exposure while at Moderate or Greater Exertion,
2006-2010 Air Quality Adjusted to Just Meet the Existing Standard 5-21
Vlll
-------
Figure 5-3. Percent of Asthmatic School-age Children in Atlanta with at least One Daily
Maximum 8-hr Average Cb Exposure while at Moderate or Greater Exertion,
2006-2010 Air Quality 5-22
Figure 5-4. Percent of Asthmatic School-age Children in Atlanta with Multiple (> 2, > 4, > 6
per Os season - red, green, and blue lines) Daily Maximum 8-hr Average Cb
Exposures at or above 60 ppb while at Moderate or Greater Exertion, 2006-2010
Air Quality (x-axes) Adjusted to Just Meet 8-hr Cb Standard Levels of 75, 70, 65,
60 ppb (panels left to right) 5-23
Figure 5-5. Percent of All School-age Children with at Least One Daily Maximum 8-hr
Average Os Exposure at or above 60, 70, and 80 ppb (red, green and blue lines)
while at Moderate or Greater Exertion, 2006-2010 Air Quality Adjusted to Just
Meet 8-hr Average Os Standards of 75, 70, 65, and 60 ppb (panels left to right) in
15 Urban Study Areas 5-31
Figure 5-6. Percent of Asthmatic School-age Children with at Least One Daily Maximum 8-
hr Average Os Exposure at or above 60, 70, and 80 ppb (red, green and blue lines)
while at Moderate or Greater Exertion, 2006-2010 Air Quality Adjusted to Just
Meet 8-hr Average Os Standards of 75, 70, 65, and 60 ppb (panels left to right), 15
Urban Study Areas 5-32
Figure 5-7. Percent of Asthmatic Adults with at Least One Daily Maximum 8-hr Average Os
Exposure at or above 60, 70, and 80 ppb (red, green and blue lines) while at
Moderate or Greater Exertion, 2006-2010 Air Quality Adjusted to Just Meet 8-hr
Average Os Standards of 75, 70, 65, and 60 ppb (panels left to right) in 15 Urban
Study Areas 5-33
Figure 5-8. Percent of Older Adults with at Least One Daily Maximum 8-hr Average Os
Exposure at or above 60, 70, and 80 ppb (red, green and blue lines) while at
Moderate or Greater Exertion, 2006-2010 Air Quality Adjusted to Just Meet 8-hr
Average Os Standards of 75, 70, 65, and 60 ppb (panels left to right) in 15 Urban
Study Areas 5-34
Figure 5-9. Percent of All School-age Children with > 2, > 4, > 6 (red, green, and blue lines)
Daily Maximum 8-hr Average Os Exposures at or above 60 ppb while at
Moderate or Greater Exertion, 2006-2010 Air Quality (x-axes) Adjusted to Just
Meet 8-hr Average Os Standards of 75, 70, 65, 60 ppb (panels left to right) in 15
Urban Study Areas 5-35
Figure 5-10. Mean Number of People with at Least One Daily Maximum 8-hr Average
Exposure at or above 60 ppb while at Moderate or Greater Exertion, 2006-2010
Air Quality Adjusted to Just Meet 8-hr Standards of 75, 70, 65, and 60 ppb (x-
axes, from right to left), in 15 Urban Study Areas 5-36
Figure 5-11. Mean Number of People with at Least One Daily Maximum 8-hr Average
Exposure at or above 70 ppb while at Moderate or Greater Exertion, 2006-2010
Air Quality Adjusted to Just Meet 8-hr Standards of 75, 70, 65, and 60 ppb (x-
axes, from right to left), in 15 Urban Study Areas 5-37
Figure 5-12. Distribution of Afternoon Time Spent Outdoors for School-age Children on Days
When Daily Maximum Temperature was > 84 °F, Stratified by Four U.S.
Regions 5-47
IX
-------
Figure 5-13. Comparison of the Percent of All School-age Children in Detroit Having Daily
Maximum 8-hr Average Cb Exposures at or above 60 ppb while at Moderate or
Greater Exertion, June-August 2007 5-50
Figure 5-14. Percent of Outdoor Workers Ages 19-35 Experiencing Exposures at or above
Benchmark Levels while at Moderate or Greater Exertion using an Outdoor
Worker Scenario-based Approach (top panel) and a General Population-based
Approach (bottom panel), Air Quality Adjusted to Just Meet the Existing
Standard in Atlanta, GA, Jun-Aug, 2006 5-52
Figure 5-15. Percent of All School-age Children (left panel) and Asthmatic School-age
Children (right panel) Having at Least One Daily Maximum 8-hr Average Os
Exposure at or above Benchmark Levels while at Moderate or Greater Exertion
During a 2-day Simulation in Detroit, Base Air Quality, August 1-2, 2007 5-54
Figure 5-16. Comparison of APEX Exposure Results Generated for Three Urban Study Areas
(Atlanta, Detroit, and Philadelphia) using Three Different 2005 Air Quality Input
Data Sets: AQS, VNA, and eVNA 5-55
Figure 5-17. Comparison of Exposure Results Generated by APEX using Two Different Air
Quality Adjustment Approaches to Just Meet the Existing Standard in Atlanta:
Quadratic Rollback (left panel) and HDDM-model Simulation (right panel)... 5-57
Figure 5-18. Distribution of Estimated Daily Average Os Exposures (top panels) and Daily
Total Outdoor Time (bottom panels) for DEARS Study Adult Participants (left
panels) and APEX Simulated Adults (right panels) in Wayne County, MI, July-
August 2006 5-58
Figure 5-19. Distribution of Indoor/Outdoor Concentration Ratios for APEX Simulated Adults
in Wayne County, MI, July-August 2006 5-62
Figure 5-20. Daily Maximum 8-hr Exposures at or above 60 ppb Estimated by APEX for
Simulated Adults in Wayne County, MI, July-August 2006 5-63
Figure 5-21. Means (and range) of 6-day Average Personal Os Exposures, Measured and
Modeled (APEX), Upland Ca 5-64
Figure 5-22. Comparison of Body Mass Normalized Mean Daily Ventilation Rates Estimated
by APEX (closed symbols) and by Brochu et al., 2006 (open symbols) 5-66
Figure 5-23. Comparison of Body Mass Normalized Mean Daily Ventilation Rates in Male and
Female School-age Children (5-18) when Correcting Brochu et al. (2006a)
Results with Child Appropriate VQ Estimates 5-67
Figure 5-24. Comparison of Body Mass Normalized Daily Mean Ventilation Rates in School-
age Children (5-18) Estimated using APEX and Literature Reported Values... 5-68
Figure 5-25. Incremental Decreases in Percent of All School-age Children Experiencing Daily
Maximum 8-hr Os Exposures at or above 60 ppb, with Increasing Stringency in
Adjusted 2006-2010 Air Quality Standard Levels 5-88
Figure 6-1. Two-Compartment Model 6-10
Figure 6-2. Distribution of Responses (Lung Function Decrements in FEVi) Predicted by the
MSS Model for 20-Year Old Individuals. Exposure to 100 ppb Os at Moderate
Exercise (40 L/min, BSA=2 m2) Under the Conditions of a Typical 6.6-hour
Clinical Study 6-14
Figure 6-3. Median Response (Lung Function Decrements in FEVi) Predicted by the MSS
Model for 20-Year Old Individuals. Exposure to 100 ppb Os at Moderate Exercise
-------
(40 L/min, BSA=2 m2) Under the Conditions of a Typical 6.6-hour Clinical
Study 6-15
Figure 6-4. Median Response (FEVi Decrements) Predicted by the MSS Threshold and Non-
Threshold Models for 20-Year Old Individuals, Constant 100 ppb Os Exposure
For 6.6 Hours, 2 Hours Heavy Exercise From Hour 1 to 3 (30 L/min-m2 BSA)...
6-16
Figure 6-5. Probability of Response > 10% Predicted by the MSS Threshold and Non-
Threshold Models for 20-Year Old Individuals, Constant 100 ppb Os Exposure
For 6.6 Hours, 2 Hours Heavy Exercise From Hour 1 to 3 (30 L/min-m2 BSA)...
6-17
Figure 6-6. Probabilistic E-R Relationships for FEVi Decrements > 10% for 8-hr Exposures
At Moderate Exertion, Ages 18-35 6-21
Figure 6-7. Risk Results For All School-Aged Children With > 1 Occurrences of FEVi
Decrements > 10, 15, 20% For All Study Areas, Year, and Scenarios (y-axis is
percent of children affected) 6-24
Figure 6-8. Risk Results For All School-Aged Children With > 1 Occurrences of FEVi
Decrements > 10% Under The 0.07 ppm Alternative Standard Showing
Variability Across Study Areas (Horizontally) and Years (Vertically) 6-25
Figure 6-9. Distribution of Daily FEVi Decrements > 10% Across Ranges of 8-hr Average
Ambient Os Concentrations (Los Angeles, 2006 recent air quality) 6-28
Figure 6-10. Comparison of E-R and MSS Model (restricted to 8-hr average EVR > 13)
Response Functions (Atlanta 2006 base case, ages 18-35) 6-34
Figure 6-11. Distribution of Daily Maximum 8-hr Average EVR For Values of EVR > 13
(L/min-m2) (midpoints on vertical axis) (Atlanta 2006 base case, ages 18-35). 6-35
Figure 6-12. Sensitivity (Percent Change) of Population With One or More FEVi Decrements
> 10% to a 5% Increase in Individual MSS Model Parameter Estimates 6-41
Figure 6-13. Lung Function Risk Results, Incremental Increases In Risk For Increasing
Standard Levels: Percent of All School-aged Children With FEVi Decrement >
10%, Highest Value For Each Study area Over Years 6-54
Figure 6-14. Lung Function Risk Results, Incremental Increases In Risk For Increasing
Standard Levels: Percent of All School-aged Children With FEVi Decrement >
10%, Mean Value For Each Study Area Over Years 6-55
Figure 7-1. Flow Diagram of Risk Assessment for Short-term Exposure Studies 7-10
Figure 7-2. Heat Maps for Short-term Os-attributable Mortality, 2007 Air Quality Adjusted to
Just Meet the Existing Standard and Risk Reductions from Just Meeting
Alternative Standards. Smith etal. (2009) C-R functions 7-55
Figure 7-3. Heat Maps for Short-term Os-attributable Mortality, 2009 Air Quality Adjusted to
Just Meet the Existing Standard and Risk Reductions from Just Meeting
Alternative Standards. Smith etal. (2009) C-R functions 7-56
Figure 7-4. Short-Term Os-attributable All-Cause Mortality for 2007 (top panel) and 2009
(bottom panel) Air Quality Adjusted to Just Meet the Existing and Alternative
Standards. Smith et al. (2009) C-R functions 7-57
Figure 7-5. Short-Term Os-attributable Respiratory Hospital Admissions, 2007 (top panel)
and 2009 (bottom panel) Air Quality Adjusted to Meet the Existing and
Alternative Standards. Medina-Ramon, etal. (2006) C-R functions 7-61
xi
-------
Figure 7-6. Long-Term Cb-attributable Respiratory Mortality for 2007 and 2009 Air Quality
Adjusted to Just Meet the Existing and Alternative Standards, Jerrett et al. (2009)
C-R functions 7-64
Figure 7-7. Sensitivity Analysis: Effect of Air Quality Factors on Short-Term Cb-attributable
Mortality, 2009 Air Quality. SAl-smaller (Smith et al., 2009-based) study area,
SA2-alternative method for simulating standards 7-80
Figure 7-8. Sensitivity Analysis: Effect of C-R Function Specification on Short-Term Cb-
attributable Mortality, 2009 Air Quality. SA1-regional Bayes adjustment, SA2-
co-pollutant model (PMio), SA3-Zanobetti and Schwartz-based function 7-81
Figure 7-9. Sensitivity Analysis: Long-Term Cb-attributable Respiratory Mortality (threshold
models: Cb-only effect estimates), 2009 Air Quality 7-84
Figure 8-1. Conceptual Diagram for National-scale Mortality Risk Assessment 8-2
Figure 8-2. Estimated Annual non-Accidental Premature Deaths (individuals) in 2007
Associated with Average 2006-2008 May-September Means of the Daily
Maximum 8-hr Average Cb Levels by U.S. County using Smith et al. (2009)
Effect Estimates 8-10
Figure 8-3. Estimated Annual all-Cause Premature Deaths (individuals) in 2007 Associated
with Average 2006-2008 June-August Means of the Daily 8-hr Average (10am-
6pm) Cb Levels by U.S. County using Zanobetti and Schwartz (2008) Effect
Estimates 8-10
Figure 8-4. Estimated Annual Adult (age 30+) Respiratory Premature Deaths (individuals) in
2007 Associated with Average 2006-2008 April-September Means of the Daily
Maximum 1-hr Cb Levels by U.S. County using Jerrett et al. (2009) Effect
Estimates 8-11
Figure 8-5. Estimated Percentage of May-September Total non-Accidental Mortality (all
ages) Attributable to 2006-2008 average Cb Levels by U.S. County using Smith et
al. (2009) Effect Estimates 8-11
Figure 8-6. Estimated Percentage of June-August Total all-Cause Mortality (all ages)
Attributable to 2006-2008 Average Cb Levels by U.S. County using Zanobetti and
Schwartz (2008) Effect Estimates 8-12
Figure 8-7. Estimated Percentage of April-September Respiratory Mortality among Adults
Age 30+ Attributable to 2006-2008 Average Cb Levels by U.S. County using
Jerrett etal. (2009) Effect Estimates 8-12
Figure 8-8. Cumulative Distribution of County-level Percentage of all-cause, all-year, and all-
age Mortality Attributable to 2006-2008 Average Cb for the U.S 8-13
Figure 8-9. Cumulative Percentage of Total Cb deaths by Baseline Cb Concentration. Ozone
Concentrations are Reported as May-September Mean of the Daily Maximum 8-
hr Average Concentration for Results based on Smith et al. (2009) Effect
Estimates, June-August Mean of the Daily 8-hr Average (10am to 6pm) for
Results based on Zanobetti and Schwartz (2008) Effect Estimates, and April-
September Mean of the Daily Maximum 1-hr Concentration for Results based on
Jerrett etal. (2009) Effect Estimates 8-14
Figure 8-10. Seven Regions Used in the Sensitivity Analysis based on the Smith et al. (2009)
Regional-prior Bayes-shrunken City-specific and Regional Average Effect
Estimates (Source: Samet et al., 2000) 8-17
xii
-------
Figure 8-11. Ozone-attributable Premature Deaths by Region as Calculated by Applying Smith
et al. (2009) Regional-prior Bayes-shrunken and Regional-average Effect
Estimates, as Compared with the National-prior Bayes-shrunken and National-
average Effect Estimates as Done for the Main Results 8-18
Figure 8-12. Comparison of County-level Populations of Urban Study Area Counties to the
Frequency Distribution of Population in 3,143 U.S. Counties 8-31
Figure 8-13. Comparison of County-level Seasonal Means of the Daily Maximum 8-hr
Average Os Concentrations in Urban Study Area Counties to the Frequency
Distribution of Seasonal Mean of the Daily Maximum 8-hr Os Concentrations in
671 U.S. Counties with Os Monitors 8-31
Figure 8-14. Comparison of 2007 County-level 4th highest Daily Maximum 8-hr Average Os
Concentrations in Urban Study Area Counties to the Frequency Distribution of
2007 4th highest Daily Maximum 8-hr Average Os Concentrations in 725 U.S.
Counties with Os Monitors 8-32
Figure 8-15. Comparison of County-level all-cause Mortality in Urban Study Area Counties to
the Frequency Distribution of all-cause Mortality in 3,137 U.S. Counties 8-32
Figure 8-16. Comparison of County-level non-accidental Mortality in Urban Study Area
Counties to the Frequency Distribution of non-accidental Mortality in 3,135 U.S.
Counties 8-33
Figure 8-17. Comparison of City-level all-cause Mortality Risk Coefficients from Zanobetti
and Schwartz (2008) in Urban Study Areas to the Frequency Distribution of all-
cause Mortality Risk Coefficients from Zanobetti and Schwartz (2008) in 48 U.S.
Cities 8-33
Figure 8-18. Comparison of City-level National-prior Bayes-shrunken non-accidental Mortality
Risk Coefficients from Smith et al. (2009) in Urban Study Areas to the Frequency
Distribution of National-prior Bayes-shrunken non-accidental Mortality Risk
Coefficients from Smith et al. (2009) in 98 U.S. Cities 8-34
Figure 8-19. Comparison of County-level Percent of Population 0 to 14 years old in Urban
Study Area Counties to the Frequency Distribution of Percent of Population 0 to
14 years old in 3,141 U.S. Counties 8-35
Figure 8-20. Comparison of County-level Percent of Population age 65 years old and older in
Urban Study Area Counties to the Frequency Distribution of Percent of
Population age 65 and older in 3,141 U.S. Counties 8-36
Figure 8-21. Comparison of County-level Income per capita in Urban Study Areas to the
Frequency Distribution of Income per capita in 3,141 U.S. Counties 8-36
Figure 8-22. Comparison of County-level July Temperature in Urban Study Area Counties to
the Frequency Distribution of July Temperature in all U.S. Counties 8-37
Figure 8-23. Comparison of City-level Asthma Prevalence in Urban Study Areas to the
Frequency Distribution of Asthma Prevalence in 184 U.S. Cities 8-37
Figure 8-24. Comparison of City-level Air Conditioning Prevalence in Urban Study Areas to
the Frequency Distribution of Air Conditioning Prevalence in 76 U.S. Cities.. 8-38
Figure 8-25. Cumulative Distribution of County-level Percentage of May-September non-
accidental Mortality for all ages Attributable to 2006-2008 Average Os for the
U.S. and the Locations of the Selected Urban Study Areas along the Distribution,
using Smith etal. (2009) Effect Estimates 8-41
xni
-------
Figure 8-26. Cumulative Distribution of County-level Percentage of June-August all-cause
Mortality for all ages Attributable to 2006-2008 Average Os for the U.S. and the
Locations of the Selected Urban Study Areas along the Distribution, using
Zanobetti and Schwartz (2008) Effect Estimates 8-42
Figure 8-27. Cumulative Distribution of County-level Percentage of April-September Adult
(age 30+) Respiratory Mortality Attributable to 2006-2008 Average Os for the
U.S. and the Locations of the Selected Urban Study Areas along the Distribution,
using Jerrett et al. (2009) Effect Estimates and City Definitions from Zanobetti
and Schwartz (2008) 8-42
Figure 8-28. Change in 50th Percentile Summer Season (April-October) Daily Maximum 8-hr
Average Os Concentrations between 2001-2003 and 2008-2010 8-46
Figure 8-29. Change in 95th Percentile Summer Season (April-October) Daily Maximum 8-hr
Average Os Concentrations between 2001-2003 and 2008-2010 8-46
Figure 8-30. Population Density within Census Tract where each Os Monitor is Located.... 8-47
Figure 8-31. Distributions of Os Concentrations for High Population Density Monitors by
Different Subsets of Months over a 13-year Period 8-49
Figure 8-32. Distributions of Os Concentrations for Medium Population Density Monitors by
Different Subsets of Months over a 13-year Period 8-49
Figure 8-33. Distributions of Os Concentrations for Low Population Density Monitors by
Different Subsets of Months over a 13-year Period 8-50
Figure 8-34. Map of Os Trends at Specific Monitors in the New York Area 8-51
Figure 8-35. Map of Os Trends at Specific Monitors in the Chicago Area. All upward and
downward facing triangles represent statistically significant trends from 1998-
2011 (p < 0.05), circles represent locations with no significant trends 8-52
Figure 8-36. Ratio of April-October Seasonal Average Os Concentrations in the Brute-force
50% NOx Emissions Reduction CMAQ Simulations to April-October Seasonal
Average Os Concentrations in the 2007 Base CMAQ Simulation 8-56
Figure 8-37. Ratio of April-October Seasonal Average Os Concentrations in the Brute-force
90% NOx Emissions Reduction CMAQ Simulations to April-October Seasonal
Average Os Concentrations in the 2007 Base CMAQ Simulation 8-56
Figure 8-38. Ratio of January Monthly Average Os Concentrations in Brute-force 50% NOx
Emissions Reduction CMAQ Simulations to January Monthly Average Os
Concentrations in the 2007 Base CMAQ Simulation 8-57
Figure 8-39. Ratio of January Monthly Average Os Concentrations in Brute-force 90% NOx
Emissions Reduction CMAQ Simulations to January Monthly Average Os
Concentrations in the 2007 Base CMAQ Simulation 8-57
Figure 8-40. Histograms of U.S. Population Living in Locations with Increasing and
Decreasing Mean Os Concentrations. Values on the x-axis represent change in
mean Os (ppb) from the 2007 base CMAQ simulation to the 50% NOx cut CMAQ
simulation 8-59
Figure 8-41. Histograms of U.S. Population Living in Locations with Increasing and
Decreasing Mean Os Concentrations. Values on the x-axis represent change in
mean Os (ppb) from the 2007 base CMAQ simulation to the 90% NOx cut CMAQ
simulation 8-60
xiv
-------
Figure 8-42. Population (as % of total case-study area population) Living in Locations of
Increasing April-October Seasonal Mean Os in the 50% NOx Reduction CMAQ
Simulation 8-61
Figure 8-43. Population (as % of total case-study area population) Living in Locations of
Increasing April-October Seasonal Mean Os in the 90% NOx Reduction CMAQ
Simulation 8-61
Figure 8-44. Histograms of U.S. Population Living in Locations with Increasing and
Decreasing Mean Os Concentrations. Values on the x-axis represent the change in
seasonal mean (April-October) Os from the 2007 base CMAQ simulation to the
50% NOx cut CMAQ simulation 8-63
Figure 8-45. Histograms of U.S. Population Living in Locations with Increasing and
Decreasing Mean Os Concentrations. Values on the x-axis represent the change in
seasonal mean (April-October) Os from the 2007 base CMAQ simulation to the
90% NOx cut CMAQ simulation 8-64
Figure 9-1. Distributions of Composite Monitor Daily Maximum 8-hr Average Os
Concentrations from Ambient Measurements (black), Quadratic Rollback (blue),
and the HDDM Adjustment Methodology (red) Used for Just Meeting the
Existing Os Standard 9-9
Figure 9-2. Effect of Just Meeting Existing (column 1) and Alternative (columns 2 through 4)
Standards on the Percent of All School-age Children (ages 5-18) with at Least
One Daily Maximum 8-hr Os Exposure at or above 60, 70, and 80 ppb while at
Moderate or Greater Exertion, 2006-2010 Air Quality 9-13
Figure 9-3. Effect of Just Meeting the Existing (75 ppb) and Alternative Standards on the
Percent of All School-age Children (ages 5-18) at or above the 60 ppb Exposure
Benchmark, Maximum Value Across All Years in Each Study Area, 2006-2010
Air Quality 9-15
Figure 9-4. Effect of Just Meeting the Existing (column 1) and Alternative (columns 2-4)
Standards on the Percent of All School-age Children with FEVi decrements > 10,
15, and 20%, 2006-2010 Air Quality 9-17
Figure 9-5. Effect of Just Meeting Existing (75 ppb) and Alternative Standards on the Percent
of All School-age Children (ages 5-18) with FEVi Decrements > 10%, Maximum
Value for Each Urban Study Area, 2006-2010 Air Quality 9-19
Figure 9-6. Effect of Just Meeting Existing (75 ppb) and Alternative Standard Levels on
Mortality Risk per 100,000 Population (left panels) and on Adult (ages 65 and
older) Respiratory Hospital Admissions Risk per 100,000 Population, 2007 (top
panels) and 2009 (bottom panels) Air Quality 9-21
Figure 9-7. Comparison of Exposure (Row 1), Lung Function Risk (Row 2) and
Epidemiology-Based Risk (Rows 3 and 4) Metrics after Just Meeting the Existing
75 ppb Standard 9-29
Figure 9-8. Comparison of the Percent Reduction in Key Risk Metrics for Alternative
Standard Levels Relative to Just Meeting the Existing 75 ppb Standard 9-32
xv
-------
This page left intentionally blank
xvi
-------
LIST OF TABLES
Table 3-1. Short-term Os Exposure Health Endpoints Evaluated in Urban Study Areas. ..3-20
Table 4-1. Monitor and Area Information for the 15 Urban Study Areas in the Exposure
Modeling and Clinical Study Based Risk Assessment 4-5
Table 4-2. Monitor and Area Information for the 12 Urban Study Areas in the Epidemiology
Based Risk Assessment 4-8
Table 4-3. NOx-only and NOx/VOC Emission Reductions Applied to 15 Urban Study Areas.
Numbers in blue bold text represent the base scenario while numbers in plain font
represent the scenario used in the sensitivity analyses for each urban study area...
4-18
Table 4-4. Summary Statistics based on the Three Fused Seasonal Average Os Surfaces
Based on all CMAQ 12 km Grid Cells 4-35
Table 4-5. Correlation Coefficients between the Three Fused Seasonal Average Os Surfaces
Based on all CMAQ 12 km Grid Cells 4-35
Table 4-6. Correlation Coefficients and Ratios of the 2006-2008 Os Design Values to the
2006-2008 Fused Seasonal Average Os Levels for the CMAQ 12km Grid Cells
Containing Os Monitors 4-37
Table 4-7. Summary of Qualitative Uncertainty Analysis of Key Air Quality Elements in the
Os NAAQS Risk Assessment 4-42
Table 5-1. General Characteristics of the Population Exposure Modeling Domain
Comprising Each Study Area 5-10
Table 5-2. Asthma Prevalence for Children and Adults Estimated by APEX in Each
Simulated Study Area 5-13
Table 5-3. Consolidated Human Activity Database (CHAD) Study Information and Diary-
days Used by APEX 5-14
Table 5-4. Ventilation Equation Coefficient Estimates (hi) and Residuals Distributions (ei)..
5-16
Table 5-5. Microenvironments Modeled, Calculation Method Used, and Variables Included.
5-18
Table 5-6. Number and Percent of All CHAD Diaries Used by APEX and Stratified by Four
U.S. Regions 5-44
Table 5-7. Number of CHAD Diaries for All School-age Children, Stratified by Four U. S.
Regions and Diary Pool Attributes 5-45
Table 5-8. Number and Percent of CHAD Diaries of School-age Children Reporting Time
Spent Outdoors on Days When Daily Maximum Temperature > 84 °F, Stratified
by Four U.S. Regions, Sex, and Day Type 5-46
Table 5-9. Percent of All School-age Children with Daily Maximum 8-hr Exposures at or
above Exposure Benchmarks while at Moderate of Greater Exertion, 2010 Air
Quality Adjusted to Just Meet the Existing Standard, When Using Either 2000 or
2010 Census Population Data 5-70
Table 5-10. Characterization of Key Uncertainties in Historical and Current Exposure
Assessments using APEX 5-73
xvii
-------
Table 5-11. Mean and Maximum Percent of all School-age Children Estimated to Experience
at Least One Daily Maximum 8-hr Average Os Exposure at or above Selected
Health Benchmark Levels while at Moderate or Greater Exertion 5-84
Table 5-12. Mean and Maximum Percent of All School-age Children Estimated to Experience
at Least Two Daily Maximum 8-hr Average Os Exposures at or above Selected
Health Benchmark Levels while at Moderate or Greater Exertion 5-85
Table 5-13. Mean and Maximum Number of People and Person-days with at Least One Daily
Maximum 8-hr Average Os Exposure at or above 60 ppb while at Moderate or
Greater Exertion for All Exposure Study Groups, All 15 Urban Study Areas
Combined 5-87
Table 5-14. Mean and Maximum Number of People and Person-days with at Least One Daily
Maximum 8-hr Average Os Exposure at or above 70 ppb while at Moderate or
Greater Exertion for All Exposure Study Groups, All 15 Urban Study Areas
Combined 5-87
Table 6-1. Estimated Parameters in the MSS Models 6-12
Table 6-2. Age Term Parameters for Application of the 2012 MSS Threshold Model to All
Ages 6-13
Table 6-3. Study-specific Os E-R Data for Lung Function Decrements Based on Correcting
Individual Responses for the Effect on Lung Function of Exercise in Clean Air,
Ages 18-35 6-18
Table 6-4. Ranges of Percents of Population Experiencing One or More and Six or More
Days during the Os Season with Lung Function Decrement (AFEVi) More than
10%. The numbers in this table are the minimum and maximum percents
estimated over all study areas and years, for each age group and scenario 6-26
Table 6-5. Ranges of Percents of Population Experiencing One or More and Six or more
Days during the Os Season with Lung Function Decrement (AFEVi) More than
15%. The numbers in this table are the minimum and maximum percents
estimated over all study areas and years, for each age group and scenario 6-27
Table 6-6. Percents of the General Population and Outdoor Workers (ages 19-35)
Experiencing One or More and Six or More FEVi Decrements > 15% (based on
Atlanta 2006 APEX simulations) 6-29
Table 6-7. Ranges of Percents of all School-aged Children Experiencing One or More Days
during the Os Season with Lung Function Decrement (AFEVi) More than 10 and
15%, based on the E-R Function Approach. The numbers in this table are the
minimum and maximum percents estimated over all study areas and years 6-30
Table 6-8. Comparison of Responses from the MSS Model with Responses from the
Population E-R Method, 2006 Existing Standard Air Quality Scenario, Ages 5 to
18 6-31
Table 6-9. Comparison of MSS Model and E-R Model of Previous Reviews for Atlanta, Mar
l-Oct30, 2006, ages 18-35 6-33
Table 6-10. Comparison of MSS Model and E-R Model of Previous Reviews for Los Angeles,
Jan 1-Dec 31,2006, ages 18-35 6-33
Table 6-11. Comparison of Responses from the MSS 2010 Model with Responses from
McDonnell etal. (1985) 6-37
xvin
-------
Table 6-12. Percents of the Population by Age Group with One or More Days during the Os
Season with Lung Function (FEVi) Decrements more than 10, 15, and 20%
(Atlanta 2006 base case), MSS Threshold Model, Ambient Monitor Data 6-38
Table 6-13. Percents of the Population by Age Group with One or More Days during the Os
Season with Lung Function (FEVi) Decrements more than 10, 15, and 20%
(Atlanta 2006 base case), MSS No-Threshold Model, Ambient Monitor Data
6-38
Table 6-14. MSS Threshold Model Estimated Parameters with Confidence Intervals 6-40
Table 6-15. Sensitivity Analysis of the Inter-individual Variability Term U using the MSS
Threshold Model. Percents of the population aged 5 to 18 with one or more and 6
or more days during the Os season with lung function (FEVi) decrements more
than 10, 15, and 20% (Atlanta 2006 base case, ambient monitor data) 6-42
Table 6-16. Sensitivity Analysis of the Intra-individual Variability Term 8 using the MSS
Threshold Model. Percents of the population aged 5 to 18 with one or more and 6
or more days during the Os season with lung function (FEVi) decrements more
than 10, 15, and 20% (Atlanta 2006 base case, monitors air quality) 6-44
Table 6-17. Convergence results for the Atlanta 2006 base case with 200,000 simulated
individuals. Percents of the population by age group with one or more days (and
six or more days) during the Os season with lung function (FEVi) decrements
more than 10, 15, and 20%. Minimum and maximum values and ranges over 40
APEX runs 6-45
Table 6-18. Comparison of 2000 vs. 2010 Census Demographics. Percents of all School-aged
Children (ages 5 to 18) with One or More Days during the Os Season with Lung
Function (FEVi) Decrements more than 10, 15, and 20% (7 study areas, 2010
existing standard air quality) 6-47
Table 6-19. Comparison of 2000 vs. 2010 Census Demographics. Changes from using 2000
Demographics to using 2010 Demographics of Percents of all School-aged
children (ages 5 to 18) with One or More Days during the Os Season with Lung
Function (FEVi) Decrements more than 10, 15, and 20% (7 study areas, 2010
existing standard air quality) 6-47
Table 6-20. Summary of Qualitative Uncertainties of Key Modeling Elements in the Os Lung
Function Risk Assessment 6-49
Table 7-1. Information on the 12 Urban Study Areas in the Risk Assessment 7-14
Table 7-2. Overview of Epidemiological Studies Used in Specifying C-R Functions 7-24
Table 7-3. CBSA-based Study Areas with Multiple Effect Estimates from the Smith et al.
(2009) Study 7-28
Table 7-4. Summary of Qualitative Uncertainty Analysis of Key Modeling Elements in the
Os NAAQS Risk Assessment 7-42
Table 7-5. Specification of the Core and Sensitivity Analyses (air quality simulation) 7-47
Table 7-6. Specification of the Core and Sensitivity Analyses (alternative C-R function
specification) 7-47
Table 7-7. Short-Term Os-attributable All-Cause Mortality, 2007 and 2009 Air Quality.
Smith et al. (2009) C-R Functions, Os season, CBSA-based study area, no
threshold 7-53
xix
-------
Table 7-8. Percent of Total All-Cause Mortality Attributable to Os and Percent Change in
Os-Attributable Risk, 2007 and 2009 Air Quality. Smith et al. (2009) C-R
functions, Os season, CBSA-based study area, no threshold 7-54
Table 7-9. Short-Term Os-attributable Morbidity Counts, Percent of Baseline and Reduction
in Ch-attributable Risk, Respiratory-Related Hospital Admissions, 2007 and 2009
Air Quality 7-58
Table 7-10. Short-Term Os-attributable Morbidity Counts, Percent of Baseline and Reduction
in Cb-attributable Risk, Emergency Department Visits, 2007 and 2009 Air
Quality 7-59
Table 7-11. Short-Term Os-attributable Morbidity Counts, Percent of Baseline and Reduction
in Os-attributable Risk, Asthma Exacerbations, 2007 and 2009 Air Quality.... 7-60
Table 7-12. Long-Term Os-attributable Respiratory Mortality, 2007 and 2009 Air Quality.
Jerrett et al. (2009) C-R functions, CBSA-based study area, no threshold 7-62
Table 7-13. Long-Term Os-attributable Respiratory Mortality Percent of Baseline Incidence
and Percent Reduction in Os-attributable Risk, 2007 and 2009 Air Quality. Jerrett
et al. (2009) C-R functions, CBSA-based study area, no threshold 7-63
Table 7-14. Sensitivity Analysis for Long-Term Os-attributable Respiratory Mortality,
Alternative C-R Function Specification (regional effect estimates), 2009 Air
Quality. Percent of baseline all-cause mortality and change in Os-attributable risk,
Jerrett etal. (2009), Os season 7-82
Table 7-15. Sensitivity Analysis for Long-Term Os-attributable Respiratory Mortality,
Alternative C-R Function Specification (national Os-only effect estimates), 2009
Air Quality. Percent of baseline all-cause mortality and change in Os-attributable
risk, Jerrett et al. (2009), Os season 7-83
Table 8-1. Estimated Annual Os-related Premature Mortality in 2007 Associated with 2006-
2008 Average Os Concentrations (95 percent confidence interval) 8-7
Table 8-2. Mean, Median, 2.5th Percentile, and 97.5th Percentile of the Estimated Percentage
of Mortality Attributable to Ambient Os for all 3,109 Counties in the Continental
U.S 8-9
Table 8-3. Sensitivity of Estimated Os-attributable Premature Deaths to the Application of
the 5th lowest and 5th highest City-specific Risk estimates found by Smith et al.
(2009) and Zanobetti and Schwartz (2008) to the Grid Cells that do not
correspond to the Cities Included in Those Studies 8-16
Table 8-4. Sensitivity of Estimated Os-attributable Premature Deaths to the Application of
Smith et al. (2009) Regional-prior Bayes-shrunken City-specific and Regional-
average Effect Estimates, as Compared with the National-prior Bayes-shrunken
City-specific and National-average Effect Estimates as Done for the Main
Results 8-16
Table 8-5. Sensitivity of Estimated Os-attributable Premature Deaths to the Application of
Jerrett et al. (2009) Effect Estimates with and without a Low Concentration
Threshold 8-19
Table 8-6. Data Sources for Os Risk-related Attributes 8-25
Table 8-7. Summary Statistics for Selected Os Risk-related Attributes 8-28
Table 8-8. Broad Regional Annual Trends of Concurrent Os Concentrations and Emissions
of NOx and VOCs over years 2000-2011 8-53
xx
-------
Table 9-1. Summary of Urban Scale Os-Exposure Risk Across Alternative Os Standard
Levels 9-3
Table 9-2. Area and Monitoring Information for the 15 Urban Study Areas 9-5
Table 9-3. General Patterns in Seasonal (May-Sept) Mean of Daily Maximum 8-hr Cb
Concentrations after Adjusting to Meet Existing and Alternative Standards 9-7
xxi
-------
This page left intentionally blank
xxn
-------
LIST OF ACRONYMS/ABBREVIATIONS
AER
AHRQ
APEX
AQI
AQS
ATUS
BenMAP
BRFSS
BSA
CAA
CASAC
CASTNET
CDC
CDF
CH4
CHAD
CI
CMAQ
CO
CO2
COPD
C-R
DEARS
DLW
ED
EE
EGU
EPA
ER
air exchange rate
Agency for Healthcare Research and Quality
Air Pollution Exposure Model
Air Quality Index
Air Quality System
American Time Use Survey
Benefits Mapping and Analysis Program
Behavioral Risk Factor Surveillance System
body surface area
Clean Air Act
Clean Air Science Advisory Committee
EPA's Clean Air Status and Trends Network
Center for Disease Control and Prevention
cumulative distribution functions
methane
Consolidated Human Activity Database
confidence interval
Community Multi-scale Air Quality
carbon monoxide
carbon dioxide
chronic obstructive pulmonary disease
Concentration Response (function)
Detroit Exposure and Aerosol Research Study
doubly-labeled water
emergency department
energy expenditure
electric generating unit
U.S. Environmental Protection Agency
emergency room
xxin
-------
E-R
eVNA
EVR
FEM
FEVi
FRM
FVC
HA
HDDM
HNO3
HO2
HREA
HUCP
IPCC
IRP
ISA
I/O
LML
MATS
ME
METs
MSA
MSS
MT
NAAQS
NCDC
NHANES
NEI
NO
NO2
NOx
Exposure Response (function)
enhanced Voronoi Neighbor Averaging
equivalent ventilation rate
Federal Equivalent Method
one-second forced expiratory volume
Federal Reference Method
forced vital capacity
hospital admissions
Higher-order Decoupled Direct Method
nitric acid
hydroperoxy radical
Health Risk and Exposure Assessment
Healthcare Cost and Utilization Program
Intergovernmental Panel on Climate Change
Integrated Review Plan
Integrated Science Assessment
indoor to outdoor concentration ratio
lowest measured level
Modeled Attainment Test Software
microenvironment
metabolic equivalents of work
Metropolitan Statistical Area
McDonnell-Stewart-Smith model
metric ton
National Ambient Air Quality Standards
National Climatic Data Center
National Health and Nutrition Examination Survey
National Emissions Inventory
nitric oxide
nitrite
nitrogen oxides
xxiv
-------
NRC
O3
OAQPS
OMB
OH
PA
PDI
PI
PM
ppb
ppm
PRB
REA
RR
SAB
SEDD
SES
SID
SO2
STE
TSD
TRIM Expo
VE
VNA
VO2
voc
VQ
WHO
WREA
National Research Council
Ozone
Office of Air Quality Planning and Standards
Office of Management and Budget
hydroxyl radical
Policy Assessment
pain on deep inspiration
posterior interval
particulate matter
parts per billion
parts per million
Policy Relevant Background
Risk and Exposure Assessment
relative risk
Science Advisory Board
State Emergency Department Databases
socioeconomic status
State Inpatient Databases
sulfur dioxide
stratosphere-troposphere exchange
technical support document
Total Risk Integrated Methodology Inhalation Exposure
ventilation rate
Voronoi Neighbor Averaging
oxygen consumption rate
volatile organic compound
ventilatory equivalent
World Health Organization
Welfare Risk and Exposure Assessment
xxv
-------
This page left intentionally blank
xxvi
-------
1 INTRODUCTION
The U.S. Environmental Protection Agency (EPA) is presently conducting a review of
the national ambient air quality standards (NAAQS) for ozone (Os), and related photochemical
oxidants. The NAAQS review process includes four key phases: planning, science assessment,
risk/exposure assessment, and policy assessment/rulemaking.1 This process and the overall plan
for this review of the Os NAAQS is presented in the Integrated Review Plan for the Ozone
National Ambient Air Quality Standards (IRP; U.S. EPA, 201 la). The IRP additionally presents
the schedule for the review; identifies key policy-relevant issues; and discusses the key scientific,
technical, and policy documents. These documents include an Integrated Science Assessment
(ISA), Risk and Exposure Assessments (REAs), and a Policy Assessment (PA). This Health
REA (HREA) is one of the two quantitative REAs developed for the review by the EPA's Office
of Air Quality Planning and Standards (OAQPS); the second is a Welfare REA (WREA). This
HREA focuses on assessments to inform consideration of the review of the primary (health-
based) NAAQS for O3.
The existing primary (health-based) NAAQS for Os is set at a level of 75 ppb (0.075
ppm), based on the annual fourth-highest daily maximum 8-hour average concentration,
averaged over three years, and the secondary standard is identical to the primary standard (73 FR
16436). Initial nonattainment area designations have been made for 46 areas in the U.S. with
ambient Os concentrations exceeding the existing standard (77 FR 30160). For illustrative
purposes, Figure 1-1 provides the locations of nonattainment areas and their respective
classifications2 and includes 227 counties with an estimated 2010 population of just over 123
million people.
1 For more information on the NAAQS review process see http://www.epa.gov/ttn/naaqs/review.html.
2 Classification thresholds for the 8-hr design value are as follows: marginal (76 to <86 ppb), moderate (86 to <100
ppb), serious (100 to <113 ppb), severe-15 (113 to <119 ppb), severe-17 (119 to <175 ppb), and extreme (>175
ppb). Note that a few nonattainment areas were reclassified to higher classifications following a voluntary request
by the particular state (77 FR 30160).
1-1
-------
8-Hour Ozone Nonattainment Areas (2008 Standard)
12/05/2013
Nonattainment areas are indicated by color,
When only a portion of a county is shown in
color, it indicates that only that part of the
county is within a rvonattainment area boundary.
8-hour Ozone Classification
• Extreme
"3 Severe 15
Z] Serious
= Moderate
LZI Marginal
Figure 1-1. Nonattainment Area Classifications Based on the Existing Os NAAQS.
(Source: http://www.epa.gov/airquality/greenbook/).
The EPA initiated the current review of the Cb NAAQS on September 29, 2008, with an
announcement of the development of an Os ISA and a public workshop to discuss policy-
relevant science to inform EPA's integrated plan for the review of the Os NAAQS (73 FR
56581). Discussions at the workshop, held on October 29-30, 2008, informed identification of
key policy issues and questions to frame the review of the Os NAAQS. Drawing from the
workshop discussions, the EPA developed a draft and then final IRP (U.S. EPA, 201 la).3 In
early 2013, the EPA completed the Integrated Science Assessment for Ozone and Related
Photochemical Oxidants (U.S. EPA, 2013). The ISA provides a concise review, synthesis and
evaluation of the most policy-relevant science to serve as a scientific foundation for the review
of the NAAQS. The scientific and technical information in the ISA, including that newly
available since the previous review on the health effects of Os includes information on exposure,
physiological mechanisms by which Os might adversely impact human health, an evaluation of
the toxicological and controlled human exposure study evidence, and an evaluation of the
epidemiological evidence, including information on reported ambient concentration-response (C-
3 On March 30, 2009, EPA held a public consultation with the CASAC Ozone Panel on the draft IRP. The final IRP
took into consideration comments received from CASAC and the public on the draft plan as well as input from
senior Agency managers.
1-2
-------
R) relationships for Os-related morbidity and mortality associations, and also includes
information on potentially at-risk populations and lifestages.4
This HREA is a concise presentation of the conceptual model, scope, methods, key
results, observations, and related uncertainties associated with the quantitative analyses
performed. This HREA builds upon the health effects evidence presented and assessed in the
ISA, as well as CAS AC advice (Samet, 2011), and public comments on a scope and methods
planning document for the HREA (here after, "Scope and Methods Plan," U.S. EPA, 201 Ib).
Preparation of this HREA draws upon the ISA and reflects consideration of CAS AC and public
comments on the draft HREAs (Frey and Samet, 2012; Frey, 2014). This HREA is being
released, concurrently with the WREA and PA to inform the proposed Os NAAQS rulemaking.
The PA presents a staff evaluation and conclusions of the policy implications of the key
scientific and technical information in the ISA, and REAs. The PA is intended to help "bridge
the gap" between the Agency's scientific assessments presented in the ISA and REAs, and the
judgments required of the EPA Administrator in determining whether it is appropriate to retain
or revise the NAAQS. The PA integrates and interprets the information from the ISA and REAs
to frame policy options for consideration by the Administrator. In so doing, the PA recognizes
that the selection of a specific approach to reaching final decisions on primary and secondary
NAAQS will reflect the judgments of the Administrator. The development of the various
scientific, technical and policy documents and their roles in informing this NAAQS review are
described in more detail in the PA.
1.1 HISTORY
As part of the last Os NAAQS review completed in 2008, EPA's OAQPS conducted
quantitative risk and exposure assessments to estimate exposures above health benchmarks and
risks of various health effects associated with exposure to ambient Os in a number of urban study
areas, selected to illustrate the public health impacts of this pollutant (U.S. EPA 2007a, U.S.
EPA, 2007b). The assessment scope and methodology were developed with considerable input
from CASAC and the public, with CASAC generally concluding that the exposure assessment
reflected generally accepted modeling approaches, and that the risk assessments were well done,
balanced and reasonably communicated (Henderson, 2006a). The final quantitative risk and
exposure assessments took into consideration CASAC advice (Henderson, 2006a; Henderson,
2006b), and public comments on two drafts of the risk and exposure assessments.
4 The ISA also evaluates scientific evidence for the effects of O3 on public welfare which EPA will consider in its
review of the secondary O3 NAAQS. Building upon the effects evidence presented in the ISA, OAQPS has also
developed a second REA titled Ozone Welfare Effects Risk and Exposure Assessment (U.S. EPA, 2014).
1-3
-------
The exposure and health risk assessment conducted in the last review developed exposure
and health risk estimates for 12 urban study areas across the U.S., based on 2002 to 2004 air
quality data. That assessment provided annual or Cb season-specific exposure and risk estimates
for these years of air quality and for air quality scenarios, simulating just meeting the then-
existing 8-hour Os standard set in 1997 at a level of 0.08 ppm and several alternative 8-hour
standards.5 The strengths and limitations in the assessment were characterized, and analyses of
key uncertainties were presented.
Exposure estimates from the last assessment were used as an input to the risk assessment
for lung function responses (a health endpoint for which exposure-response (E-R) functions were
available from controlled human exposure studies). Exposure estimates were developed for the
general population and population groups including school-age children with asthma as well as
all school-age children. The exposure estimates also provided information on exposures to
ambient Os concentrations at and above specified benchmark levels (referred to as "exposures of
concern"), to provide some perspective on the public health impacts of health effects associated
with Os exposures in controlled human exposure studies that could not be evaluated in the
quantitative risk assessment (e.g., lung inflammation, increased airway responsiveness, and
decreased resistance to infection). For several other health endpoints, Os-related health risk
estimates were generated using C-R relationships reported in epidemiological or field studies,
together with ambient air quality concentrations, baseline health incidence rates, and population
data for the various locations included in the assessment. Health endpoints included in the
assessment based on epidemiological or field studies included: hospital admissions for
respiratory illness in four urban study areas, premature mortality in 12 urban study areas, and
respiratory symptoms in asthmatic children in one urban study area.
Based on the 2006 Air Quality Criteria for Ozone (U.S. EPA, 2006), the Staff Paper
(U.S. EPA, 2007c), and related technical support documents (akin to analyses currently
performed in HREAs), the proposed decision was published in the Federal Register on July 11,
2007 (72 FR 37818). The EPA proposed to revise the level of the primary standard to a level
within the range of 0.075 to 0.070 ppm. Two options were proposed for the secondary standard:
(1) replacing the then-existing standard with a cumulative seasonal standard, expressed as an
index of the annual sum of weighted hourly concentrations cumulated over 12 daylight hours
during the consecutive 3-month period within the Os season with the maximum index value
(W126), set at a level within the range of 7 to 21 ppm-hours, and (2) setting the secondary
standard identical to the revised primary standard. The EPA completed the review with
publication of a final decision on March 27, 2008 (73 FR 16436), revising the level of the 8-hour
5 It is worth noting that the rounding convention applied at that time would allow for design value concentrations to
include values up to 0.084 ppm and still meet the then-existing standard.
1-4
-------
primary Os standard from 0.08 ppm to 0.075 ppm, as the 3-year average of the fourth highest
daily maximum 8-hour average concentration, and revising the secondary standard to be
identical to the revised primary standard.
Following promulgation of the revised Os standard in March 2008, state, public health,
environmental, and industry petitioners filed suit against EPA regarding that final decision.
At EPA's request, the consolidated cases were held in abeyance pending EPA's
reconsideration of the 2008 decision. A notice of proposed rulemaking to reconsider the
2008 final decision was issued by the Administrator on January 6, 2010. Three public
hearings were held. The Agency solicited CAS AC review of the proposed rule on January
25, 2010, and additional CASAC advice on January 26, 2011. On September 2, 2011, the
Office of Management and Budget (OMB) returned the draft final rule on reconsideration to
EPA for further consideration. EPA decided to coordinate further proceedings on its
voluntary rulemaking on reconsideration with this ongoing periodic review, by deferring the
completion of its voluntary rulemaking on reconsideration until it completes its statutorily-
required periodic review. In light of that, the litigation on the 2008 final decision proceeded.
On July 23, 2013, the Court ruled on the litigation of the 2008 decision, denying the
petitioners suit except with respect to the secondary standard, which was remanded to the
Agency for reconsideration. The PA provides additional description of the court ruling with
regard to the secondary standard.
1.2 CURRENT RISK AND EXPOSURE ASSESSMENT: GOALS AND PLANNED
APPROACH
The goals of the current quantitative exposure and health risk assessments are to provide
information relevant to answering questions regarding the adequacy of the existing Os standard
and the potential improvements in public health from meeting alternative standards. To meet
these goals, this assessment provides results from several analyses, including (1) estimates of the
number and percent of people in the general population and in at-risk populations and lifestages
with Os exposures above benchmark levels, while at moderate or greater exertion levels; (2)
estimates of the number and percent of people in the general population and in at-risk
populations and lifestages with impaired lung function resulting from exposures to Os; and (3)
estimates of the potential magnitude of Os-related premature mortality and selected morbidity
health effects in the population, including at-risk populations and lifestages, where data are
available to assess these groups. For each of the analyses, we provide estimates for recent
ambient levels of Os and for air quality conditions simulated to just meet the existing Os standard
and alternative standards.
1-5
-------
In presenting these results, we evaluate the influence of various inputs and assumptions
on the exposure and risk estimates to more clearly differentiate alternative standards that might
be considered, including potential impacts on various at-risk populations and lifestages. We also
evaluate the distribution of risks and patterns of risk reduction and uncertainties in those risk
estimates. In addition, we have conducted an assessment to provide nationwide estimates of the
potential magnitude of premature mortality associated with recent ambient Os concentrations, to
more broadly characterize this risk on a national scale. This assessment includes an evaluation of
the distribution of Os-related health risk across the U.S. and associated influential factors, to
assess the overall representativeness of the risk results generated by our urban study area
analyses.
This current quantitative risk and exposure assessment builds on the approach used and
lessons learned in the last Cb risk and exposure assessment, and focuses on improving the
characterization of the overall confidence in the exposure and risk estimates, including related
uncertainties, by incorporating a number of enhancements, in terms of both the models, methods,
and data used in the analyses. This risk assessment considers a variety of health endpoints for
which, in staff s judgment, there is adequate information to develop quantitative risk estimates
that can meaningfully inform the review of the primary Cb NAAQS.
1.3 ORGANIZATION OF DOCUMENT
The remainder of this document is organized as follows. Chapter 2 provides a conceptual
framework for the risk and exposure assessment, including discussions of Cb chemistry, sources
of Cb precursors, exposure pathways and microenvironments where Cb exposure can be high, at-
risk populations and lifestages, and health endpoints associated with Cb. This conceptual
framework sets the stage for the scope of the risk and exposure assessments. Chapter 3 provides
an overview of the scope of the quantitative risk and exposure assessments, including a summary
of the previous risk and exposure assessments, and an overview of the current risk and exposure
assessments. Chapter 4 discusses the characterization of air quality relevant to the exposure and
risk assessments, including available Cb monitoring data, and important inputs to the risk and
exposure assessments. Chapter 5 describes the inputs, models, and results for the urban-scale
human exposure assessment, and discusses the literature on exposure to Cb, the scope of the
exposure assessment, exposure model inputs, sensitivity and uncertainty evaluations, and
estimation of results. Chapter 6 describes the estimation of urban-scale health risks based on
application of the results of controlled human exposure studies combined with the Chapter 5
exposure results, including discussions of health endpoint selection, approaches to calculating
risk, and exposure-based risk results. Chapter 7 describes the estimation of health risks in
selected urban study areas based on application of the results of observational epidemiology
1-6
-------
studies, including discussions of air quality characterizations, epidemiological-based risk model
inputs, variability and uncertainty, and results. Chapter 8 describes the national-scale
epidemiological-based risk characterization and an urban study area representativeness analysis.
Chapter 9 provides an integrative discussion of the exposure and risk estimates generated in the
analyses drawing on the results of the analyses based on both controlled human exposure and
epidemiological studies, and incorporating considerations from the national-scale risk
characterization.
1-7
-------
1.4 REFERENCES
Frey, C. and J. Samet. 2012. CASAC Review oftheEP'A's Health Risk and Exposure Assessment
for Ozone (First External Review Draft - Updated August 2012) and Welfare Risk and
Exposure Assessment for Ozone (First External Review Draft - Updated August 2012). U.S.
Environmental Protection Agency Science Advisory Board. EPA-CASAC-13-002.
Frey, C. 2014. CASAC Review of the EPA's Health Risk and Exposure Assessment for Ozone
(Second External Review Draft-February, 2014). U.S. Environmental Protection Agency
Science Advisory Board. EPA-C AS AC-14-005.
Henderson, R. 2006a. Clean Air Scientific Advisory Committee (CASAC) Ozone Review
Panel's Consultation on EPA's First Draft Ozone Staff Paper, Risk Assessment, and
Exposure Assessment Documents. (February, 2006). U.S. Environmental Protection
Agency Science Advisory Board. EPA-C AS AC-CON-06-003.
Henderson, R. 2006b. Clean Air Scientific Advisory Committee's (CASAC) Peer Review of the
Agency's 2ndDraft Ozone Staff Paper. U.S. Environmental Protection Agency Science
Advisory Board. EPA-C AS AC-07-001.
Samet, J. 2011. Consultation on EPA's Ozone National Ambient Air Quality Standards: Scope
and Methods Plan for Health Risk and Exposure Assessment (April 2011) and Ozone
National Ambient Air Quality Standards: Scope and Methods Plan for Welfare Risk and
Exposure Assessment (April 2011). EPA-CASAC-11-008.
U.S. EPA. 2006. Air Quality Criteria for Ozone and Related Photochemical Oxidants (Final).
Washington, DC: Office of Research and Development. (EPA document number
EPA/600/R-05/004aF-cF).
.
U.S. EPA. 2007a. O 3 Population Exposure Analysis for Selected Urban Areas. Research
Triangle Park, NC: Office of Air Quality Planning and Standards. (EPA document
number EPA-452/R-07-010).
.
U.S. EPA. 2007b. Os Health Risk Assessment for Selected Urban Areas. Research Triangle Park,
NC: Office of Air Quality Planning and Standards. (EPA document number EPA 452/R-
07-009). .
U.S. EPA. 2007c. Review of the National Ambient Air Quality Standards for Ozone: Policy
Assessment of Scientific and Technical Information. OA QPS Staff Paper. Research
Triangle Park, NC: Office of Air Quality Planning and Standards. (EPA document
number EPA-452/R-07-007).
.
U.S. EPA. 201 la. Integrated Review Plan for the Os National Ambient Air Quality Standards.
Research Triangle Park, NC: National Center for Environmental Assessment, Office of
Research and Development and Office of Air Quality Planning and Standards, Office of
Air and Radiation. (EPA document number EPA 452/R-l 1-006).
.
U.S. EPA. 201 Ib. Os National Ambient Air Quality Standards: Scope andMethods Plan for
Health Risk and Exposure Assessment. Research Triangle Park, NC: Office of Air
-------
Quality Planning and Standards. (EPA document number EPA-452/P-11-001).
.
U.S. EPA. 2014. Welfare Risk and Exposure Assessment for Ozone (Final Report). Research
Triangle Park, NC: EPA Office of Air and Radiation. (EPA document number EPA-
452/R-14-005a).
1-9
-------
This page left intentionally blank
1-10
-------
2 OVERVIEW OF EXPOSURE AND RISK ASSESSMENT DESIGN
In this chapter, we summarize our framework for assessing exposures to Os and the
associated risks to human populations. Figure 2-1 provides an overview of the general design of
this exposure and risk assessment, which includes air quality characterization, review of relevant
scientific evidence on health effects, modeling of exposure, modeling of risk, and risk
characterization. Each element identified in the diagram is described in a specific, identified
chapter of this exposure and risk assessment.
In this Os exposure and risk assessment, air quality is characterized primarily by the
combined use of ambient monitoring data available in EPA Air Quality System (AQS), a spatial
interpolation approach (Voronoi Neighbor Averaging, VNA), along with Higher-Order
Decoupled Direct Method (HDDM) capabilities in the Community Multi-scale Air Quality
(CMAQ)1 model. The modeling of personal exposure and estimation of associated risks based on
published results from controlled human exposure studies are implemented using the Air
Pollution Exposure model (APEX; U.S. EPA, 2012 a,b).2 Modeling of population level risks for
health endpoints based on application of results reported in published epidemiological studies, is
implemented using the environmental Benefits Mapping and Analysis Program (BenMAP),3 a
peer reviewed software tool for estimating risks and impacts associated with changes in ambient
air quality (U.S. EPA, 2013a). The overall characterization of risk draws from the results of the
exposure assessment and both types of risk assessment.
The remainder of this chapter includes summary discussions of each of the main elements
of Figure 2-1, including policy-relevant exposure and risk questions (Section 2.1),
characterization of ambient Os, including important sources of Os precursors, and its relation to
population exposures, as well as simulation of just meeting existing and potential alternative Os
standards (Section 2.2), review of health evidence identified in the literature describing
associations with ambient Os (Section 2.3), key components of exposure modeling (Section 2.4),
key components of risk modeling (Section 2.5), and risk characterization (Section 2.6). Specific
details related to the scope of the exposure and risk assessments and how each element will be
addressed in the quantitative exposure and risk analysis are provided in Chapter 3.
1 The CMAQ model and associated documentation is available for download at https://www.cmascenter.org/cmaq/.
2 The APEX model and associated documentation is available for download at
http://www.epa.gov/ttn/fera/human_apex.html.
3 The BenMAP model and associated documentation is available for download at http://www.epa.gov/air/benmap/.
2-1
-------
Policy Relevant Exposure and
Risk Questions
(Chapter 2)
Exposure Assessment
APEX
i I
Urban Scale
Assessment of
Individual Exposure
(Chapter 5)
^=-L—
AQS, VNA,
CMAQ-HDDM
Air Quality Characterization
{Chapter 4)
Review of Health Evidence
(Chapter 2)
Risk Assessment
Urban Scale Risk
Analyses Based on
Application of Results
from Controlled
Human Exposure
Studies
(Chapter 6}
BenMAP
Urban Scale Risk
Analyses Based on
Application of Results
from Epidemiological
Studies
(Chapter 7)
National Scale Risk
Burden Based on
Application of Results
from Epidemiological
Studies
(Chapter 8)
Risk Characterization
(Chapter 9)
Figure 2-1. Overview of Exposure and Risk Assessment Design.
2.1 POLICY-RELEVANT EXPOSURE AND RISK QUESTIONS
The first step in the design is to determine the set of policy-relevant exposure and risk
questions that will be informed by the assessment. Consistent with recommendations from the
recent National Research Council report "Science and Decisions: Advancing Risk Assessment"
(NRC, 2009), these exposure and risk assessments have been designed to address the risk
questions identified in the Integrated Review Plan for the Ozone National Ambient Air Quality
Standards (IRP; U.S. EPA, 2011). We have focused on designing the exposure and risk
assessments to inform consideration of those risk-related policy-relevant questions in the
separately developed Os NAAQS Policy Assessment (PA; U. S. EPA, 2014). The risk-related
policy-relevant questions identified in the Integrated Review Plan are related to two main
activities, evaluation of the adequacy of the existing standards and, if appropriate, evaluation of
2-2
-------
potential alternative standards (U.S. EPA, 2011). With regard to evaluation of the adequacy of
the existing standards, the risk-related policy-relevant questions are:
"To what extent do risk and/or exposure analyses suggest that exposures of
concern for Oi-related health effects are likely to occur with existing ambient
levels ofOs or with levels that just meet the Os standard? Are these
risks/exposures of sufficient magnitude such that the health effects might
reasonably be judged to be important from a public health perspective? What are
the important uncertainties associated with these risk/exposure estimates?"
With regard to evaluation of potential alternative standards, the risk-related policy-relevant
questions are:
"To what extent do alternative standards, taking together levels, averaging times
and forms, reduce estimated exposures and risks of concern attributable to Os
and other photochemical oxidants, and what are the uncertainties associated with
the estimated exposure and risk reductions? What conclusions can be drawn
regarding the health protection afforded at-riskpopulations? "
This risk and exposure assessment is designed to inform consideration of these questions
through application of exposure and risk modeling for a set of urban study areas. Exposure and
risk estimates will be generated for recent4 O^ concentrations, O^ concentrations after simulating
just meeting the existing standards, and Qs concentrations after simulating just meeting potential
alternative standards. Careful consideration will be given to addressing variability and
uncertainty in the estimates, and to the degree to which at-risk populations experience exposures
and risks. Exposure modeling is discussed in Chapter 5 (Characterization of Urban-Scale Human
Exposure), while risk modeling is discussed in Chapter 6 (Characterization of Urban-scale
Health Risk Based on Controlled Human Exposure Studies) and Chapter 7 (Characterization of
Urban-Scale Health Risk Based on Epidemiological Studies). Chapter 8 (Characterization of
National-scale Mortality Risk Based on Epidemiological Studies and Urban-Scale
Representativeness Analysis) provides a national-scale assessment of risks under recent Os
concentrations to provide context for the urban-scale analyses and to help characterize the
representativeness of the urban-scale analyses.
In order to inform consideration of the risk-related policy-relevant questions, the first step
for all of the exposure and risk analyses is simulation of meeting the existing and alternative
4 To represent recent air quality, the exposure and risk assessments conducted in Chapters 5 to 7 used ambient O3
concentrations from 2006-2010. For risk calculations performed in Chapter 8, data from years 2006-2008 were
used.
2-3
-------
standards. To do this, recent air quality measurements of Os are adjusted such that they mimic a
realistic and general atmospheric response to changes in precursor emissions for the specific
urban study area and so that they just meet the existing and alternative standard levels.
Conceptually, there is an almost infinite set of combinations of precursor emissions reductions
that will result in just meeting the existing or alternative standards. The specific combinations of
reductions that might actually be implemented are not relevant for the exposure and risk
analyses, as those will result from the implementation processes which follow the establishment
of a standard. However, it is appropriate to ask the question of how the patterns of ambient Os on
multiple temporal scales (hourly, daily, monthly, seasonally) and across each urban study area,
may respond to precursor emissions reductions that result in meeting the existing and potential
alternative standards, and how these different patterns of Os could affect the exposure and risk
results. The answers to these questions are critical inputs to the exposure and risk analyses.
Consideration of the available methods for simulating just meeting existing and alternative
standards is discussed in Chapter 4 (Air Quality Characterization).
Analyses presented in this document to inform the policy-relevant risk questions
regarding potential alternative standards, are focused on alternative levels for an 8-hour standard.
Other elements of the standard (indicator, averaging time, and form),5 are addressed in the Os PA
as part of the overall evaluation of the health protection afforded by the primary Os standards.
With regard to potential alternative levels for an 8-hour Cb standard, the quantitative risk
assessment evaluates the range of levels in 5 ppb increments from 60 to 70 ppb. These levels
were selected based on the evaluations of the evidence provided in the PA, which received
support from the CASAC in their advisory letter (Frey and Samet, 2012). For a subset of urban
study areas, we also evaluated a standard level of 55 ppb, consistent with recommendations from
CASAC to also give consideration to evaluating a level somewhat below 60 ppb. Thus, for most
areas, we evaluate exposures and risks for potential alternative standard levels of 70, 65, and 60
ppb. Some additional analyses were also included for evaluation of exposures and risks for a
potential alternative 8-hour standard level of 55 ppb.
2.2 AIR QUALITY CHARACTERIZATION
In order to address the policy-relevant questions discussed in Section 2.1, the first step is
characterizing Os concentrations relevant to estimation of exposure and risk. This requires
characterization of recent Cb concentrations, Os concentrations after simulating just meeting the
5 The "form" of a standard defines the air quality statistic that is compared to the level of the standard in determining
whether an area attains the standard. The existing form of the 8-hour O3 standard is the 4th highest daily maximum
8-hour average O3, averaged over 3 years. The "indicator" of a standard defines the chemical species or mixture
that is to be measured in determining whether an area attains the standard.
2-4
-------
existing standards, and Os concentrations after simulating just meeting potential alternative
standards. This section provides conceptual information on Os formation and responsiveness of
Os to changes in precursor emissions that inform the simulations of just meeting existing and
alternative standards.
2.2.1 Ozone Chemistry and Response to Changes in Precursor Emissions
Ozone occurs naturally in the stratosphere where it provides protection against harmful
solar ultraviolet radiation, and it is formed closer to the surface in the troposphere from precursor
emissions from both natural and anthropogenic sources. Ozone is created when its two primary
precursors, volatile organic compounds (VOC) and oxides of nitrogen (NOx), combine in the
presence of sunlight. Volatile organic compounds and NOx are, for the most part, emitted
directly into the atmosphere. Carbon monoxide (CO) and methane (CH4) can also be important
for Os formation (U.S. EPA, 2013b, section 3.2.2).
Rather than varying directly with emissions of its precursors, Os changes in a nonlinear
fashion with the concentrations of its precursors. NOx emissions lead to both the formation and
destruction of Os, depending on the local concentrations of NOx, VOC, and radicals such as the
hydroxyl (OH) and hydroperoxy (IKh) radicals. In areas dominated by fresh emissions of NOx,
these radicals are removed via the production of nitric acid (HNOs), which lowers the Os
formation rate. In addition, the depletion of Os by reaction with NO is called "titration" and is
often found in downtown metropolitan areas, especially near busy streets and roads, and in
power plant plumes. This "titration" results in Os concentrations that can be much lower than in
surrounding areas. Titration is usually confined to areas close to strong NOx sources, and the
NO2 formed can lead to Os formation later and further downwind. Consequently, Os response to
reductions in NOx emissions is complex and may include Os decreases at some times and
locations and increases of Os in other times and locations. In areas with low NOx concentrations,
such as those found in remote continental areas and rural and suburban areas downwind of urban
centers, the net production of Os typically varies directly with NOx concentrations, and increases
with increasing NOx emissions.
In general, the rate of Os production is limited by either the concentration of VOCs or
NOx, and Os formation, and using these two precursors relies on the relative sources of OH and
NOx. When OH radicals are abundant and are not depleted by reaction with NOx and/or other
species, Os production is referred to as being "NOx-limited" (U.S. EPA, 2013b, section 3.2.4). In
this situation, Os concentrations are most effectively reduced by lowering NOx emissions, rather
than lowering emissions of VOCs. When the abundance of OH and other radicals is limited
either through low production or reactions with NOx and other species, Os production is
sometimes called "VOC-limited" or "radical limited" or "NOx-saturated" (Jaegle et al., 2001),
2-5
-------
and Os is most effectively reduced by lowering VOCs. However, even in NOx-saturated
conditions, very large decreases in NOx emissions can cause the Os formation regime to become
NOx-limited. Consequently, reductions in NOx emissions (when large), can make further
emissions reductions more effective at reducing Os. Between the NOx-limited and NOx-saturated
extremes there is a transitional region, where Os is less sensitive to marginal changes in either
NOx or VOCs. In rural areas and downwind of urban areas, Os production is generally NOx-
limited. However, across urban areas with high populations, conditions may vary. For contrast,
while data from monitors in Nashville, TN, suggest NOx-limited conditions exist there, data from
monitors in Los Angeles suggest NOx-saturated conditions (U.S. EPA, 2013b, Figure 3-3).
2.2.2 Sources of Ozone and Ozone Precursors
Ozone precursor emissions can be divided into anthropogenic and natural source
categories, with natural sources further divided into biogenic emissions (from vegetation,
microbes, and animals), and abiotic emissions (from biomass burning, lightning, and geogenic
sources). The anthropogenic precursors of Os originate from a wide variety of stationary and
mobile sources.
In urban areas, both biogenic and anthropogenic VOCs, as well as CO, are important for
Os formation. Hundreds of VOCs are emitted by evaporation and combustion processes from a
large number of anthropogenic sources. Based on the 2005 national emissions inventory (NEI),
solvent use and highway vehicles are the two main anthropogenic sources of VOCs, with
roughly equal contributions to total emissions (U.S. EPA, 2013b, Figure 3-2). The emissions
inventory categories of "miscellaneous" (which includes agriculture and forestry, wildfires,
prescribed burns, and structural fires), and off-highway mobile sources are the next two largest
contributing emissions categories with a combined total of over 5.5 million metric tons a year
(MT/year).
On the U.S. and global scales, emissions of VOCs from vegetation are much larger than
those from anthropogenic sources. Emissions of VOCs from anthropogenic sources in the 2005
NEI were -17 MT/year (wildfires constitute -1/6 of that total), compared to emissions from
biogenic sources of 29 MT/year. Vegetation emits substantial quantities of VOCs, such as
isoprene and other terpenoid and sesqui-terpenoid compounds. Most biogenic emissions occur
during the summer because of their dependence on temperature and incident sunlight. Biogenic
emissions are also higher in southern and eastern states than in northern and western states for
these reasons and because of species variations.
Anthropogenic NOx emissions are associated with combustion processes. Based on the
2005 NEI, the three largest sources of NOx are on-road and off-road mobile sources (e.g.,
construction and agricultural equipment), and electric power generation plants (EGUs) (U.S.
2-6
-------
EPA, 2013b, Figure 3-2). Emissions of NOx therefore are highest in areas having a high density
of power plants and in urban areas having high traffic density. However, it is not possible to
make an overall statement about their relative impacts on Os in all local areas because EGUs are
sparser than mobile sources, particularly in the west and south and because of the nonlinear
nature of Os chemistry discussed in Section 2.2.1.
Major natural sources of NOx in the U.S. include lightning, soils, and wildfires. Biogenic
NOx emissions are generally highest during the summer and occur across the entire country,
including areas where anthropogenic emissions are low. It should be noted that uncertainties in
estimating natural NOx emissions are much larger than for anthropogenic NOx emissions.
Ozone concentrations in a region are maintained by a balance between photochemical
production and transport of Os into the region; and loss of Os by chemical reactions, deposition
to the surface and transport out of the region. Ozone transport occurs on many spatial scales
including local transport between cities, regional transport over large regions of the U.S. and
international/long-range transport. In addition, Os is also transfered into the troposphere from the
stratosphere, which is rich in Os, through stratosphere-troposphere exchange (STE).
Stratosphere-troposphere exchange occurs in tropopause "foldings" that occur behind cold
fronts, bringing stratospheric air with them (U.S. EPA, 2013b, section 3.4.1.1). Contributions to
Os concentrations in an area from STE are defined as being part of background Os (U.S. EPA,
2013b, section 3.4).
2.2.3 Simulation of Just Meeting the Existing and Alternative Standards
Conceptually, simulation of meeting existing and alternative standards should reflect the
physical and chemical processes of Os formation in the atmosphere and estimate how hourly
values of Os at each monitor in an urban area would change in response to reductions in
precursor emissions, allowing for nonlinearities in response to emissions reductions and allowing
for nonlinear interactions between reductions in NOx and VOC emissions. For this assessment,
we have employed a sophisticated approach (Higher-Order Decoupled Direct Method (HDDM)
capabilities in the Community Multi-scale Air Quality (CMAQ) model) to conduct simulations
of hourly Os responses to reductions in precursor emissions. This modeling incorporates all
known emissions, including emissions from both natural and anthropogenic sources within and
outside of the U.S. By using the model-adjustment methodology we are able to more realistically
simulate the temporal and spatial patterns of Os response to precursor emissions. We chose to
simulate just meeting the existing and alternative standards in most cases, by decreasing U.S.
anthropogenic emissions of NOx exclusively, though in a few instances, proportional decreases
in both NOx and VOC were applied. This approach, as observed by our modeling, was most
effective in achieving the desired design values of each air quality scenario and while also
2-7
-------
avoiding any suggestion that we are approximating a specific emissions control strategy that a
state or urban area might adopt to meet a standard. These analyses allow us to apply an
adjustment to ambient Os measurements in the urban study areas, to better represent how air
quality concentrations at each monitor would change to meet the existing and alternative
standard levels. Furthermore, background Os concentrations (e.g. that arising from non-
anthropogenic emissions) is accounted for by the modeling approach used to simulate each of the
air quality scenarios, though background Os concentrations are not separately used in the
calculation of or in apportioning exposures or health risks. The details of the specific approach
used to simulating air quality that just meets the existing and alternative standards, are discussed
in greater detail in Chapter 4 and in the Chapter 4 appendices.
It is fundamentally a policy decision, as to which sources of precursor emissions are most
appropriate to decrease to simulate just meeting existing and alternative Os standards. In
addressing the policy-relevant questions regarding the evaluation of alternative standards,
consistent with previous reviews of the Os standards, this analysis is focused on simulating
reductions in risk associated with precursor emissions originating from anthropogenic sources
within the U.S. In doing so, we recognize that the CAA provides mechanisms primarily for
reducing emissions from U.S. emissions sources. As such, we estimate changes in exposure and
risks likely to result from just meeting alternative standards relative to just meeting the existing
standards, by simulating changes in atmospheric concentrations that represent atmospheric
response to reductions in U.S. anthropogenic emissions. However, we recognize that, in this
approach, we are simulating air quality that just meets existing and alternative standard levels,
based on the selected-year air quality concentrations and the chemical environment and
emissions in these years. We have not mimicked the future-year atmospheric conditions and
emissions inventory as would be done for the implementation process.
In addition, while it is possible to decrease Os concentrations using decreases in either
NOx or VOC or both NOx and VOC, the specific combination of the reductions in those
emissions is a policy decision, with recognition that atmospheric chemistry considerations will
make NOx and VOC decreases more or less effective in specific urban areas, depending on the
degree to which Os formation is NOx or VOC limited. As discussed above, in most locations,
decreases in NOx are the most effective means to decrease ambient Os concentrations. However,
in some downtown urban areas, Os formation is VOC-limited, and therefore smaller decreases in
NOx will not decrease Os.
2.3 CONSIDERATION OF HEALTH EVIDENCE
A critical input for both the exposure and risk assessments is the health evidence
summarized in the Integrated Science Assessment (ISA; U.S. EPA, 2013b). This health evidence
2-8
-------
provides the basis for evaluating the significance of exposures to Os, by informing health
benchmarks for estimating exposures of concern. The evidence also provides the basis for
selecting health endpoints that will be modeled in the risk assessment. This evidence includes
controlled human exposure studies and observational epidemiology studies. The health evidence
is also the source of the specific studies that are used to develop exposure-response (E-R) and
concentration-response (C-R) functions, used in the risk assessment. Finally, the health evidence
provides information on at-risk populations to guide the selections of study populations used in
the exposure and risk assessments. The following subsections summarize key conceptual aspects
regarding exposures of concern, health endpoints, E-R and C-R functions, and at-risk
populations.
2.3.1 Exposures of Concern
The Os ISA identifies health effects associated with exposures to varying concentrations
of Os. However, not all of the evidence is suitable for evaluation in a quantitative risk
assessment. Estimating exposures to ambient Os concentrations at and above benchmark levels
where health effects have been observed in studies provides a perspective on the public health
impacts of Os-related health effects that have been demonstrated in human clinical and
toxicological studies but cannot currently be evaluated in quantitative epidemiological-based
health risk assessments, such as lung inflammation, increased airway responsiveness, and
decreased resistance to infection.
To inform the selection of benchmark levels for Os exposure, it is appropriate to consider
the evidence from clinical studies which have evaluated adverse health responses in study
subjects typically exposed to controlled levels of Os over 6.6 hours while also engaged in
moderate quasi-continuous exercise (Ch ISA, section 6.2.1.1). There is substantial evidence from
these controlled human exposure studies demonstrating a range of Os-related effects including
lung inflammation and airway responsiveness in healthy individuals at an exposure level of 80
ppb. There is additional evidence that asthmatics have larger and more serious effects than
healthy people at 70 ppb, as well as a substantial body of epidemiological evidence of
associations with Ch levels that extend well below 80 ppb. There is a more limited set of
evidence based on clinical studies of healthy individuals exposed at 60 ppb in which Os-related
effects have been observed. This is the lowest level at which any Os-related health effects have
been observed in clinical studies of healthy individuals (U.S. EPA, 2013b, section 6.2.1.1).
Thus, estimated exposure concentrations of 60 ppb, 70 ppb, and 80 ppb time-averaged
over 8 hours in simulated individuals are used in our exposure assessment to indicate exposure
2-9
-------
levels of concern (i.e., benchmark levels) for a range of potential health effects in healthy and at-
risk populations exposed to Os.6
2.3.2 Health Risk Endpoints
The Os ISA identifies a wide range of health outcomes associated with short-term
exposure to ambient Os, including an array of morbidity effects as well as premature mortality.
The ISA also identifies several morbidity effects and some evidence for premature mortality
associated with longer-term exposures to Os. In identifying health endpoints for risk assessment,
we have focused on endpoints that pertain to at-risk populations, have public health significance,
and for which information is sufficient to support a quantitative C-R relationship, in the case of
epidemiological studies, or E-R relationship, in the case of controlled human exposure studies.7
In considering such adverse health effect endpoints for Os, we draw from two types of
studies: controlled human exposure and epidemiological studies. Each study type informs our
characterization of Os risk and can do so in different ways. Estimates of risk based on results of
controlled human exposure studies are valuable because they provide clear evidence of the
detrimental effects of controlled (and measured) exposures to Os over multiple hours on lung
function at moderate levels of exertion. Results of these studies can be applied to modeled
estimates of population exposure to provide insights into population exposure characteristics,
including types of activity patterns and microenvironments that are associated with high levels of
risk. Controlled human exposure studies, however, cannot directly provide relationships for
endpoints such as premature death or hospitalizations, focusing more on intermediate biological
endpoints including inflammatory, blood, neurological, cardiovascular, and respiratory
biomarkers or symptoms. Further these studies usually do not include particular at-risk subjects
such as children. Estimates of risk based on C-R functions from observational epidemiology
studies can provide insights on risk for more serious or chronic health endpoints as well as for
people at differing lifestages. For example, epidemiological studies of Os described in the ISA
have evaluated associations between Os and various endpoints including respiratory symptoms,
respiratory-related hospitalizations and emergency department (ED) visits, and premature
6 Note that the existing and potential alternative ambient standards that are evaluated using this assessment
approach, regardless of the level, averaging time, and form of the standard are designed to limit exposures based
on these controlled exposure concentrations of 60, 70, and 80 ppb that were evaluated in the clinical studies. This
is not to say that there are not health effects that could occur with exposures within the range of these benchmark
levels (e.g., an 8-hour average exposure of 75 ppb) but that responses occurring at any alternative exposure
concentrations have not been directly assessed in the clinical studies.
7 The distinction between C-R and E-R functions reflects the typical use of ambient concentrations as measured at
monitor locations as surrogates for population exposures in observational epidemiology studies, as compared to
the personal exposures to controlled concentrations of O3 that are typically used in controlled human exposure
studies. Both types of studies are intended to produce an E-R relationship, however, the epidemiology studies are
actually providing a C-R relationship, which captures the E-R relationship with errors in exposure measurement.
2-10
-------
mortality (U.S. EPA, 2013b, sections 6.2.9 and 6.3.4). Epidemiological studies also generally
focus on a population residing in specific area, which may reflect a broad range of
susceptibilities and sensitivities. Controlled human exposure studies typically involve a smaller
number of individuals over a more limited range of health status, in some cases focused on at-
risk populations, such as asthmatics and individuals with chronic obstructive pulmonary disease
(COPD). Lastly, while controlled human exposure studies directly measure the exposures
eliciting the recorded effects, epidemiology studies have not traditionally been based on
observations of personal exposure to ambient Os, relying instead on surrogate measures of
population exposure. Such surrogates are often based on simple averages of ambient Os monitor
observations. Thus, with attention to their differing strengths and limitations, risk analyses based
on each type of study can inform the risk characterization.
The Os ISA makes overall causal determinations based on the full range of evidence
including epidemiological, controlled human exposure, and toxicological studies. Figure 2-1
shows the Os health effects which have been categorized by strength of evidence for causality in
the Os ISA (U.S. EPA, 2013b, chapter 2). The health evidence supporting each of these three
categories considered here are defined briefly:
• Causal: Health effect evidence is sufficient to conclude that there is a causal relationship
with relevant pollutant exposures. That is, the pollutant has been shown to result in health
effects in studies in which chance, bias, and confounding could be ruled out with
reasonable confidence.
• Likely: Evidence is sufficient to conclude that a causal relationship is likely to exist with
relevant pollutant exposures, but important uncertainties remain. That is, the pollutant has
been shown to result in health effects in studies in which chance and bias can be ruled out
with reasonable confidence but potential issues remain.
• Suggestive: Evidence is suggestive of a causal relationship with relevant pollutant
exposures, but is limited.
The Os ISA determined there are causal relationships between short-term exposure to
ambient Os and respiratory effects, including respiratory-related morbidity and mortality and a
likely causal relationship with all-cause total mortality and with cardiovascular effects; the
evidence was concluded to be suggestive of a causal relationship between short-term exposure to
ambient Os and central nervous system effects. The ISA determined to also be a likely causal
relationship between long-term Os exposures and respiratory effects (including respiratory
symptoms, new-onset asthma, and respiratory mortality), and determined the evidence to be
suggestive of causal relationships between long-term Os exposures and total mortality as well as
cardiovascular, reproductive and developmental, and central nervous system effects.
2-11
-------
If]
OJ
l/l
o
Q.
X
OJ
o
LO
Central nervous
system effects
Cardiovascular effects
Total Mortality
Respiratory effects
Suggestive
Likely
IS]
OJ
a
X
OJ
m
O
E
i_
V
tio
O
Cardiovascular effects Respiratory effects
Reproductive and
developmental effects
Central nervous system
effects
Total Mortality
Figure 2-2. Causal Determinations for Os Health Effects.
The ISA identifies several specific respiratory responses to short-term Os exposure that
have been evaluated in controlled human exposure studies (U.S. EPA, 2013b, section 6.2.1).
These include decreased inspiratory capacity, decreased forced vital capacity (FVC) and forced
expiratory volume in 1 second (FEVi); mild bronchoconstriction; rapid, shallow breathing
patterns during exercise; symptoms of cough and pain on deep inspiration (PDI); and pulmonary
inflammation. While such studies document quantitative relationships between short-term Os
exposure and an array of respiratory-related effects, E-R data across a range of concentrations
sufficient for developing quantitative risk estimates are only available for Os-related decrements
in FEVi (U.S. EPA, 2013b, section 6.2.1).
Within the broad category of respiratory morbidity effects, the epidemiology literature
has provided effect estimates for a wide range of health endpoints associated with short-term Os
exposures which we have considered for risk assessment. These health endpoints include lung
function, respiratory symptoms and medication use, respiratory-related hospital admissions, and
emergency department visits. In the case of respiratory symptoms, the evidence is most
consistently supportive of the relationship between short-term ambient Os metrics and
2-12
-------
respiratory symptoms and asthma medication use in children with asthma, but not for a
relationship between Os and respiratory symptoms in children without asthma (U.S. EPA, 2013b,
section 6.2.9). In the case of hospital admissions, there is evidence of associations between short-
term ambient Os metrics and general respiratory-related hospital admissions as well as more
specific asthma-related hospital admissions (U.S. EPA, 2013b, section 6.2.7.2).
With regard to mortality, studies have evaluated associations between short-term ambient
Os metrics and all-cause, non-accidental, and cause-specific (usually respiratory or
cardiovascular) mortality. The evidence from respiratory-related morbidity studies provides
strong support for respiratory-related mortality for which a causal determination has been made
(U.S. EPA, 2013b, Table 2-3). There are also a number of large studies that have found
associations between Os and all-cause and all non-accidental mortality for which a likely causal
determination has been made (U.S. EPA, 2013b, Table 2-3). Thus, it is appropriate to assess
risks for respiratory-related mortality as well as for all-cause total mortality associated with Os
exposure. The ISA also reports a likely causal determination for short-term Os and
cardiovascular effects, including cardiovascular mortality (U.S. EPA, 2013b, Table 2-3). This
determination is supported by studies relating total and cardiovascular mortality, coupled with
evidence from animal toxicological studies and controlled human exposure studies which find
effects of Os on systemic inflammation and oxidative stress. Cardiovascular mortality effects are
covered through the estimation of risks associated with total mortality, which is dominated by
cardiovascular mortality. There are not sufficient epidemiological studies of cardiovascular
morbidity showing consistent associations to justify inclusion of any cardiovascular morbidity
endpoints in the quantitative risk assessment.
With regard to effects associated with long-term Os exposures, the ISA states that the
relationship between Os and respiratory-related effects, including respiratory symptoms, new-
onset asthma, and respiratory mortality is likely causal (U.S. EPA, 2013b, Table 2-3). This
suggests that for long-term exposures, when comparing the evidence for respiratory-related
mortality and total mortality, the evidence is strongest for respiratory-related mortality, which is
supported by the strong evidence for respiratory morbidity. As a result, it is appropriate to
include respiratory mortality rather than total mortality in the risk assessment and to give
consideration to inclusion of additional respiratory-related health endpoints.
2.3.3 Exposure and Concentration-Response Functions for Health Risk Endpoints
Estimation of risk requires characterization of the E-R and C-R functions along the full
range of potential exposures. For E-R functions, the evidence from individual controlled human
exposure studies provides responses for exposures at and above 60 ppb. McDonnell et al. (2012)
develop an integrated model of FEVi response that is fit to the results from controlled human
2-13
-------
exposure studies and find that a model with a threshold provides the best fit to the data. In
addition, the ISA notes that it is difficult to characterize the E-R relationship at and below 40 ppb
due to the dearth of data at these lower concentrations (U.S. EPA, 2013b, section 2.5.4.4). Thus,
for the portion of the risk assessment based on application of results of controlled human
exposure studies, the threshold model is applied.
The evidence for a threshold in the C-R functions for mortality and morbidity outcomes
derived from the epidemiological literature is limited. In general, the epidemiological evidence
suggests a generally linear C-R function with no indication of a threshold. However, evaluation
of evidence for a threshold in the C-R function is complicated by the high degree of
heterogeneity between cities in the C-R functions and by the sparse data available at lower
ambient Os concentrations (U.S. EPA, 2013b, sections 2.5.4.4 and 2.5.4.5).
The ISA also evaluated whether the magnitude of the relationship between short-term
exposures to Os and mortality changes at lower concentrations (e.g., whether the C-R function is
non-linear). The ISA concludes that epidemiologic studies that examined the shape of the C-R
curve and the potential presence of a threshold have indicated a generally linear C-R function
with no indication of a threshold in analyses that have examined 8-hour max and 24-hour
average Os concentrations, and that the evidence supports less certainty in the shape of the C-R
function at the lower end of the distribution of Os concentrations, e.g., 24-hour average Os below
20 ppb, due to the low density of data in this range (U.S. EPA, 2013b, section 2.5.4.4).
Regarding long-term exposures to Os and mortality changes at lower concentrations, Jerrett et al.
(2009) evaluated a number of C-R functions with varying threshold levels. Statistical tests
indicated little discernable improvement in overall model fit when evaluating models that
included thresholds, however there remained uncertainty about the specific location of the
threshold, if one did exist (Jerrett et al., 2009; Sasser et al., 2014). In the absence of substantial
information in the scientific literature on alternative forms of C-R functions at low Os
concentrations, the best estimate of the C-R function is a linear, no-threshold function. The
scientific literature does not provide sufficient information with which to quantitatively
characterize any potential additional uncertainty in the C-R functions at lower Os concentrations
for use in the quantitative risk assessment.
Multiple exposures to elevated Os levels over the course of an Os season may result in
adaptation within exposed population. Evidence suggests that repeated or chronic exposures to
elevated Os can result in morphologic and biochemical adaptation which reduces the impacts of
subsequent Os exposures (U.S. EPA, 2013b, section 6.2.1.1). This has implications for exposure
modeling, in that the effects of modeled repeat exposures on risk may be attenuated relative to
the effects of the initial exposures. The ISA notes that "neither tolerance nor attenuation should
be presumed to imply complete protection from the biological effects of inhaled Os, because
2-14
-------
continuing injury still occurs despite the desensitization to some responses (U.S. EPA, 2013b,
section 6.2.1.1). Because of the complexity of this phenomenon, neither adaptation nor injury
accumulation resulting from multiple exposures is accounted for in the exposure and associated
risk modeling performed in Chapters 5 and 6, The ISA reports that there are limited
epidemiological studies evaluating adaptation to the mortality effects of Os, although the limited
evidence does suggest that mortality effects are decreased in later months during the Os season
relative to earlier months (U.S. EPA, 2013b, section 6.3.3). The impact of this phenomenon on
risks based on application of results from epidemiological studies is likely to be small, because
the relative risk estimates from those studies already incorporate any adaptive phenomenon.
Therefore, adaptation is not explicitly incorporated as an effect modifier in the epidemiological-
based risk modeling performed in Chapters 7 and 8.
2.3.4 At-Risk Populations
The Os ISA refers to "at-risk" populations as an all-encompassing term used for groups
with specific factors that increase the risk of an air pollutant- (e.g., Os) related health effect in a
population group (U.S. EPA, 2013b, chapter 8). Populations or lifestages can experience
elevated risks from Os exposure for a number of reasons. These include experiencing high levels
of exposure due to personal activities that include long durations of time spent in high Os
concentration locations, e.g., outdoor recreation or work, elevated exertion levels that increase
the dose of Os, e.g., engagement in moderate or strenuous levels of exercise, genetic or other
biological factors, e.g., lifestage, that may predispose an individual to sensitivity to a given dose
of Os, pre-existing diseases, e.g., asthma or COPD, and socioeconomic factors that may result in
more severe health outcomes, e.g., low access to primary care that can lead to increased
emergency department visits or hospital admissions. To consider risks to these populations,
modeling of exposures to Os needs to incorporate information on time spent by potentially at-risk
populations in high Os concentration locations. This requires identification of populations with
the identified exposure-related risk factors, e.g., children or adults engaging in activities
involving moderate to high levels of outdoor exertion, especially on a repeated basis typical of
student athletes or outdoor workers, as well as identifying populations with high sensitivity to
Os, e.g., asthmatic children. It also requires that information on Os concentrations be mapped to
locations where at-risk populations are likely to be exposed, e.g., near roadways where running
may occur, or at schools or parks where children are likely to be engaged in outdoor activities.
In addition to consideration of factors that lead to increased exposure to Cb, modeling of
risk from Os exposures should incorporate additional information on factors that can lead to
increased dose of Os for a given exposure, e.g., increased breathing rates during periods of
exertion. These factors are especially important for risk estimates based on application of the
2-15
-------
results of controlled human exposure studies. For risk modeling based on application of
observational epidemiology results, it is also important to understand characteristics of study
populations that can impact observed relationships between ambient Cb and population health
responses.
The Os ISA identifies a number of factors which have been associated with modifications
of the effect of ambient Os on health outcomes. Building on the causal framework used
throughout the Cb ISA, conclusions are made regarding the strength of evidence for each factor
that may contribute to increased risk of an Os-related health effect based on the evaluation and
synthesis of evidence across scientific disciplines. The OsISA categorizes potential risk
modifying factors by the degree of available evidence. These categories include "adequate
evidence," "suggestive evidence," "inadequate evidence," and "evidence of no effect." See Table
8-1 of the Os ISA for a discussion of these categories (U.S. EPA, 2013b, chapter 8).
Factors categorized as having adequate evidence include asthma, lifestage (children less
than 18 years of age, adults older than 65 years of age), diets with nutritional deficiencies, and
working outdoors. For example, children are the group considered to be at greatest risk because
they breathe more air per unit of body weight, are more likely to be active outdoors when Os
levels are high, are more likely than adults to have asthma, and are in a critical time period of
rapid lung growth and organ development. Factors categorized as having suggestive evidence
include genetic markers, sex (some studies have shown that females are at greater risk of
mortality from Os compared to males), low socioeconomic status, and obesity. Factors
characterized as having inadequate evidence include influenza and other respiratory infections,
COPD, cardiovascular disease, diabetes, hyperthyroidism, race, and smoking (U.S. EPA, 2013b,
section 8.5, Table 8-6).
2.4 URBAN-SCALE MODELING OF INDIVIDUAL EXPOSURE
Estimates of human exposure to Os provide important information to inform
consideration of policy-relevant questions identified in Section 2.2 regarding the occurrence of
exposures of concern under air quality conditions that meet existing and potential alternative
standards, and also to provide inputs to the portion of the risk assessment based on application of
results of controlled human exposure studies. Studies that measure human exposure to Cb are
limited. More commonly, human exposure is estimated using sophisticated models which
combine information on ambient Os concentrations in various microenvironments, e.g., near
roads, in schools, etc., with information on activity patterns for individuals sampled from the
general population or specific subpopulations, e.g., children with asthma.
Ozone exposure is highly dependent on the ambient Os concentrations in an urban area.
Given that these concentrations are variable from year to year, it is important to model multiple
2-16
-------
years representing the range of variability on Os concentrations to provide a better
characterization of potential exposures of concern. In addition, other important sources of
variability and uncertainty affecting the exposure estimates should be characterized, including
uncertainty and variability in the data on time-activity patterns, Os concentrations, and
population inputs. This can be accomplished in part by modeling exposure for multiple urban
areas selected to represent variability in these underlying sources of variability.
This section briefly describes the conceptual foundation for key components of exposure
modeling, characterization of microenvironmental Os concentrations, and characterization of
human activity patterns, including behaviors intended to avert exposures to Os. In addition, a
brief discussion of key factors to consider in selecting urban study areas for the exposure
analysis is provided. The specific exposure model used in this assessment, APEX, is described
more fully in Chapters 3 and 5. Characterization of ambient Os concentrations is discussed
earlier in this chapter and in greater detail in Chapter 4.
2.4.1 Microenvironmental Ozone Concentrations
Human exposure to Cb involves the contact (via inhalation) between a person and the
pollutant in the various locations (or microenvironments) in which people spend their time.
Ozone concentrations in some indoor microenvironments, such as within homes or offices, are
considerably lower than Os concentrations in similarly located outdoor microenvironments,
primarily due to deposition processes and the transformation of Os into other chemical
compounds within those indoor microenvironments. Concentrations of Os may also be quite
different in roadway environments, such as might occur while an individual is in a vehicle.
Thus, three important classes of microenvironments that should be considered when
evaluating population exposures to ambient Os are indoors, outdoors, and in-vehicle. Within
each of these broad classes of microenvironments, there are many subcategories, reflecting types
of buildings, types of vehicles, etc. The Os ISA evaluated the literature on indoor-outdoor Os
concentration relationships and found that studies consistently show that indoor concentrations
of Os are often substantially lower than outdoor concentrations unless indoor sources are present.
This relationship is greatly affected by the air exchange rate, which can be affected by open
windows, use of air conditioning, and other factors. Ratios of indoor to outdoor Os
concentrations generally range from about 0.1 to 0.4 (U.S. EPA, 2013b, section 4.3.2). In some
indoor locations, such as schools, there can be large temporal variability in the indoor-outdoor
ratios because of differences in air exchange rates over the day. For example, during the school
day, there is an increase in open doors and windows, so the indoor-outdoor ratio is higher during
the school day compared with an overall average across all hours and days. In-vehicle
concentrations are also likely to be lower than ambient concentrations, although the literature
2-17
-------
providing quantitative estimates is smaller. Studies of personal exposure to Cb have identified
that Os exposures are highest when individuals are in outdoor microenvironments, such as
walking outdoors midday, moderate when in vehicle microenvironments, and lowest in
residential indoor microenvironments (U.S. EPA, 2013b, section 4.3.3). Thus the time spent
indoors, outdoors, and in vehicles is likely to be a critical component in estimating Os exposures.
Because of localized chemistry, Os concentrations on or near roadways can be much
lower than away from roadways. This is due to the high levels of NOx emissions from motor
vehicles, which can lead to NOx titration of Os, reducing Cb levels during times of peak traffic.
The ISA reports evidence that concentrations of NO, NO2, and NOx are negatively correlated
with concentrations of Os near busy roadways. Because few monitors are located in direct
proximity to roadways, it is important to account for differences between near-road Os
concentrations and ambient Os measurements in modeling exposure.
2.4.2 Human Activity Patterns
Human exposure can be measured using several metrics. Exposure to ambient
concentrations is one such metric. It is also possible to model dose, which combines exposure
information with physiological parameters related to activity levels. In order to model exposure
to ambient concentrations, detailed information on the patterns of time spent in different
microenvironments is critical. In order to model Os dose, additional information on the activities
conducted while in those microenvironments is needed, along with data on physiological
parameters associated with different activities.
Several large-scale databases of human time-activity-location patterns have been
compiled. The most comprehensive of these databases in the Consolidated Human Activity
Database (CHAD), which has been the basis of several previous exposure analyses for previous
NAAQS reviews. These databases compile large numbers of diaries collected from individual
study subjects who recorded their time spent performing different activities while in different
locations, data that were originally collected as part of smaller studies. The ISA notes the high
degree of variability in activity patterns across the population, as well as the variability in time
spent in different microenvironments. Time-activity-location patterns vary by age group, as well
as by region of the U.S. Children generally spend more time in outdoor locations and also
generally have higher activity levels in those environments.
The dose of Os received for any given exposure in a microenvironment depends not only
on the activity levels and Os concentrations in the microenvironment, but also on ventilation
rates, which are related to age, body weight, and other physiological parameters. Children
generally have lower ventilation rates than adults when considering the volume of air breathed
per unit time; however, they tend to have a greater oral breathing contribution than adults, and
2-18
-------
due to smaller lung volumes and generally greater breathing frequencies, children breathe at
higher body mass or surface area normalized minute ventilation rates, relative to their lung
volumes. Both of these factors tend to increase their applied or intake dose normalized to lung
surface area. For example, when comparing daily body mass normalized ventilation rates,
children can have up to a factor of two greater ventilation rates when compared to that of adults.
During periods of high activity, ventilation rates for children and young adults can be nearly
double those during moderate activity. Thus, it is important to model levels of activity and
associated ventilation rate as well as time spent in different microenvironments.
In addition to modeling daily exposures, it may also be important to understand the
patterns of exposure over an Os season, including multiple repeated exposures for the same
individuals. Some individuals or subpopulations may exhibit multiple high daily exposures due
to persistent patterns of high activity in microenvironments with high Os concentrations. For
example, children engaged in numerous outdoor sports over a summer Os season may have
multiple exposures to elevated Os levels.
Another important issue in characterizing exposure involves consideration of the extent
to which people in relevant population groups modify their behavior for the purpose of
decreasing their personal exposure to Os based on information about predicted air quality levels
made public through the Air Quality Index (AQI). The AQI is the primary tool EPA has used to
communicate information on predicted occurrences of high levels of Os and other pollutants. The
AQI provides both the predicted level of air quality in an area along with a set of potential
actions that individuals and communities can take to reduce exposure to air pollution and thus
reduce the risk of health effects associated with breathing ambient air pollution. There are
several studies, discussed in the Os ISA, that have evaluated the degree to which populations are
aware of the AQI and what actions individuals and communities take in response to AQI values
in the unhealthy range. These studies suggest that at-risk populations, such as children, older
adults, and asthmatics, modify their behavior in response to days with bad air quality, most
commonly by reducing their time spent outdoors or limiting their outdoor activity exertion level.
A challenge remains in how to consider existing averting behaviors within the assessment tools
we use and how best to use improved knowledge of participation rates, the varying types of
actions performed particularly by potentially at-risk individuals, and the duration of these
averting behaviors to quantify the impact on estimated exposures and health risks.
2.4.3 Modeling of Exposures Associated with Simulating Just Meeting Ozone Standards
In order to address policy-relevant questions regarding changes in exposure associated
with potential alternative standards, the exposure assessment evaluates changes in the Os
concentrations, and the resulting changes in exposure, associated with simulating just meeting
2-19
-------
alternative standards relative to just meeting the existing standards. The new, model-adjustment
methodology being implemented in this risk and exposure assessment provides for more realistic
responses of hourly Os concentrations to changes in the precursor emissions that lead to Os
formation. Characterization of exposure and changes in exposure when simulating just meeting
the alternative standards are discussed in greater detail in Chapter 5.
2.4.4 Considerations in Selecting Urban Study Areas for the Exposure Analysis
The goal of the urban-scale exposure analysis is to characterize the variability in
exposures for different geographic locations, taking into account variability in essential factors
that affect exposures. Important factors identified earlier that may influence exposure include
time activity patterns, especially activities occurring in outdoor environments; demographics of
the exposed population, e.g., age and income level; and Os concentrations. In addition to these
factors, the selection of urban areas to include in the exposure analysis takes into consideration
the location of Os epidemiological studies (for comparability with the risk assessments), the
availability of ambient Os data and specific exposure information (e.g., air conditioning
prevalence), and the desire to represent a range of geographic areas. To make the exposure
analysis most useful in addressing the key policy-relevant questions, urban study areas were also
chosen such that most of them exceeded the existing 8-hr Os standards and potential alternative
standards during the time period of interest.
2.5 URBAN- AND NATIONAL-SCALE RISK ASSESSMENT
Assessment of risk entails joint consideration of the exposure to a hazard, frequency of
adverse outcomes given exposure, and severity of resulting adverse outcomes. A risk assessment
for Os requires characterization of exposures to ambient Os for relevant populations,
identification of appropriate dose-response or C-R functions linking Os with adverse health
outcomes, and characterizing risks for individuals and populations.
As discussed above, there are two classes of studies that have provided information to
inform the risk modeling: controlled human exposure studies and observational epidemiology
studies. The conceptual approach to risk assessment varies based on which type of study result is
being applied. This section briefly describes the conceptual foundation for several aspects of risk
modeling, including the concept of attributable risk, modeling of total risk and incremental risk
reductions, development of risk estimates based on controlled human exposure studies, and
development of risk estimates based on results of observational epidemiology studies.
This section briefly describes the conceptual foundation for key elements of risk
modeling, including a discussion of the concept of attributable risk, modeling of risk for total Os
exposure and the distribution of risk over Os concentrations, modeling of risk reductions
2-20
-------
associated with alternative standards, and key factors to consider in selecting urban study areas
for the risk analysis. Characterization of ambient Os concentrations is discussed earlier in this
chapter and in greater detail in Chapter 4. The specific risk models used in the urban-scale risk
analyses, APEX for analyses based on application of controlled human exposure studies and
BenMAP for analyses based on application of observational epidemiology studies, are described
more fully in Chapters 6 and 7, respectively. Chapter 8 provides an additional national-scale
assessment of mortality risk associated with recent Os concentrations, to provide context for
evaluating the magnitude of health risks in the urban study areas and to evaluate the
representativeness of the urban study areas in estimating Ch risks.
2.5.1 Attributable Risk
This risk and exposure assessment relies on the concept of attributable risk in evaluating
both total risk and incremental changes in risk associated with just meeting existing and potential
alternative Os standards. Attributable risk is defined as the difference in incidence of an adverse
effect between an exposed and unexposed population for a specific stressor. Attributable risk is
an important concept when addressing risks that are associated with multiple causes, such as
mortality and respiratory hospital admissions.
Estimates of attributable risk require either an E-R function (for analyses based on results
of controlled human exposure studies) or a C-R function (for analyses based on results of
epidemiology studies).
E-R functions require estimates of exposure to calculate health risk, and in this case
exposures are supplied by the APEX modeling described above. In the case of the lung function
endpoint evaluated in this risk analysis, the E-R function also requires information on age and
exertion levels to predict the impact of Os exposure on decrements in lung function. E-R
functions may provide estimates of the incidence of an endpoint or the probability of exceeding
benchmark decrement levels.
C-R functions derived from relative risk estimates reported in the epidemiological
literature generally require estimates of ambient Os concentrations, baseline incidence rates, and
estimates of exposed populations to calculate health risk. Ambient Os concentrations should
generally be constructed to match the spatial and temporal averaging used in the underlying
epidemiology study; e.g., a study may have used a spatial average over a metropolitan statistical
area of the daily maximum 8-houraverage concentration.
As with exposure, attributable risk is highly dependent on the ambient Cb concentrations
in an urban area. Given that these concentrations are variable from year to year, it is important to
model multiple years representing the range of variability of Os concentrations to provide a
better characterization of risk. In addition, other important sources of variability and uncertainty
2-21
-------
affecting the risk estimates should be characterized, including uncertainty and variability in the
C-R and E-R functions, Os concentrations and Os exposure, and population inputs. This can be
accomplished in part by modeling risk for multiple urban areas selected to represent variability in
these underlying risk drivers.
2.5.2 Modeling of Health Risk Associated with Total Ozone Exposure
As discussed earlier in this chapter, ambient Os is contributed to by emissions from a
variety of sources, including natural, U.S. anthropogenic, and non-U.S. anthropogenic sources.
Once in the atmosphere, Os molecules created from these different sources of emissions are not
distinguishable. Individuals and populations are exposed to total Os from all sources, and risks
associated with Os exposure are due to total Os exposure and do not vary for Os exposure
associated with any specific source including background sources, regardless of how they may
be defined. Given the absence of a detectable threshold in the available C-R functions, total risk
attributable to Os will thus be the risk associated with total exposure to Os, with no threshold or
cut-point applied. To address certain policy-related questions, it is possible to approximately
attribute risk to specific sources through the use of air quality modeling techniques, and this is
explored in the PA. However, these techniques are based on applying model results to total Os
risk, rather than on directly modeling risk attributable to specific sources.
As discussed earlier in this chapter, a critical policy-relevant risk question is the Os
attributable risk remaining after just meeting the existing Os standards. This risk includes risks
associated with Os from all sources after we have simulated just meeting the existing daily
maximum 8-hour standard level of 75 ppb. The estimates of total risk remaining after meeting
the existing standard form the reference values for evaluating reductions in risk associated with
just meeting alternative levels of the standard.
In addition to providing risk estimates for urban study areas, it is also useful to evaluate
Os risks across the entire U.S., both to better understand the total magnitude of the health burden
associated with Os exposures and to evaluate the representativeness of selected urban study areas
in characterizing the range and variability in risks across the U.S. The national-scale risk
assessment presented in Chapter 8 is focused on estimating risk associated with recent Os
concentrations, rather than on risk after just meeting existing or alternative standards. This is the
appropriate focus for the national-scale analysis, because the techniques used to simulate just
meeting existing and alternative standards in urban study areas are less certain in a national
context due to concerns about interdependence between air quality responses in different urban
areas; e.g., just meeting a standard in one urban area would likely have impacts on Os air quality
in surrounding urban areas. It is beyond the scope of this HREA to simulate a full suite of all
2-22
-------
possible control strategies that would result in just meeting existing or alternative primary health
standards.
2.5.3 Distributions of Risk Across Ozone Concentrations
Total Os risk for the Os season is calculated by summing daily risks across all days in the
Os season. Because of the high degree of variability in daily Os concentrations across an Os
season, total Os risk will include risks calculated for some days with high Os concentrations as
well as for some days with very low Os concentrations. Therefore it is appropriate to provide the
distribution of total risk over the range of daily Os concentrations to allow for an understanding
of how Os concentrations on different days are contributing to the estimates of total risk. In
addition, as noted in the ISA and discussed above, because C-R functions are commonly
comprised of fewer days with low Os concentrations of compared with the number of high
concentration days, there is decreased confidence in the shape of the C-R function at lower Os
concentrations, and therefore lower confidence in risk estimates for days with lower Os
concentrations, especially in the range below 20 ppb. As a result, it is appropriate to provide the
distribution of total risk over the range of daily Os concentrations to allow for better
characterization of confidence in the estimates of total risk.
2.5.4 Modeling Risk Associated with Simulating Just Meeting Ozone Standards
In order to address policy-relevant questions regarding changes in risk associated with
potential alternative standards, the risk assessment evaluates changes in the distribution of Os
concentrations, and the resulting changes in risk, associated with simulating just meeting
alternative standards relative to just meeting the existing standards. The new, model-adjustment
methodology being implemented in this risk and exposure assessment provides for more realistic
responses of hourly Os concentrations to changes in the precursor emissions that lead to Os
formation than when compared with that using a purely statistical-based approach.8 As noted
earlier there are multiple combinations of reductions in precursor emissions that can result in just
meeting alternative standards. As a result, there is variability in the potential changes in the
distribution of Os concentrations and risk that would result from just meeting existing and
alternative standards. Characterization of this variability, as well as uncertainties in the
simulation of just meeting the standards, will be included in Chapters 6 and 7.
! The new, model-adjustment approach captures the nonlinearity of O3 response to emissions changes, representing
both increases and decreases in O3 concentrations resulting from emissions reductions. In addition, HDDM
characterizes different O3 response at different locations (downtown urban versus downwind suburban) and at
different times of day allowing us to incorporate temporal and spatial variations in response into the O3 adjustment
methodology. Finally, HDDM eliminates the need to use background O3 as a floor for adjustment because
predicted sensitivities are based on model formulations that explicitly account for background sources.
2-23
-------
2.5.5 Considerations in Selecting Urban Study Areas for the Risk Assessment
The goal of the urban-scale risk analysis is to characterize the magnitude of risk and the
impact on risk of meeting existing and potential alternative standards. The selection of specific
urban study areas is based on a set of factors reflecting both variability in factors that affect risk
and availability of high quality input data, to provide risk estimates that have higher overall
confidence. Important factors identified earlier that may influence risk include Os concentrations,
demographics, exposure factors, and magnitude of the effect estimate in the C-R function. In
addition to consideration of variability in these factors, urban study areas are preferentially
selected if they have Os concentrations that are above the existing standards and potential
alternative standards, if they have suitable epidemiological studies to provide C-R functions for
mortality or morbidity, if they have adequate monitoring data available to characterize
population exposures, and if they have appropriate baseline health incidence data available.
2.6 RISK CHARACTERIZATION
Risk characterization is the process of communicating the results of risk (and exposure)
modeling in metrics that have meaning to decision-makers. In the specific context of this review,
this translates into providing metrics that are most useful in the Os PA to assess the adequacy of
the existing Os standards in protecting public health with an adequate margin of safety and to
evaluate the additional protection provided by potential alternative standards.
Risk characterization requires careful translation of very complex outputs of exposure
and risk models into simpler metrics, for example, translating hourly Os exposures in various
microenvironments into estimates of population exposures above alternative exposure
benchmarks. Risk characterization also requires the condensation of a large number of analytical
steps and results to (a) summarize the results of the risk analysis, usually taking detailed results
and condensing them into a more aggregate interpretation while still providing information about
heterogeneity across space and time; (b) communicate the sensitivity of results to different
modeling assumptions; and (c) characterize the qualitative and quantitative uncertainty in results.
As described more fully in Chapter 5 and in the Os PA, EPA has selected, based on
providing a reasonable measure of exposures of concern for at-risk populations and lifestages,
aggregate exposure metrics including the number and percent of certain highly vulnerable
populations exposed to levels of Cb above exposure levels that have been identified in the
scientific literature as associated with adverse respiratory responses (i.e., pulmonary function
decrements and increases in respiratory symptoms, lung inflammation, lung permeability, and airway
hyper-responsiveness). As noted in section 2.3.1 above, these benchmark exposure levels are
0.060 ppm, 0.070 ppm, and 0.080 ppm. Important at-risk populations include active children,
older adults, and outdoor workers.
2-24
-------
As described more fully in Chapters 6 and 7 and in the Os PA, EPA has selected, based
on providing characterization of risks to the public including at-risk populations and lifestages,
aggregate risk metrics including the number and percent of vulnerable populations experiencing
adverse respiratory responses based on application of results of controlled human exposure
studies and the attributable incidence and percent of baseline incidence of mortality and
morbidity endpoints based on application of results of epidemiology studies.
For all three types of metrics (exposure, risk based on controlled human exposure studies,
and risk based on epidemiology studies) and for the purpose of evaluating the adequacy of the
existing standards, the focus is on the exposure and risk remaining upon just meeting the existing
standards. For the purpose of evaluating alternative standards, the focus in on the changes in
exposure and risk after simulating just meeting the alternative standards, compared to exposures
and risk after simulating just meeting the existing standards.
As detailed in Chapter 3, quantitative sensitivity analyses are provided to evaluate the
impacts of critical inputs to the exposure and risk modeling. Limited quantitative uncertainty
analyses are also included, along with a comprehensive qualitative uncertainty assessment. The
overall treatment of uncertainty is guided by the World Health Organization (WHO) guidelines
for uncertainty assessment (WHO, 2008). These guidelines recommend a tiered approach in
which progressively more sophisticated methods are used to evaluate and characterize sources of
uncertainty depending on the overall complexity of the risk assessment.
In order to inform considerations of overall confidence in the risk estimates derived from
application of C-R functions derived from the epidemiological literature, we provide the
distributions of total risk across the entire range of daily maximum 8-hour Os concentrations. In
addition, we provide an assessment of the representativeness of the urban study areas selected for
the risk and exposure analysis in characterizing the overall distribution of risk across the U.S.
This assessment evaluates how well the selected urban study areas capture important
characteristics that are associated with risk, including demographics, air quality levels, and
factors affecting exposure such as air conditioning prevalence.
2-25
-------
2.7 REFERENCES
Frey, C. and J. Samet. 2012. CASAC Review of the EPA's Policy Assessment for the Review of
the Ozone National Ambient Air Quality Standards (First External Review Draft -
August 2012). U.S. Environmental Protection Agency Science Advisory Board. EPA-
CASAC-13-003.
Jaegle, L.; DJ. Jacob; W.H. Brune; and P.O. Wennberg. 2001. Chemistry of HOx radicals in the
upper troposphere. Atmos Environ. 35:469-489.
Jerrett, M.; R.T. Burnett; C.A. Pope, III; K. Ito; G. Thurston; D. Krewski; Y. Shi; E. Calle; and
M. Thun. 2009. Long-term Os exposure and mortality. NEJM. 360:1085-1095.
NRC. 2009. Science and Decisions: Advancing Risk Assessment. National Research Council,
Washington, DC: The National Academies Press.
Sasser, E. 2014. Response to Comments Regarding the Potential Use of a Threshold Model in
Estimating the Mortality Risks from Long-term Exposure to Ozone in the Health Risk
and Exposure Assessment for Ozone, Second External Review Draft. Memorandum to
Holly Stallworth, Designated Federal Officer, Clean Air Scientific Advisory Committee
from EPA/OAQPS Health and Environmental Impacts Division.
U.S. EPA. 2011. Integrated Review Plan for the Os National Ambient Air Quality Standards.
Research Triangle Park, NC: National Center for Environmental Assessment, Office of
Research and Development and Office of Air Quality Planning and Standards, Office of
Air and Radiation. (EPA document number EPA 452/R-l 1-006).
.
U.S. EPA. 2012a. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
Documentation (TRIM. Expo /APEX, Version 4.4) Volume I: User's Guide. Research
Triangle Park, NC: EPA Office of Air Quality Planning and Standards. (EPA document
number EPA-452/B-12-001a). .
U.S. EPA. 2012b. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
Documentation (TRIM. Expo / APEX, Version 4.4) Volume II: Technical Support
Document. Research Triangle Park, NC: Office of Air Quality Planning and Standards.
(EPA document number EPA-452/B-12-001b).
< http://www.epa.gov/ttn/fera/human apex.html>.
U.S. EPA. 2013a. Environmental Benefits Mapping Analysis Program (BenMAP v4.0). Posted
January, 2013. < http://www.epa.gov/air/benmap/download.html />.
U.S. EPA. 2013b. Integrated Science Assessment of Ozone and Related Photochemical Oxidants
(Final Report). Research Triangle Park, NC: EPA Office of Research and Development.
(EPA document number EPA/600/R-10/076F).
.
WHO. 2008. Part 1: Guidance Document on Characterizing and Communicating Uncertainty in
Exposure Assessment, Harmonization Project Document No. 6. Published under joint
sponsorship of the World Health Organization (WHO), the International Labor
Organization and the United Nations Environment Programme. WHO Press, World
Health Organization, 20 Avenue Appia, 1211 Geneva 27, Switzerland.
2-26
-------
3 SCOPE
This chapter provides an overview of the scope and key design elements of this
quantitative exposure and health risk assessment. The design of this assessment began with a
review of the exposure and risk assessments completed during the last Os NAAQS review (U.S.
EPA, 2007a,b), with an emphasis on considering key limitations and sources of uncertainty
recognized in that analysis.
As an initial step in the current Cb NAAQS review in October 2009, EPA invited outside
experts, representing a broad range of expertise (e.g., epidemiology, human and animal
toxicology, statistics, risk/exposure analysis, atmospheric science), to participate in a workshop
with EPA staff to help inform EPA's plan for the review. The participants discussed key policy-
relevant issues that would frame the review and the most relevant new science that would be
available to inform our understanding of these issues. One workshop session focused on planning
for quantitative risk and exposure assessments, taking into consideration what new research
and/or improved methodologies would be available to inform the design of quantitative exposure
and health risk assessment. Based in part on the workshop discussions, EPA developed a draft
IRP (U.S. EPA, 2009a) outlining the schedule, process, and key policy-relevant questions that
would frame this review. On November 13, 2009, EPA held a consultation with CASAC on the
draft IRP (74 FR 54562, October 22, 2009), which included opportunity for public comment.
The final IRP incorporated comments from CASAC (Samet, 2009) and the public on the draft
plan, as well as input from senior Agency managers. The final IRP included initial plans for
quantitative risk and exposure assessments for both human health and welfare (U.S. EPA, 201 la,
chapters 5 and 6).
As a next step in the design of these quantitative assessments, OAQPS staff developed
more detailed planning documents, the Ozone National Ambient Air Quality Standards: Scope
and Methods Plan for Health Risk and Exposure Assessment (Health Scope and Methods Plan,
U.S. EPA, 201 Ib) and the Ozone National Ambient Air Quality Standards: Scope and Methods
Plan for Welfare Risk and Exposure Assessment (Welfare Scope and Methods Plan, U.S. EPA,
201 Ic). These Scope and Methods Plans were the subject of a consultation with CASAC on May
19-20, 2011 (76 FR 23809, April 28, 2011). Based on consideration of CASAC (Samet, 2011)
and public comments on the Scope and Methods Plans, and information in the second draft ISA,
we modified the scope and design of the quantitative risk assessment and provided a memo with
updates to information presented in the Scope and Methods Plans (Wegman, 2012). The Scope
and Methods Plans together with the update memo provide the basis for the discussion of the
scope of this exposure and risk assessment provided in this chapter. This chapter also reflects
3-1
-------
comments received from CAS AC based on their review of the draft Risk and Exposure
Assessments (Frey and Samet, 2012a; Frey, 2014).
In presenting the scope and key design elements of the current risk assessment, this
chapter first provides a brief overview of the quantitative exposure and risk assessment
completed for the previous Os NAAQS review in section 3.1, including key limitations and
uncertainties associated with that analysis. The remaining sections describe the current exposure
and risk assessment, following the general conceptual framework described in Chapter 2. Section
3.2 provides a summary of the design of the urban-scale exposure assessment. Section 3.3
provides a summary of the design of the urban-scale risk assessment based on application of
results of human clinical studies. Section 3.4 provides a summary of the design of the urban-
scale risk assessment based on application of results of epidemiology studies. Section 3.5
provides a summary of the design of the national-scale risk burden assessment and
representativeness analysis.
3.1 OVERVIEW OF THE EXPOSURE AND RISK ASSESSMENTS FROM THE
LAST REVIEW
The exposure and health risk assessment conducted in the review, completed in March
2008, developed exposure and health risk estimates for 12 urban areas across the U.S. which
were chosen based on the location of Os epidemiological studies and availability of ambient Os
data and to represent a range of geographic areas, population demographics, and Os climatology.
That analysis was in part based upon the exposure and health risk assessments included in the
review completed in 1997.!The exposure and risk assessment incorporated air quality data (i.e.,
2002 through 2004), and provided annual or Os season-specific exposure and risk estimates for
these recent years of air quality and for air quality scenarios simulating just meeting the existing
8-hour (8-hr) Os standard and several alternative 8-hr Os standards.
3.1.1 Overview of Exposure Assessment from the Last Review
Exposure estimates were used as an input to the risk assessment for lung function
responses (a health endpoint for which exposure-response (E-R) functions were available from
controlled human exposure studies). Exposure estimates were developed for the general
population and study groups including asthmatic school-age children (ages 5-18) as well as all
1 In the 1994-1997 Os NAAQS review, EPA conducted exposure analyses for the general population, children who
spent more time outdoors, and outdoor workers. Exposure estimates were generated for 9 urban areas for base air
quality and for air quality adjusted to just meeting the then-existing 1-hr standard and several alternative 8-hr
standards. Several reports that describe these analyses can be found at:
http://www.epa.gov/ttn/naaqs/standards/ozone3/s_03_pr.html.
3-2
-------
school-age children. The exposure estimates also provided information on population exposures
exceeding potential health effect benchmark levels that were identified based on the observed
occurrence of health endpoints not explicitly modeled in the health risk assessment (e.g., lung
inflammation, increased airway responsiveness, and decreased resistance to infection) associated
with 6- to 8-hr exposures to Os in controlled human exposure studies.
The exposure analysis took into account several important factors including the
magnitude and duration of exposures, frequency of repeated high exposures, and breathing rate
of individuals at the time of exposure. Estimates were developed for several indicators of
exposure to various levels of Cb air quality, including counts of people exposed one or more
times to a given Os concentration while at a specified breathing rate and counts of person-
occurrences (which accumulate occurrences of specific exposure conditions over all people in
the population groups of interest over an Os season).
As discussed in the Os exposure assessment (U.S. EPA, 2007a), a technical support
document (Langstaff, 2007), the 2007 Staff Paper (U.S. EPA, 2007c) and in Section II a of the
Os Final Rule (73 FR 16440 to 16442, March 27, 2008), the most important uncertainties
affecting the exposure estimates were related to modeling human activity patterns over an Os
season, modeling of variations in ambient concentrations near roadways, and modeling of air
exchange rates that affect the amount of Os that penetrates indoors. Another important
uncertainty, discussed in more detail in the Staff Paper (U.S. EPA, 2007c, section 4.3.4.7), was
the uncertainty in energy expenditure values which directly affected the modeled breathing rates.
These were important because they were used to classify exposures occurring when children
were engaged in moderate or greater exertion. Health effects observed in the controlled human
exposure studies generally occurred under these exertion levels for 6 to 8-hr exposures to Os
concentrations at or near 0.08 ppm.
3.1.2 Overview of Risk Assessment from the Last Review
The human health risk assessment presented in the review completed in March 2008 was
designed to estimate population risks in a number of urban areas across the U.S., consistent with
the scope of the exposure analysis described above (U.S. EPA, 2007a,b,c). The risk assessment
included risk estimates based on both controlled human exposure studies and epidemiological
and field studies. Os-related risk estimates for lung function decrements were generated using
probabilistic E-R relationships based on data from controlled human exposure studies, together
with population-based probabilistic exposure estimates from the exposure analysis. For several
other health endpoints, Os-related risk estimates were generated using concentration-response
(C-R) relationships reported in epidemiological or field studies, together with ambient air quality
concentrations, baseline health incidence rates, and population data for the various locations
3-3
-------
included in the assessment. Health endpoints included in the assessment based on
epidemiological or field studies included hospital admissions for respiratory illness in four urban
areas, premature mortality in 12 urban areas, and respiratory symptoms in asthmatic children in 1
urban area.
In the health risk assessment conducted in the previous review, EPA recognized that there
were many sources of uncertainty and variability in the inputs to the assessment and that there
was significant uncertainty in the resulting risk estimates. The statistical uncertainty surrounding
the estimated Os coefficients in epidemiological-based C-R functions as well as the shape of the
E-R relationship chosen for the lung function risk assessment were addressed quantitatively.
Additional uncertainties were addressed through sensitivity analyses and/or qualitatively. The
risk assessment conducted for the previous Os NAAQS review incorporated some of the
variability in key inputs to the assessment by using location-specific inputs (e.g., location-
specific C-R functions, baseline incidence rates and population data, and air quality data for
epidemiological-based endpoints, location specific air quality data and exposure estimates for the
lung function risk assessment). In that review, several urban areas were included in the health
risk assessment to provide some insight as to variability in the risk estimates across the U.S.
Key observations and insights from the Os risk assessment, in addition to important
caveats and limitations, were addressed in Section II.B of the Final Rule notice (73 FR 16440 to
14 16443, March 27, 2008). In general, estimated risk reductions associated with going from
then-current Os levels to just meeting the then-existing and alternative 8-hr standards showed
patterns of decreasing estimated risk associated with just meeting the lower alternative 8-hr
standards considered. Furthermore, the estimated percentage reductions in risk were strongly
influenced by the baseline air quality year used in the analysis, which was due to significant
year-to-year variability in Os concentrations. There was also noticeable city-to-city variability in
the estimated Os-related incidence of morbidity and mortality across the 12 urban areas.
Uncertainties associated with estimated policy-relevant background (PRB) concentrations2 were
also addressed and revealed differential impacts on the risk estimates depending on the health
effect considered as well as the location. EPA also acknowledged that at the time of the previous
review there were considerable uncertainties surrounding estimates of Os C-R coefficients and
the shape of C-R relationships and whether or not a population threshold or non-linear
relationship exists within the range of concentrations examined in the epidemiological studies.
' Policy-relevant background (PRB) Cb has been defined in previous reviews as the distribution of Os concentrations
that would be observed in the U.S. in the absence of anthropogenic (man-made) emissions of Os precursor
emissions (e.g., VOC, CO, NOX) in the U.S., Canada, and Mexico.
3-4
-------
3.2 PLAN FOR THE CURRENT EXPOSURE AND RISK ASSESSMENTS
The Scope and Methods Plan, including updates (U.S. EPA, 201 Ib; Wegman, 2012),
outlined a planned approach for conducting the current quantitative Os exposure and risk
assessments, including broad design issues as well as more detailed aspects of the analyses. A
critical step in designing the quantitative risk and exposure assessments is to clearly identify the
goals for the analysis based on the policy-relevant questions identified in Chapter 2. We have
identified the following goals for the urban area exposure and risk assessments: (1) to provide
estimates of the percent and number of people in the general population and in sensitive
populations with Os exposures above health-based benchmark levels; (2) to provide estimates of
the percent and number of the general population and in sensitive populations with impaired lung
function (defined based on decrements in FEVi) resulting from exposures to Cb; (3) to provide
estimates of the potential magnitude of premature mortality associated with both short-term and
long-term Os exposures, and selected morbidity health effects associated with short-term Os
exposures; (4) to evaluate the influence of various inputs and assumptions on exposure and risk
estimates to the extent possible given available methods and data; (5) to gain insights into the
spatial and temporal distribution of risks and patterns of risk reduction and uncertainties in those
risk estimates. For the exposure and risk analyses, we will estimate exposures and risks for
recent (2006 - 2010) ambient levels of Os and for Os concentrations after simulating just
meeting the existing Os standard and potential alternative standards.
With regard to selecting alternative levels for the 8-hr Os standards for evaluation in the
quantitative risk assessment, we base the range of levels on the evaluations of the evidence
provided in the PA, which received support from the CASAC in their advisory letter (Frey and
Samet, 2012b). The PA recommended evaluation of daily maximum 8-hr average concentrations
in the range of 60 to 70 ppb, with possible consideration of levels somewhat below 60 ppb. The
upper end of this range is supported by the clear evidence from both clinical and epidemiological
studies of effects at exposures of 70 ppb reported in the ISA and summarized in the PA. The
lower end of this range is based on considerations of evidence from clinical studies that have
shown lung function decrements in healthy adult populations at 60 ppb Os exposures, and that 10
percent of healthy adults exposed to 60 ppb Os experienced lung function decrements that could
be adverse to asthmatics. The evidence showing effects in healthy adults at exposures of 60 ppb
supports the consideration of risks to sensitive populations at exposure levels below 60 ppb,
although specific exposure levels below 60 ppb at which risks may be occurring are not
supported by the evidence. An important distinction is that the evidence from controlled human
exposure studies is based on actual exposure concentrations, while the standard is set to address
ambient concentrations. Typically, exposures are lower than ambient concentrations because
3-5
-------
people tend to spend a large fraction of their time indoors where Os concentrations are
substantially lower.3
Because of the year-to-year variability in Os concentrations that results from temporal
variability in meteorology and emissions that drive Os formation, the exposure and risk
assessments evaluate scenarios for meeting the existing and alternative standards based on
multiple years of Os data. Ozone concentrations from 2006-2010 are used in estimating exposure
and risk. This range of years captures a high degree of variability in meteorological conditions,
as well as reflecting years with higher and lower emissions of Os precursors.
In order to provide greater confidence in the exposure and risk estimates, this health REA
(HREA) uses an urban-scale analysis approach for assessing both exposure and risk. This
approach provides greater confidence in estimates by allowing us to make use of air quality data,
population information, health data, and epidemiology results that are well matched, and it does
not require extrapolation of results to areas without these data. In addition, the urban-scale
analysis approach allows us to simulate just meeting existing and alternative Os standards for
each urban study area, which is not currently feasible for health risk assessment at the national-
scale.4 Specific selection criteria for urban study areas included in the exposure and risk
assessments are described in the following sections. In order to gain an understanding of how
well the urban study areas represent risks at a national level and to provide context for the urban-
scale analysis results, we also include two national-scale analyses, (1) estimation of the national-
scale mortality burden associated with recent (2007) ambient Os and (2) characterization of how
well the risk estimates for the set of urban areas modeled reflect the national-scale distribution of
mortality risk.
Throughout the exposure and risk analyses, we recognize that there are many sources of
variability and uncertainty. Each analysis considers carefully the potential sources and
significance of variability and uncertainties and, where data are available, provides quantitative
assessment of variability and uncertainties, either through probabilistic analyses or through
sensitivity or scenario analyses. In general the analyses follow the World Health Organization
(WHO) guidelines for uncertainty assessment (WHO, 2008), which recommend a tiered
3 While almost all people spend a large fraction of their time indoors, there is high variability in this fraction
between children and adults, and between outdoor workers and indoor workers. The ratio of exposures to ambient
concentrations will likely be higher for children than adults, and for outdoor workers compared to indoor workers.
4 In order to simulate just meeting alternative standards everywhere nationwide using the model-based adjustment
approach employed in this HREA, some areas would see Os design values decreased below the targeted standard
level due to Os transport between locations. We were not able to devise an approach that would just meet the
standard in every location simultaneously. Using the urban-scale analysis approach, we can, acknowledging the
counterfactual nature of the analysis, assume independence of each urban study area, which allows us to simulate
just meeting the standards in each urban study area.
3-6
-------
approach in which progressively more sophisticated methods can be used to evaluate and
characterize sources of uncertainty depending on the overall complexity, end use of the
assessment, and resources and data available to conduct particular uncertainty characterizations.
The planned approaches for conducting the exposure and risk analyses are briefly
summarized below. We begin with a general discussion of how uncertainty and variability are
addressed in the different elements of the exposure and risk assessment. This is followed by a
discussion of the air quality data that will be used in both the exposure and risk assessments and
then discussions of each component of the exposure and risk assessments.
3.3 CHARACTTERIZATION OF UNCERTAINTY AND VARIABILITY IN THE
CONTEXT OF THE OZONE EXPOSURE AND RISK ASSESSMENT
An important component of this population exposure and health risk assessment is the
characterization of both uncertainty and variability. Variability refers to the heterogeneity of a
variable of interest within a population or across different populations. For example, populations
in different regions of the country may have different behavior and activity patterns (e.g., air
conditioning use and time spent indoors) that affect their exposure to ambient Os and thus the
population health response. The composition of populations in different regions of the country
may vary in ways that can affect the population response to exposure to Os - e.g., two
populations exposed to the same levels of Cb might respond differently if one population has a
greater proportion of older people than the other. Variability is inherent and cannot be reduced
through further research. Refinements in the design of a population risk assessment are often
focused on more completely characterizing variability in key factors affecting population risk -
e.g., factors affecting population exposure or response - to produce risk estimates whose
distribution adequately characterizes the distribution in the underlying population(s).
Uncertainty refers to the lack of knowledge regarding the actual values of inputs to an
analysis. Models are typically used in analyses, and there is uncertainty about the true values of
the model inputs and their parameterization (parameter uncertainty) - e.g., the value of the
coefficient for Os in a C-R function. There is also uncertainty about the extent to which the
model is an accurate representation of the underlying physical systems or relationships being
modeled (model uncertainty) - e.g., the shapes of C-R functions. In addition, there may be some
uncertainty surrounding other inputs to an analysis due to possible measurement error - e.g., the
values of daily Os concentrations in a risk assessment location or the value of the baseline
incidence rate for a health effect in a population.5
1 It is also important to point out that failure to characterize variability in an input used in modeling can also
introduce uncertainty into the analysis. This reflects the important link between uncertainty and variability with
3-7
-------
In any risk assessment, uncertainty is, ideally, reduced to the maximum extent possible
through improved measurement of key variables and ongoing model refinement. However,
significant uncertainty often remains, and emphasis is then placed on characterizing the nature of
that uncertainty and its impact on risk estimates. The characterization of uncertainty can be both
qualitative and, if a sufficient knowledge base is available, quantitative.
The characterization of uncertainty associated with risk assessment is ideally addressed in
the regulatory context using a tiered approach in which progressively more sophisticated
methods are used to evaluate and characterize sources of uncertainty depending on the overall
complexity and intended use of the risk assessment (WHO, 2008). Guidance documents
developed by EPA for assessing air toxics-related risk and Superfund Site risks as well as recent
guidance from WHO specify multitier approaches for addressing uncertainty.
Following the approach used for previous NAAQS risk and exposure assessments (U.S.
EPA, 2008, 2009b, 2010a, b), for the Os risk assessment, we are using a tiered framework
developed by WHO to guide the characterization of uncertainty. The WHO guidance presents a
four-tiered approach, where the decision to proceed to the next tier is based on the outcome of
the previous tier's assessment. The four tiers described in the WHO guidance include:
• Tier 0: recommended for routine screening assessments, uses default uncertainty factors
(rather than developing site-specific uncertainty characterizations);
• Tier 1: the lowest level of site-specific uncertainty characterization, involves qualitative
characterization of sources of uncertainty (e.g., a qualitative assessment of the general
magnitude and direction of the effect on risk results);
• Tier 2: site-specific deterministic quantitative analysis involving sensitivity analysis,
interval-based assessment, and possibly probability bounded (high-and low-end)
assessment; and
• Tier 3: uses probabilistic methods to characterize the effects on risk estimates of sources
of uncertainty, individually and combined.
With this four-tiered approach, the WHO framework provides a means for systematically
linking the characterization of uncertainty to the sophistication of the underlying risk assessment.
Ultimately, the decision as to which tier of uncertainty characterization to include in a risk
the effort to accurately characterize variability in key model inputs actually reflecting an effort to reduce
uncertainty.
3-8
-------
assessment will depend both on the overall sophistication of the risk assessment and the
availability of information for characterizing the various sources of uncertainty.
This risk and exposure assessment for the Os NAAQS review is relatively complex,
possibly warranting consideration of a full probabilistic (WHO Tier 3) uncertainty analysis. For
the exposure assessment, we include probabilistic representations of important sources of
variability; however, due to lack of information regarding reasonable alternative parameter
settings for model input variable distributions, we are not able to include a complete probabilistic
analysis incorporating both variability and uncertainty. Instead, we provide sensitivity analyses
to explore the impact of specific model assumptions, and we include a comprehensive qualitative
discussion of uncertainty regarding the model inputs and outputs.
While a full probabilistic uncertainty analysis is not undertaken for the epidemiology-
based risk assessment due to limits in available information on distributions of model inputs, we
provide a limited assessment using the confidence intervals associated with effects estimates
(obtained from epidemiological studies) to incorporate statistical uncertainty associated with
sample size considerations in the presentation of risk estimates. Technically, this type of
probabilistic simulation represents a Tier 3 uncertainty analysis, although as noted here, it will be
limited and only address uncertainty related to the fit of the C-R functions. Incorporation of
additional sources of uncertainty related to key elements of C-R functions (e.g., competing lag
structures, alternative functional forms, etc.) into a full probabilistic WHO Tier 3 analysis would
require that probabilities be assigned to each competing specification of a given model element
(with each probability reflecting a subjective assessment of the probability that the given
specification is the correct description of reality). However, for most model elements there is
insufficient information on which to base these probabilities. One approach that has been taken
in such cases is expert elicitation; however, this approach is resource- and time-intensive, and,
consequently, it is not feasible to use this technique in support of this Os risk assessment.
For most elements of the quantitative risk assessments, rather than conducting a full
probabilistic uncertainty analysis, we include a qualitative discussion of the potential impact of
uncertainty on risk results (WHO Tier 2). For some critical elements of the epidemiology-based
risk assessment, e.g., the effect-estimate in the C-R function, we include sensitivity analyses to
explore the potential impact of our assumptions. This falls under the WHO Tier 2 classification,
although we are not able to assign probabilities to the sensitivity analyses. For these sensitivity
analyses, we will include only those alternative specifications for input parameters or modeling
approaches that are deemed to have scientific support in the literature (and so represent
alternative reasonable input parameter values or modeling options). This means that the array of
risk estimates presented in this assessment is expected to represent reasonable risk estimates that
3-9
-------
can be used to provide some information regarding the potential impacts of uncertainty in the
model elements.
3.4 CHARACTERIZATION OF AIR QUALITY
Figure 3-1 diagrams the basic information used in developing the air quality inputs for
the HREA. Air quality inputs to the urban area exposure and risk assessments include (1) recent
(2006 - 2010) air quality data developed from Os ambient monitors in each selected urban study
area and (2) simulated air quality that reflects changes in the distribution of Os air quality
estimated to occur when the urban area just meets the existing or alternative Os standards under
consideration. In addition, Os air quality surfaces for recent years covering the entire continental
U.S. were generated for use in the national-scale assessment. Details of the air quality data used
in the HREA are discussed in Chapter 4.
Precursor Emissions
1
V ^
Ozone Air Quality Data
V J
\
f
X
National Ambient \
O3 Spatial Fields: 1
recent conditions /
*^ -—^
Model-based O3 Sensitivities
V J
\
t
/ O3 Metrics in Urban Case
*J Study Areas: recent conditions
\ and after just meeting
V existing and alternative standards
Figure 3-1. Conceptual Diagram for Air Quality Characterization in the HREA.
The urban-scale exposure and risk analyses are based on five recent years of air quality
data, 2006-2010. We are including five years to reflect the considerable variability in
meteorological conditions and the variation in Os precursor emissions that have occurred in
recent years. The analyses focus on the Os season in each study area, which ranges from April to
October in much of the nation but is longer in some warmer areas such as Los Angeles and
Houston. The required Os monitoring seasons for the urban study areas are described in more
detail in Chapter 4.
In developing the Os air quality surfaces for the national-scale analysis, a combination of
monitoring data and modeled Os concentrations are used to provide greater coverage across the
3-10
-------
U.S. The procedure for fusing Os monitor data with modeling results is described further in
Chapter 4.
Several Os metrics are generated for use in the urban study area exposure and risk
analyses. The exposure analyses use hourly Os concentrations, while the risk analyses use
several different averaging times. The specific metrics used in each analysis are discussed further
in following chapters. For the exposure analysis, hourly Os concentrations are interpolated to
census tracts using Voronoi neighbor averaging (VNA), a distance weighted interpolation
method (Gold, 1997; Chen et al., 2004). For the epidemiology-based risk analysis, we developed
a composite of all monitors in the urban area for application with health effect estimates from
epidemiology studies. We also evaluated several different definitions of the spatial boundaries of
the urban areas that determined the monitors included in the spatial average. Some of the
epidemiological studies specify a relatively narrow set of counties within an urban area, while
others use a broader definition, such as all counties in a core based statistical area (CBSA) as
defined by the Census Bureau. For those epidemiological studies that used a relatively narrow set
of counties, most were based on counties in the center of the urban area. In most of these areas,
the non-attaining Os monitors are not located in the center of the urban area, but instead in the
surrounding areas, reflecting the transport and atmospheric chemistry governing Os formation.
As a result, using a monitor set that exactly reflects the specific counties used in the
epidemiology studies can exclude counties in an urban area that would realize the most risk
reduction resulting from just meeting the Os standard. To better represent the changes in risk that
could be experienced in the urban areas, the core risk estimates for all endpoints will be based on
the CBSA definition. Sensitivity analyses are included to evaluate the effect of using only the
counties in each urban area that specifically match the county set used in the epidemiology
studies.
Simulation of just meeting the existing and alternative Os standards is accomplished by
adjusting hourly Os concentrations measured over the Os season using a model-based adjustment
methodology that estimates Os sensitivities to precursor emissions changes.6 These sensitivities,
which estimate the response of Os concentrations to reductions in anthropogenic NOx and VOC
emissions, are developed using the Higher-order Decoupled Direct Method (FtDDM) capabilities
in the Community Multi-scale Air Quality (CMAQ) model. This modeling approach incorporates
all known emissions, including sources of natural and anthropogenic emissions in and outside of
the U.S. By using the model-based adjustment methodology we are able to more realistically
6 In the first draft of this HREA, we used a statistical quadratic rollback approach to simulate just meeting the
existing Os standards. In that draft, we proposed using the model based approach that is being used in this draft,
and received support for the model based approach from CASAC.
3-11
-------
simulate the temporal and spatial patterns of Os response to precursor emissions. We chose to
simulate just meeting the existing and alternative standards in the urban study areas by
decreasing U.S. anthropogenic emissions of NOx exclusively, though in a few instances,
proportional decreases in both NOx and VOC were applied. This approach was used to avoid any
suggestion that we are approximating a specific emissions control strategy that a state or urban
area might choose to meet a standard. More details on the HDDM-adjustment approach are
presented in Chapter 4 of this HREA and in Simon et al. (2013).
In the previous review, background Os (referred to in that review as policy relevant
background, or PRB) was incorporated into the HREA by calculating risk only in excess of PRB.
CASAC members recommended that EPA move away from using PRB in calculating risks
(Henderson, 2007). In addition, comments received from CASAC, based on their review of the
Risk and Exposure Assessments (Frey and Samet, 2012a; Frey, 2014), agreed with the
development of risk estimates with reference to zero Os concentration. Based on these
recommendations and comments, the HREA includes risks associated with Os from all sources
after we have simulated just meeting the existing standard and estimates of total risk remaining
after meeting alternative levels of the standards. EPA believes that presenting total risk is most
relevant given that individuals and populations are exposed to total Cb from all sources, and risks
associated with Cb exposure are due to total Os exposure and do not vary for Os exposure
associated with any specific source. In addition, background Os is fully represented in estimates
of total risk given that the measured and adjusted air quality concentrations being used in the risk
and exposure analyses include Os produced from precursor emissions from both anthropogenic
and background sources. The evidence and information on background Os that is assessed in the
Integrated Science Assessment (ISA) is considered in the Policy Assessment (PA) in conjunction
with the total risk estimates provided in this HREA. With regard to background Os
concentrations, the PA will consider available information on ambient Os concentrations
resulting from natural sources, anthropogenic sources outside the U.S., and anthropogenic
sources outside of North America.
In providing a broader national characterization of Os air quality in the U.S., this HREA
draws upon air quality data analyzed in the Os ISA as well as national Os databases and
modeling of Os using the Community Multiscale Air Quality (CMAQ) model. This information,
along with additional analyses, is used to develop a broad characterization of recent air quality
across the nation. This characterization includes Os levels in the urban study areas for the time
periods relevant to the risk analysis and information on the spatial and temporal characterization
of Os across the national monitoring network. This information is then used to place the relative
comparative attributes of the selected study areas into a broader national comparative context to
help judge the overall representativeness of the selected study areas in characterizing Os risk for
3-12
-------
the nation. In addition, to better characterize the spatial patterns of responses of the distribution
of Os to just meeting existing and alternative Os standards, we also provide assessments of the
historical patterns of responses of Os to emissions changes overtime and an assessment of
national patterns of responses to emissions changes relative to the spatial distribution of
populations. These analyses are presented in more detail in Chapter 8 and Chapter 8 appendices.
3.5 CHARACTERIZATION OF URBAN-SCALE HUMAN EXPOSURE
Figure 3-2 diagrams the basic structure of the population exposure assessment. Basic
inputs to the exposure assessment include the following: (1) recent (2006 - 2010) measurements
of Os concentrations from monitors in each selected urban study area; (2) model adjusted Os
concentrations that reflect changes in the distribution of Os air quality estimated to occur when
an area just meets the existing or alternative Os standards under consideration; (3) population and
demographic information, e.g., age, gender, etc.; (4) time-location activity pattern data; and (5)
physiological data, e.g., body mass index, ventilation rates, lifestage development, etc. Basic
outputs include numbers and percent of people with Os exposures exceeding health-based
benchmark levels and time-series of Cb exposures and ventilation rates for individuals (for use in
the lung function risk analysis). Details of the exposure modeling are discussed in Chapter 5.
O3 Air Quality for Recent Conditions,
and After Just Meeting Existing
and Alternative Standards
Population and
Demographic Information
Time Activity
Pattern Data
Physiological
Data
Estimation of O3
Exposure
Concentrations
_v
Number and % of persons with 8-
hour average exposures at or
above selected benchmark level
per year while at moderate or
greater exertion
Time-series of O3 exposure
concentrations and ventilation
rate for individuals
Figure 3-2. Conceptual Diagram for Population Exposure Assessment.
3-13
-------
The scope of the exposure assessment includes 15 urban study areas.7 These areas were
selected to be generally representative of U.S. populations, geographic areas, climates, and
different Os and co-pollutant levels, and they include all of the urban study areas used in the
epidemiology-based risk analysis (see Chapter 7). Three additional urban study areas are
included in the exposure modeling beyond those included in the epidemiology-based risk
analysis. These urban study areas are included to provide additional information on
heterogeneity in exposure but could not be included in the epidemiology-based risk analysis
because those analyses require additional information not available in the three additional urban
study areas. In addition to providing population exposures for estimation of lung function effects,
the exposure modeling provides a characterization of urban air pollution exposure environments
and activities resulting in the highest exposures.
Population exposure to ambient Os levels is evaluated using version 4.5 of EPA's APEX
model. Exposures are estimated using recent ambient Os concentrations, based on 2006-2010 air
quality data, and for Os concentrations resulting from simulations of just meeting the existing 8-
hr Os standard and alternative Os standards, based on adjusting 2006-2010 air quality data.
Because the Os standard is based on the 3-year average of the 4th highest daily maximum 8-hr
average Os concentration, we simulate just meeting the standard for two three-year periods,
2006-2008 and 2008-2010. Eight-hour average Os exposures of interest in this assessment are
based on results of controlled human exposure studies indicating adverse health effects
associated with of 60, 70 and 80 ppb (section 2.2.5). The occurrence of exposure at or above
these health effect benchmark levels are estimated for all school-age children (ages 5 to 18),
asthmatic school-age children, asthmatic adults (ages 19 to 95), and older adults (ages 65 to 95).
This choice of population groups includes a strong emphasis on children, asthmatics, and people
> 65 years old and reflects the finding of the last Cb NAAQS review (U.S. EPA, 2007a,b,c) and
the ISA (U.S. EPA, 2013a, Chapter 8) that these are important at-risk groups.
In addition to estimating the frequency of population exposures exceeding health-based
exposure benchmarks, estimated exposures are used as an input to the portion of the health risk
assessment that is based on E-R relationships derived from controlled human exposure studies.
The exposure analysis also provides a characterization of people with high exposures in terms of
exposure environments and activities. In addition, the exposure analysis offers key observations
' These urban study areas are Atlanta, GA; Baltimore, MD; Boston, MA; Chicago, IL; Cleveland, OH; Dallas, TX;
Denver, CO; Detroit, MI; Houston, TX; Los Angeles, CA; New York, NY; Philadelphia, PA; Sacramento, CA; St.
Louis, MO; and Washington, D.C. We also considered included Seattle; however, the available monitoring data
was not sufficient to accurately characterize O3 exposures in the Seattle area.
3-14
-------
based on the results of the APEX modeling, viewed in the context of factors such as averting
behavior and key uncertainties and limitations of the exposure modeling.
3.6 CHARACTERIZATION OF URBAN-SCALE HEALTH RISKS BASED ON
CONTROLLED HUMAN EXPOSURE STUDIES
The major components in the lung function risk assessment are shown in Figure 3-3.
Basic inputs to the analysis include (1) individual exposure to ambient Cb derived from the
exposure modeling described in Section 3.5., (2) data from controlled human exposure studies,
used to construct E-R functions, (3) physiological data, including body mass index, age, etc., and (4)
exercise levels, which determine breathing rates and affect dose. Basic outputs include the percentage
of a selected study group, e.g., children with asthma, with predicted lung function decrements
(measured as decrements in forced expiratory volume in one second, or FEVi), greater than or equal
to 10, 15, and 20 percent, for recent (2006-2010) Os levels and for Os levels adjusted to just meeting
existing and alternative standards.
Exposure Estimates:
Individual and Population
Data from controlled
human exposure studies
Lung Function
Exposure-Response
Models
v
Physiological
parameters
Exercise levels
and duration
% of population with AFEVi>10,15,20%
for recent O3 and afterjust meeting
existing and alternative standards
Figure 3-3. Conceptual Diagram of Os Lung Function Health Risk Assessment Based on
Controlled Human Exposure Studies.
Prior EPA Cb risk assessments have included risk estimates for lung function decrements
and respiratory symptoms based on analysis of individual data from controlled human exposure
studies. The current assessment applies probabilistic E-R relationships that are based on analyses
3-15
-------
of individual data that describe the relationship between a measure of personal exposure to Os
and the measure(s) of lung function recorded in the study. The current quantitative lung function
risk assessment presents only a partial picture of the risks to public health associated with short-
term Os exposures, as there are additional controlled human exposure studies that have evaluated
cardiovascular and neurological outcomes due to Os exposure. However, these studies do not
provide sufficient information with which to generate E-R functions and therefore are not
suitable for quantitative risk assessment. Furthermore, additional adverse health response
indicators have been measured in a number of studies, such as markers of inflammation indicated
by the presence of neutrophils in bronchoalveolar lavage fluid (e.g., Devlin et al., 1991; see U.S.
EPA, 2013a, section 5.3.3). However, the use of generally higher exposures observed to generate
a significant response (> 80 ppb) and the limited number of studies reporting such information
prevent the development of an E-R function considered useful in quantitatively estimating health
risks associated with these additional adverse health endpoints in this assessment.
Modeling of risks of lung function decrements is also performed by APEX using a risk
module developed using results from controlled human exposure studies. These studies involve
volunteer subjects, most often healthy adults, who are exposed to specified levels of Os under
controlled conditions for specified amounts of time, all while engaged in different exercise
regimens. The responses measured in such studies have included measures of lung function, such
as forced expiratory volume in one second (FEVi), respiratory symptoms, airway hyper-
responsiveness, and inflammation. In addition to estimating lung function decrements for healthy
adults that were the study groups in the controlled human exposure studies, this lung function
risk assessment estimates lung function decrements (> 10, > 15, and > 20% changes in FEVi) in
school-age children (ages 5 to 18). The lung function estimates for children are based on
applying data from young adult study subjects (ages 18 to 35) to school-age children. This is
based on findings from other chamber studies and summer camp field studies documented in the
1996 O3 Staff Paper (U.S. EPA, 1996a) and 1996 O3 Criteria Document (U.S. EPA, 1996b), that
lung function changes in healthy children are similar to those observed in healthy young adults
exposed to Os under controlled chamber conditions. Further, model-estimated lung function
decrements using the current risk model were similar to those from a controlled clinical study of
children ages 8-11, further supporting this approach (McDonnell et al., 1985; see 6.4.2 for
details). Risk metrics estimated for lung function risk include the number and percent of all
school-age children and other study groups experiencing one or more occurrences of a lung
function decrement > 10, > 15, and > 20% in an Os season and the total number of occurrences
of these lung function decrements in school-age children and active school-age children.
The risk assessment includes two different modeling approaches. The first approach
employs a model that estimates FEVi responses for individuals associated with short-term
3-16
-------
exposures to Os (McDonnell et al., 2012). This model is based on the data from controlled
human exposure studies included in the prior lung function risk assessment as well as additional
data sets that considered different measurement averaging times and breathing rates of study
subjects. These data were from 23 controlled human Os exposure studies that included exposure
of 742 young adult volunteers (ages 18 to 35) (see McDonnell et al., 2007, 2010, and 2012, for a
description of these data). Outputs from this model include FEVi decrements for each simulated
individual for each day, which can be used to calculate the distribution of FEVi decrements for
the study group, and the percent of the study group having FEVi decrements > 10, > 15, and >
20% after air quality is adjusted to just meet the existing and alternative standards. The results
generated using this individual-based lung function risk modeling approach are the core results
reported in this portion of the risk assessment.
In addition, for the purpose of comparison to previous lung function decrement modeling,
we are applying the approach used in the last review that employed a probabilistic population-
level E-R function derived from the results of a number of controlled human exposure studies
(U.S. EPA, 2007a). This modeling approach uses a smaller set of controlled human exposure
studies from which the E-R functions were derived and uses the population distribution of Os
exposures (not individual exposures) to directly estimate the percent of the population with
moderate levels of exertion with lung function decrements > 10, > 15, and > 20%.
Because controlled human exposure studies are conducted in laboratory settings, they are
generally not restricted to particular conditions that might be unique to a select geographic
location (e.g., meteorology). Thus, the health results from controlled human exposure studies can
appropriately be applied to any study area for which there are adequate air quality data on which
to base the modeling of personal exposures. For this assessment, we have selected 15 urban
study areas (matching the areas used in the exposure analysis), representing a range of
geographic areas, population demographics, and Os climatology. These 15 areas also include the
12 urban study areas evaluated in the risk analyses based on C-R relationships developed from
epidemiological or field studies.
In the controlled human exposure study based risk assessment, there are two broad
sources of uncertainty to the risk estimates. One of the important sources of uncertainty is the
estimation of the population distribution of individual time series of Os exposures and ventilation
rates; these uncertainties are addressed as part of the exposure assessment. The second broad
source of uncertainty in the risk calculation results from uncertainties in the lung function risk
model. Sensitivity analyses are conducted to inform a qualitative discussion of these
uncertainties.
3-17
-------
3.7 CHARACTERIZATION OF URBAN-SCALE HEALTH RISK BASED ON
EPIDEMIOLOGICAL STUDIES
The major components of the portion of the urban-scale health risk assessment based on
data from epidemiological studies are illustrated in Figure 3-4. Basic inputs to this analysis
include (1) measured Os concentrations for recent (2007 and 2009) conditions and adjusted air
quality representing Os concentrations after just meeting existing and alternative standards, (2)
C-R functions derived from epidemiological studies evaluating associations between Os
concentrations and mortality and morbidity endpoints and (3) population counts and baseline
incidence rates for mortality and morbidity endpoints. Basic outputs for each urban area include
estimates of Os-attributable incidence and percent Os-attributable incidence for selected
mortality and morbidity endpoints and changes and percent changes in Os-attributable incidence.
O3 Air Quality for Recent Conditions,
and After Just Meeting Existing
and Alternative Standards
National Long-term Exposure
C-R Functions from
Epidemiological Studies
Location-specific Short-term
Exposure C-R Functions
From Epidemiological Studies
_V
BenMAP estimation of O3
Attributable Incidence of Mortality
And Morbidity
Urban Area Population and
Baseline Health Data
C Urban Area Estimates of % \
Os Attributable Incidence and
Change in % O3 attributable
;idence of mortality and morbidity /
/ Urban Area Estimates of Os
attributable incidence of mortality
and morbidity and change in O3
V attributable incidence
Figure 3-4. Conceptual Diagram of Urban-Scale Health Risk Assessment Based on Results
of Epidemiology Studies.
Epidemiological and field studies provide estimated C-R relationships based on data
collected in —ambient concentration settings. Ambient Os concentrations used in these studies
are typically spatial averages of monitor-specific measurements, using population-oriented
monitors. Population health responses associated with ambient Os have included population
counts of school absences, emergency room visits, hospital admissions for respiratory and
cardiac illness, respiratory symptoms, and premature mortality. Risk assessment based on
epidemiological studies typically requires baseline incidence rates and population data for the
3-18
-------
risk assessment study areas. To minimize uncertainties introduced by extrapolation, a risk
assessment based on epidemiological studies can be performed for the study areas in which the
original epidemiological studies were carried out, rather than extrapolating results to study areas
where studies for a particular health endpoint have not been conducted.
The set of urban study areas included in this portion of the risk assessment was chosen to
provide population coverage and to capture the observed heterogeneity in Os-related risk across
selected urban study areas. In addition, study areas had to have at least one epidemiological
study conducted in order for the study area to be included for a specific endpoint. This
assessment also evaluates the mortality risk results for the selected urban study areas within a
broader national context to better characterize the nature, magnitude, extent, variability, and
uncertainty of the public health impacts associated with Os exposures. This national-scale
assessment is discussed in the next section.
The risk assessment based on application of results of epidemiological studies is
implemented using the environmental Benefits Mapping and Analysis Program (BenMAP,
version 4.0 (U.S. EPA, 2013b), EPA's GIS-based computer program for the estimation of health
impacts associated with air pollution. BenMAP draws upon a database of population, baseline
incidence, ambient concentration data, and effect estimates (regression coefficients) to automate
the calculation of health impacts. We selected 2007 and 2009 as analysis years for the urban-
scale risk analysis. These two years are the midpoint years in the two three-year periods 2006-
2008 and 2008-2010. Year 2007 represents a year with generally higher Os concentrations, and
2009 represents a year with generally lower Cb concentrations. Analyses for these two years will
provide a reasonable representation of the effects of baseline Os concentrations on the risk
estimates.
This risk assessment is focused on health effect endpoints for which the weight of the
evidence as assessed in the Os ISA supports the causal determination that a likely causal or
causal relationship exists between a specific health effect category and exposure to Os. The
analysis includes estimates of mortality risk associated with daily maximum 8-hr average or
means of daily 8-hr average Os concentrations in all 12 urban study areas, as well as risk of
hospitalization for chronic obstructive pulmonary disease and pneumonia. In addition, the
analysis includes analysis of hospitalizations for additional respiratory diseases in Los Angeles,
New York City, and Detroit, due to limited availability of epidemiological studies including
these endpoints across the 12 urban areas. The analysis also evaluates risks of respiratory related
emergency department visits in Atlanta and New York City and risks of respiratory symptoms in
Boston, again based on limited availability of epidemiological studies that included these
particular endpoints in their analyses and that were conducted in the same urban study areas
3-19
-------
evaluated in this risk assessment. Table 3-1 summarizes the endpoints evaluated for each of the
12 urban study areas.
Table 3-1. Short-term Os Exposure Health Endpoints Evaluated in Urban Study Areas.
Urban Study
Area
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Mortality
X
X
X
X
X
X
X
X
X
X
X
X
COPD1 and
Pneumonia
Hospitalizations
X
X
X
X
X
X
X
X
X
X
X
X
Other
Respiratory
Hospitalizations
X
X
X
Respiratory
Related ED1
visits
X
X
Respiratory
Symptoms
X
1 COPD - chronic obstructive pulmonary disorder; ED - hospital emergency department. The specific health
endpoints evaluated in our assessment for particular study areas are for where appropriate response measurements
were made in the original epidemiological studies, i.e., most studies did not evaluate the relationship between
ambient O3 and other respiratory hospitalizations, respiratory related ED visits, and respiratory symptoms.
This analysis will also estimate the respiratory mortality risks associated with longer-term
exposures to Os. This is supported by the Os ISA, which concluded that the evidence for long-
term exposures to Os as likely to be causally related to respiratory effects, including respiratory
mortality and morbidity. There is one national study of long-term exposures and respiratory
mortality that provides a C-R function for use in our risk assessment. Several other studies have
examined long-term exposures and cardiopulmonary mortality, but consistent with the Os ISA,
we focused on respiratory mortality because of the additional supporting evidence related to
long-term exposure and morbidity. Because the long-term exposure C-R function is based on
comparing Os and mortality across urban areas, the same C-R function is applied in each of the
12 urban study areas. The available epidemiological studies evaluating long-term Os exposures
and morbidity endpoints do not provide information that can be used to develop suitable C-R
functions. As a result, we are not including quantitative risk estimates for morbidity associated
with long-term exposures.
3-20
-------
We have identified multiple options for specifying the C-R functions for particular health
endpoints. This risk assessment provides an array of reasonable estimates for each endpoint
based on the available epidemiological evidence. This array of results provides a limited degree
of information on the variability and uncertainty in risk due to differences in study designs,
model specification, and analysis years, amongst other differences.
As part of the risk assessment, we address both uncertainty and variability. We provide a
limited probabilistic characterization of uncertainty in the national-scale mortality risk estimates
using the confidence intervals associated with effects estimates (obtained from epidemiological
studies). However, this addresses only one source of uncertainty. For other sources of
uncertainty, we include a number of sensitivity analyses to evaluate the impact of alternative
approaches to simulating just meeting existing and alternative standards, alternative C-R
functions, definitions of Cb seasons to which C-R functions are applied, and definitions of urban
areas to which the C-R functions are applied. In addition, we evaluate the impact in a subset of
locations of using co-pollutant C-R functions. In the case of variability, we identify key sources
of variability associated with Os risk (for both short-term and long-term exposure-related
endpoints included in the risk assessment) and discuss the degree to which these sources of
variability are reflected in the design of the risk assessment. Finally, we also include a
comprehensive qualitative assessment of uncertainty and variability.
We also provide a representativeness analysis (see Chapter 8) designed to support the
interpretation of risk estimates generated for the set of urban study areas included in the risk
assessment. The representativeness analysis focuses on comparing the urban study areas to
national-scale distributions for key Os-risk related attributes (e.g., demographics including
socioeconomic status, air-conditioning use, baseline incidence rates and ambient Os levels). The
goal of these comparisons is to assess the degree to which the urban study areas provide
coverage for different regions of the country as well as for areas likely to experience elevated Os-
related risk due to their specific mix of Cb-risk related attributes.
3.8 NATIONAL-SCALE MORTALITY RISK ASSESSMENT
The major components of the national-scale mortality risk assessment are shown in
Figure 3-5. Basic inputs to this analysis are similar to those for the urban-scale epidemiology-
based assessment and include (1) gridded Os concentrations over the continental U.S. for recent
(2007) conditions, (2) C-R functions relating long-term and short-term exposures to Os to
mortality, and (3) population and baseline mortality rates. Basic outputs include county and
national estimates of incidence and percent of mortality attributable to Os.
3-21
-------
The national-scale mortality risk assessment serves two primary purposes. First, it serves
as part of the representativeness analysis discussed above, providing an assessment of the degree
to which the urban study areas included in the risk assessment provide coverage for areas of the
country expected to experience elevated mortality rates due to Os-exposure. Second, it provides a
broader perspective on the distribution of risks associated with recent Os concentrations
throughout the U.S., and provides a more complete understanding of the overall public health
burden associated with Os.8 We note that a national-scale assessment such as this was completed
for the risk assessment supporting the latest PM NAAQS review (U.S. EPA, 2010b) with the
results of the analysis being used to support an assessment of the representativeness of the urban
study areas assessed in the PM NAAQS risk assessment, as described here for Os.
National Long-term Exposure
C-R Functions from
Epidemiological Studies
Nationwide Set of City
Specific Short-term Exposure
C-R Functions from
Epidemiological Studies
National ambient Os
12 km2 gridded spatial field
for recent conditions
BenMAP estimation of Os
Attributable Incidence of Mortality
Nationwide County Specific
Population and Baseline
Health Data
/ County and national level \
I estimates of burden of premature I
V mortality attributable to Os /
/ County and national level \
I estimates of % of total mortality I
V attributable to O3 /
Figure 3-5. Conceptual Diagram of National Os Mortality Risk Assessment Based on
Results of Epidemiology Studies.
The national-scale risk assessment focuses on mortality only, due to the availability of
large multi-city epidemiology studies for short-term mortality and the availability of a long-term
mortality study which provides information to develop a suitable C-R function. As noted in the
8 In the previous Os NAAQS review, CASAC commented that "There is an underestimation of the affected
population when one considers only twelve urban "Metropolitan Statistical Areas" (MSAs). The CASAC
acknowledges that EPA may have intended to illustrate a range of impacts rather than be comprehensive in their
analyses. However, it must be recognized that Os is a regional pollutant that will affect people living outside these
12 MSAs, as well as inside and outside other urban areas." Inclusion of the national-scale mortality risk
assessment partially addresses this concern by providing a broader characterization of risk for an important Os
health endpoint.
3-22
-------
discussion of the urban-scale analyses, the available epidemiological studies evaluating long-
term Os exposures and morbidity endpoints do not provide information that can be used to
develop suitable C-R functions. In the case of short-term morbidity endpoints, the available
epidemiological studies are generally located in only a few urban areas and, even in the case of
the multi-city hospitalization studies, cover only a small fraction of the urban areas in the U.S. In
addition, baseline mortality rates are available for every county in the U.S., while baseline
hospitalization rates are available in only a small subset of counties. For these reasons, the
national-scale risk assessment includes only mortality associated with short- and long-term Os
exposures.
We provide a limited probabilistic characterization of uncertainty in the national-scale
mortality risk estimates using the confidence intervals associated with effects estimates
originally obtained from epidemiological studies. However, this addresses only one source of
uncertainty. To address some other key potential sources of uncertainty in the national-scale
assessment, we conduct sensitivity analyses. Risk estimates are provided for two alternative C-R
functions for short-term exposure, reflecting two multi-city epidemiological studies. For short-
term exposure-related mortality, the assessment provides several estimates of national mortality
risk, including a full national-scale estimate including all counties in the continental U.S., and an
analysis restricted to the set of urban areas included in the time-series studies that provide the
effect estimates. We have greater confidence in the analysis based on the large urban areas
included in the epidemiological studies, but the information from the full analysis of all counties
is useful to gain understanding of the potential magnitude of risk in less urbanized areas. In
addition, the national-scale mortality risk assessment evaluates the sensitivity of the nationwide
estimates to assumptions about the transferability of effect estimates from the cities included in
the underlying epidemiological studies to other cities in the U.S. Finally, the assessment includes
a sensitivity analysis evaluating the use of regional priors city—rather than using a national prior
in developing the city specific Bayesian adjusted effect estimates.9 These sensitivity analyses are
described in detail in Chapter 8.
The national-scale risk assessment is conducted only for recent Os conditions. We do not
attempt to simulate nationwide Os concentrations that would result from just meeting the existing
or alternative Os standards everywhere in the U.S. Such a simulation would require detailed
modeling of attainment strategies in all potential non-attainment areas and would need to take
into account the interdependence of Os concentrations across urban areas. This type of analysis is
9 In multi-city Bayesian analyses, it is necessary to specify initial values or "priors" which are then "updated" using
information from the individual city specific estimates. These priors are generally a mean value across all of the
cities, in this case, cities in regions or cities across the nation.
3-23
-------
beyond the scope of this current risk assessment. Analyses of nationwide attainment are included
as part of the Regulatory Impact Analyses that accompany proposed and final rulemaking
packages and will likely be included in the rulemaking portion of this review.
3.9 PRESENTATION OF EXPOSURE AND RISK ESTIMATES TO INFORM THE
OZONE NAAQS POLICY ASSESSMENT
We present exposure estimates in three ways: person-occurrences, number, and percent
of people in different study groups (e.g., asthmatic adults, all school-age children, asthmatic
school-age children, older adults) with at least one 8-hr average exposure at or above benchmark
levels of 60 ppb, 70 ppb, and 80 ppb. In addition, the same types of results are shown for people
experiencing multiple 8-hr average Os exposures at or above the benchmark levels. The results
are presented in summary tables and graphics, while detailed tables of results are provided in an
appendix. The focus in the presentation of results is on exposures occurring after simulating just
meeting the existing standard and on the change in number and percent of exposures between
meeting the existing standard and meeting alternative standards. Results are presented for the
five modeled years, for all 15 urban study areas.
Quantitative risk estimates from the analyses based on application of results from
controlled human exposure studies are presented for the two different risk models. For each
model, we provide estimates of the percent of different study groups (all school-age (5-18)
children, asthmatic children, and adults) with lung function decrements greater than or equal to
10, 15, and 20 percent. As with exposure, the focus in the presentation of results is on risk
occurring after simulating just meeting the existing standards and on the change in risk occurring
between meeting the existing standard and meeting alternative standards.
Results from the epidemiology-based risk assessment are presented in two ways: (1) total
(absolute) health effects incidence for recent air quality and simulations of air quality just
meeting the existing and alternative standards under consideration and (2) risk reduction
estimates, reflecting the change in the distribution of Os between scenarios of just meeting the
existing standard and just meeting alternative standards. In addition, risks are presented as the
percent of baseline incidence, and risks per 100,000 population, to allow for comparisons
between urban areas with very different population sizes. We include risk modeled across the
full distribution of Os concentrations, as well as core risk estimates for Os concentrations down
to 0 ppb.
We present an array of risk estimates in order to provide additional context for
understanding the potential impact of uncertainty on the risk estimates. For core estimates and
sensitivity analyses, we provide the statistical confidence intervals, demonstrating the relative
3-24
-------
precision of estimates. The graphical presentation of sensitivity analyses focuses on the
differences from the core estimates in terms of risk per 100,000 population.
The results of the representativeness analysis are presented using cumulative probability
plots (for the national-level distribution of Os risk-related parameters) with the locations where
the individual urban study areas fall within those distributions noted in the plots using vertical
lines. Similar types of plots are used to present the distribution of national-scale mortality
estimates based on the national-scale risk assessment, showing the location of the urban study
areas within the overall national distribution.
Chapter 9 of this risk and exposure assessment provides a synthesis of the results from
the four assessments (urban-scale exposure, urban-scale lung function risk, urban-scale
epidemiology-based risk, and national-scale mortality risk). Chapter 9 focuses on comparing
patterns of results across locations, years, and alternative standards. Chapter 9 also provides
perspective on the overall degree of confidence of the analyses and the representativeness of the
set of results in characterizing patterns of exposure and risk and patterns of changes in exposure
and risk from just meeting alternative standards relative to just meeting the existing standards.
3-25
-------
3.10 REFERENCES
Chen, J., R. Zhao and Z. Li. 2004. Voronoi-based k-order neighbor relations for spatial analysis.
ISPRS J Photogrammetry Remote Sensing. 59(1-2), 60-72.
Devlin, R.B.; W.F McDonnell; R. Mann; S. Becker; D.E. House; D. Schreinemachers; H.S.
Koren. 1991. Exposure of humans to ambient levels of ozone for 6.6 hours causes
cellular and biochemical changes in the lung. American Journal of Respiratory Cell and
Molecular Biology. 4(1):72-81.
Frey, C. and J. Samet. 2012a. CASAC Review of the EP'A's Health Risk and Exposure Assessment
for Ozone (First External Review Draft - Updated August 2012) and Welfare Risk and
Exposure Assessment for Ozone (First External Review Draft - Updated August 2012). U.S.
Environmental Protection Agency Science Advisory Board. EPA-CASAC-13-002.
Frey, C. and J. Samet. 2012b. CASAC Review of the EPA's Policy Assessment for the Review
of the Ozone National Ambient Air Quality Standards (First External Review Draft -
August 2012). U.S. Environmental Protection Agency Science Advisory Board. EPA-
CASAC-13-003.
Frey, C. 2014. CASAC Review of the EPA's Health Risk and Exposure Assessment for Ozone
(SecondExternal Review Draft - February, 2014). U.S. Environmental Protection Agency
Science Advisory Board. EPA-C AS AC-14-005.
Gold, C. 1997. Voronoi methods in GIS. In: Algorithmic Foundation of Geographic Information
Systems (va Kereveld M.; J. Nievergelt; T. Roos; P. Widmayer, eds.). Lecture Notes in
Computer Science, Vol 1340. Berlin: Springer-Verlag, pp 21-35.
Henderson, R. 2007. Clean Air Scientific Advisory Committee's (CASAC) Review of the
Agency's Final O3 Staff Paper. EPA-C AS AC-07-002. March 26, 2007.
Langstaff, I.E. 2007. OAQPS Staff Memorandum to Ozone NAAQS Review Docket (EPA HQ-
OAR-2005-0172). Subject: Analysis of Uncertainty in Ozone Population Exposure
Modeling. [January 31, 2007].
.
McDonnell, W.F.; R.S. Chapman; M.W. Leigh; G.L. Strope; and A.M. Collier. 1985.
Respiratory responses of vigorously exercising children to 0.12 ppm ozone exposure.
American Review of Respiratory Disease. 132:875-879.
McDonnell, W.F.; P.W. Stewart and M.V. Smith. 2007. The temporal dynamics of ozone-
induced FEVi changes in humans: an exposure-response model. Inhalation Toxicology.
19:483-494.
McDonnell W.F.; P.W. Stewart and M.V. Smith. 2010. Prediction of Os -induced lung function
responses in humans. Inhalation Toxicology, 22(2): 160-8.
McDonnell, W.F.; P.W. Stewart; M.V. Smith; C.S. Kim andE.S. Schelegle. 2012. Prediction of
lung function response for populations exposed to a wide range of ozone conditions.
Inhalation Toxicology. 24: 619-633.
Samet, J. 2009. Consultation on EPA's Draft Integrated Review Plan for the National Ambient
Air Quality Standards for Particulate Matter. (EPA document number EPA-C AS AC-10-
004. December 3).
3-26
-------
Samet, J. 2011. Consultation on EPA's Os National Ambient Air Quality Standards: Scope and
Methods Plan for Health Risk and Exposure Assessment (April 2011) and Ozone
National Ambient Air Quality Standards: Scope and Methods Plan for Welfare Risk and
Exposure Assessment (April 2011). EPA-CASAC-11-008. June 21.
Simon H.; K.R. Baker; F. Akhtar; S.L. Napelenok; N. Possiel; B. Wells and B. Timin. 2013. A
direct sensitivity approach to predict hourly ozone resulting from compliance with the
national ambient air quality standard. Environmental Science and Technology.
47(5):2304-13.
U.S. EPA. 1996a. Review of National Ambient Air Quality Standards for Ozone: Assessment of
Scientific and Technical Information - OA QPS Staff Paper. (EPA document number
EPA/452/R-96-007). Research Triangle Park, NC: Office of Air Quality Planning and
Standards. Available from: NTIS, Springfield, VA; PB96-203435. <
http://www.epa.gov/ttn/naaqs/standards/ozone/s_O3_pr.html>.
U.S. EPA. 1996b. Air Quality Criteria for Ozone and Related Photochemical Oxidants. (EPA
document number EPA/600/P-93/004aF-cF). Research Triangle Park, NC: Office of
Research and Development, National Center for Environmental Assessment.
.
U.S. EPA. 2007a. Ozone Population Exposure Analysis for Selected Urban Areas. Research
Triangle Park, NC: Office of Air Quality Planning and Standards. (EPA document
number EPA-452/R-07-010).
.
U.S. EPA. 2007b. Ozone Health Risk Assessment for Selected Urban Areas. Research Triangle
Park, NC: Office of Air Quality Planning and Standards. (EPA document number EPA
452/R-07-009).
.
U.S. EPA. 2007c. Review of the National Ambient Air Quality Standards for Ozone: Policy
Assessment of Scientific and Technical Information. OA QPS Staff Paper. Research
Triangle Park, NC: Office of Air Quality Planning and Standards. (EPA report
numberEPA-452/R-07-007.
.
U. S. EPA. 2008. Risk and Exposure Assessment to Support the Review of the NO2 Primary
National Ambient Air Quality Standard. (EPA document number EPA-452/R-08-008a).
.
U.S. EPA. 2009a. Integrated Review Plan for the Ozone National Ambient Air Quality Standards
Review: External Review Draft. Research Triangle Park, NC: Environmental Media
Assessment Group, National Center for Environmental Assessment and Health and
Environmental Impacts Division, Office of Air Quality Planning and Standards. (EPA
document number EPA 452/D-09-001).
U. S. EPA. 2009b. Risk and Exposure Assessment to Support the Review of the SO 2 Primary
National Ambient Air Quality Standard. EPA Office of Air Quality Planning and
Standards. (EPA document number EPA-452/R-09-007).
U.S. EPA. 2010a. Quantitative Risk and Exposure Assessment for Carbon Monoxide -Amended.
EPA Office of Air Quality Planning and Standards. (EPA document number EPA-452/R-
3-27
-------
10-009, July). .
U.S. EPA. 201 Ob. Quantitative Health Risk Assessment for Particulate Matter. Research
Triangle Park, NC: Office of Air Quality Planning and Standards. (EPA document
number EPA-452/R-10-005).
.
U.S. EPA. 201 la. Integrated Review Plan for the Os National Ambient Air Quality Standards.
Research Triangle Park, NC: National Center for Environmental Assessment, Office of
Research and Development and Office of Air Quality Planning and Standards, Office of
Air and Radiation. (EPA document number EPA 452/R-l 1-006).
.
U.S. EPA. 201 Ib. Ozone National Ambient Air Quality Standards: Scope and Methods Plan for
Health Risk and Exposure Assessment. Research Triangle Park, NC: Office of Air
Quality Planning and Standards. (EPA document number EPA-452/P-11-001). <
http://www.epa.gov/ttn/naaqs/standards/ozone/data/201 l_04_HealthREA.pdf>.
U.S. EPA. 201 Ic. Ozone National Ambient Air Quality Standards: Scope andMethods Plan for
Welfare Risk and Exposure Assessment. Research Triangle Park, NC: Office of Air
Quality Planning and Standards. (EPA document number EPA-452/P-11-002).
U.S. EPA. 2013 a. Integrated Science Assessment of Ozone and Related Photochemical Oxidants
(Final Report). Research Triangle Park, NC: EPA Office of Research and Development.
(EPA document number EPA/600/R-10/076F).
U.S. EPA. 2013b. Environmental Benefits Mapping Analysis Program (BenMAP v4.0). Posted
January, 2013. < http://www.epa.gov/air/benmap/download.html />.
Wegman, L. 2012. Updates to information presented in the Scope andMethods Plans for the Os
NAAQS Health and Welfare Risk and Exposure Assessments. Memorandum from Lydia
Wegman, Division Director, Health and Environmental Impacts Division, Office of Air
Quality Planning and Standards, Office of Air and Radiation, US EPA to Holly
Stallworth, Designated Federal Officer, Clean Air Scientific Advisory Committee, US
EPA Science Advisory Board Staff Office. May 2, 2012.
WHO. 2008. World Health Organization Harmonization Project Document No. 6. Part 1:
Guidance Document on Characterizing and Communicating Uncertainty in Exposure
Assessment, .
3-28
-------
4 AIR QUALITY CHARACTERIZATION
4.1 INTRODUCTION
Air quality information is used in Chapter 5 through Chapter 8 to assess risk and
exposure resulting from recent Os concentrations, as well as to estimate the relative change in
risk and exposure that could result from just meeting the existing Os standard of 75 ppb and the
potential alternative standard levels of 70 ppb, 65 ppb, and 60 ppb.1 The same air quality data are
used to examine fifteen2 urban study areas in the population exposure analyses discussed in
Chapter 5 and the lung function risk assessment based on application of results from clinical
studies discussed in Chapter 6: Atlanta, GA; Baltimore, MD; Boston, MA; Chicago, IL;
Cleveland, OH; Dallas, TX; Denver, CO; Detroit, MI; Houston, TX; Los Angeles, CA; New
York, NY; Philadelphia, PA; Sacramento, CA; St. Louis, MO; and Washington, DC. The
epidemiology-based risk assessment discussed in Chapter 7 examines twelve3 of the fifteen
urban study areas evaluated in the population exposure analyses. Finally, Chapter 8 includes an
assessment of the national-scale Os mortality risk burden associated with recent Os
concentrations, and characterizes the representativeness of the 15 urban study areas compared to
the rest of the U.S. This chapter describes the air quality information developed for these
analyses, providing an overview of monitoring data and air quality (section 4.2) and an overview
of air quality inputs to the risk and exposure assessments (section 4.3).
4.2 OVERVIEW OF OZONE MONITORING AND AIR QUALITY DATA
To determine whether or not the NAAQS have been met at an ambient Os monitoring
site, a statistic commonly referred to as a "design value" must be calculated based on three
consecutive years of data collected from that site. The form of the existing Os NAAQS design
value statistic is the 3-year average of the annual 4th highest daily maximum 8-hour (8-hr) Os
concentration in parts per billion (ppb), with decimal digits truncated. The existing primary and
secondary Os NAAQS are met at an ambient monitoring site when the design value is less than
or equal to 75 ppb.4 In counties or other geographic areas with multiple monitors, the area-wide
1 For a subset of urban areas and analyses, the HREA evaluates a standard level of 55 ppb, consistent with
recommendations from CAS AC to also give consideration to evaluating a level somewhat below 60 ppb.
2 In the first draft REA, we proposed to include 16 urban areas in the second draft REA. However, further analysis
of the air quality information available for Seattle, WA has prompted us to not include that city. This decision and
supporting analysis are discussed in more detail in Appendix 4E.
3 These study areas are Atlanta, GA; Baltimore, MD; Boston, MA; Cleveland, OH; Denver, CO; Detroit, MI;
Houston, TX; Los Angeles, CA; New York, NY; Philadelphia, PA; Sacramento, CA; and St. Louis, MO.
4 For more details on the data handling procedures used to calculate design values for the existing Os NAAQS, see
40 CFR Part 50, Appendix P.
4-1
-------
design value is defined as the design value at the highest individual monitoring site, and the area
is said to have met the NAAQS if all monitors in the area are meeting the NAAQS.
Air quality monitoring data from 1,468 U.S. ambient Os monitoring sites were retrieved
by EPA staff for use in the risk and exposure assessments. The initial dataset consisted of hourly
Os concentrations in ppb collected between 1/1/2006 and 12/31/2010 from these monitors. Data
for nearly 1,400 of these monitors were extracted from EPA's Air Quality System (AQS)
database5, while the remaining data came from EPA's Clean Air Status and Trends Network
CASTNET) database which consists of primarily rural monitoring sites.
These data were split into two design value periods, 2006-2008 and 2008-2010, and all
subsequent analyses based on these data were conducted independently for these two periods.
Observations flagged in AQS as having been affected by exceptional events were included the
initial dataset, but were not used in design value calculations in accordance with EPA's
exceptional events policy. Missing data intervals of 1 or 2 hours in the initial dataset were filled
in using linear interpolation. These short gaps often occur at regular intervals in the ambient data
due to an EPA requirement for monitoring agencies to perform routine quality control checks on
their Os monitors. Quality control checks are typically performed between midnight and 6:00
AM when Os concentrations are low. Missing data intervals of 3 hours or more were not
replaced. Interpolated data values were not used in design value calculations.
Figure 4-1 and Figure 4-2 show the design values for the existing Os NAAQS for all
regulatory monitoring sites in the U.S. for the 2006-2008 and 2008-2010 periods, respectively. In
general, Os design values were lower in 2008-2010 than in 2006-2008, especially in the Eastern
U.S. There were 518 Os monitors in the U.S. with design values above the existing standard in
2006-2008, compared to only 179 in 2008-2010.
5 EPA's Air Quality System (AQS) database is a national repository for many types of air quality and related
monitoring data. AQS contains monitoring data for the six criteria pollutants dating back to the 1970's, as well as
more recent additions such as PM2.5 speciation, air toxics, and meteorology data. At present, AQS receives
hourly O3 monitoring data collected from nearly 1,400 monitors operated by over 100 state, local, and tribal air
quality monitoring agencies.
4-2
-------
•; ..
f
§
.80
• *
ALASKA
•
X;-
V
I 8-Hour Oiono Detign Vilu». 2006-2008
.
• ««• &S ppe (4S SMD
O «
HAWAII
PUERTO RK'( I
*
Figure 4-1. Map of Monitored 8-hr O3 Design Values for the 2006-2008 Period.
rr\.
HAWAII
o 71-?spi>b(30oat«i
* 7«-i;Ol»»<17»Si»l)
PUERTORICp
n
I *
Figure 4-2. Map of Monitored 8-hr O3 Design Values for the 2008-2010 Period.
4-3
-------
4.3 OVERVIEW OF URBAN-SCALE AIR QUALITY INPUTS TO RISK AND
EXPOSURE ASSESSMENTS
The air quality information input into the urban-scale risk and exposure assessments
includes both recent air quality data from the years 2006-2010, as well as air quality data
adjusted to reflect just meeting the existing and potential alternative standard levels. In this
section, we summarize these air quality inputs and discuss the methodology used to adjust air
quality to meet the existing and potential alterative standards. Figure 4-3 presents a flowchart of
air quality data processing steps for the urban-scale analyses. The rest of section 4.3.1 will
provide more details on each step depicted in the flow diagram. Additional information is
provided in Appendices 4A, 4B and 4D.
> Recent ^
Monitored O3: '
lourly2Q06-201Q measurements f
at individual monitors J
Community Multiscale
Air Quality model
instrumented with the
Higher Order Decoupled Direct
Method (CMAQ-HDDM)
O3 sensitivity
coefficients
Daily O3 Metrics at
composite monitor for
recent conditions in urban
case study areas.
Metrics include:
MDAS
8-hr average
1-hr max
Adjustment of hourly ozone
values for a M potential
emissions reduction levels
+
metrics at \
tor in urban case \
ter just meeting \
rnative standards I
include:
DAS /
verage /
rmax /
Voronoi Neighbor
Averaging (VNA)
interpolation
rJ
Y
f~
1 Hourly ozone con
— ^-f justmeeting
I alternative
"
Hourly census-tract level
VNA surfaces foreach urban case
study area for recent conditions
Hourly census-tract level
VNA surfaces for each urban case
study area after just meeting existing i
And alternative standards
Epi-based RiskAssessrnentfor
Urban Case Study areas.
End points:
Deaths, hospital admissions etc
:;linical-based risk assessment
Endpoints:
Tiber of individualsexperiences
FEV1 decrements > 1O%
Exposure assessment
Endpoints:
Number of individualsexposuresto
A single hourly ozone concentration
above 60, 70, and 80 ppb
Benchmark levels
Figure 4-3. Flowchart of Air Quality Data Processing for Different Parts of the Urban-
scale Risk and Exposure Assessments.
4-4
-------
4.3.1 Urban Study Areas
4.3.1.1 Exposure modeling and controlled human exposure study-based risk
assessment
The 15 urban study areas in the exposure modeling and lung function risk assessments
covered a large spatial extent, with boundaries generally similar to those covered by the
respective Combined Statistical Areas (CSA) defined by the U.S. Census Bureau. Table 4-1
gives some basic information about the 15 urban study areas in the exposure assessment,
including the number of ambient monitoring sites, the required Os monitoring season, and the
2006-2008 and 2008-2010 design values for each study area. All 15 of the urban study areas had
8-hr Os design values above the existing standard in 2006-2008, while 13 urban areas had 8-hr
Os design values above the existing standard in 2008-2010. Chicago (74 ppb) and Detroit (75
ppb) had design values meeting the existing standard during the 2008-2010 period. The design
values in the 15 urban areas decreased by an average of 6 ppb between 2006-2008 and 2008-
2010, ranging from no change in Sacramento to a decrease of 15 ppb in Atlanta.
Table 4-1. Monitor and Area Information for the 15 Urban Study Areas in the Exposure
Modeling and Clinical Study Based Risk Assessment.
Study Area
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington DC
#of
Counties
33
7
10
16
8
11
13
9
10
5
27
15
7
17
26
#of03
Monitors
13
7
14
26
13
20
26
12
22
54
31
19
26
17
22
Population
(2010)
5,618,431
2,710,489
5,723,468
9,686,021
2,881,937
6,366,542
3,390,504
5,218,852
5,946,800
17,877,006
21,056,173
7,070,622
2,755,972
2,837,592
5,838,518
Required Os
Monitoring Season
March - October
April - October
April - September
April - October
April - October
January- December
March - September
April - September
January- December
January- December
April - October
April - October
January- December
April - October
April - October
2006-2008
DV (ppb)
95
91
83
78
82
89
86
81
91
119
90
92
102
85
87
2008-2010
DV (ppb)
80
89
77
74
77
86
77
75
85
112
84
83
102
77
81
Because Os design values are based on the annual 4th highest daily maximum 8-hr
average Os concentrations from 3-consecutive years, it is useful to look at inter-annual
variability. In general, the annual 4th highest 8-hr Os concentrations decreased in 11 of the 15
urban areas from 2006 to 2010, while remaining relatively constant in the other 4 areas (Figure
4-4). The average decrease in the annual 4th highest daily maximum concentration from 2006 to
2010 was 8 ppb. However, there was significant year-to-year variability, and some areas showed
increases in some years relative to 2006, even though the 2010 values were generally lower.
4-5
-------
» Baltimore
• Boston
-*—NewYork
—Philadelphia
?|( Washington
2006
2CC7
2008
2009
2010
•Atlanta
-Ch-cago
•Cleveland
•Detroit
•SaintLouis
2006
2007
2008
2009
2010
•^-Dallas
• Denver
—tt- Houston
>( LosAngeles
< Sacramento
2005
2007
2008
2009
2C1C
Figure 4-4. Trends in Annual 4th Highest Daily Maximum 8-hr Average Os Concentrations
in ppb for the 15 Urban Study Areas for 2006-2010. Urban areas are grouped into 3
regions: Eastern (top), Central (middle), and Western (bottom).
4-6
-------
4.3.1.2 Epidemiological-based risk assessment
Table 4-2 gives some basic information on the 12 urban study areas in the epidemiology-
based risk assessment for each set of area boundaries. The spatial extent of each urban study area
was based on the respective Core Based Statistical Area (CBSA)6. The CBSAs were generally
smaller than the study areas used in the exposure modeling and clinical study based risk
assessments, except for Baltimore and Houston, where the two study areas were identical. The
rationales for the definitions of the spatial areas used in each type of analysis are provided in the
corresponding chapters. The final two columns in Table 4-2 show the annual 4th highest daily
maximum 8-hr Os concentration in ppb for the monitors within each urban study area in 2007
and 2009.
It should be noted that the CBSA boundaries used for the urban study areas in this
assessment are different than those used in the 1st draft of the HREA, where the study areas were
derived from the Zanobetti and Schwartz (2008) study. The change to the CBSA boundaries was
intended to capture a larger portion of the urban area populations by including some surrounding
suburban counties, rather than focusing strictly on the urban population centers. Two sensitivity
analyses were conducted to determine the effect of changing the spatial extent of the urban study
areas on the epidemiology-based risk estimates. These sensitivity analyses are presented in
Chapter 7, and a summary of the two alternative sets of boundaries for the 12 urban study areas
are provided in Appendix 4A.
Since Os is not directly emitted but is formed through photochemical reactions, precursor
emissions may continue to react and form Os downwind of emissions sources, thus the highest
Os concentrations are often found downwind of the highest concentrations of precursor
emissions near the urban population center. There were some instances where the highest
monitor occurred outside of the CBSA, but within the exposure area, which was designed to
always include the monitor associated with the area-wide design value. For example, in Los
Angeles, the CBSA includes Los Angeles and Orange counties, but the highest Os concentrations
are typically measured further downwind in Riverside and San Bernardino counties. Thus, the
values reported in Table 4-2 may not match the values shown in Figure 4-4.
6 Core Based Statistical Areas (CBSAs) are used by the Office of Management and Budget (OMB) to group U.S.
counties into urbanized areas. These groupings are updated by OMB every 5 years. The CBSAs used in the
epidemiology based risk assessment are based on the OMB delineations from 2008. For more information see:
http://www.wMtehouse.gov/sites/default/files/omb/assets^lletins/blO-02.pdf
4-7
-------
Table 4-2. Monitor and Area Information for the 12 Urban Study Areas in the
Epidemiology Based Risk Assessment.
Study Area
Atlanta
Baltimore
Boston
Cleveland
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
#of
Counties
28
7
7
5
10
6
10
2
23
11
4
16
#of03
Monitors
13
7
11
10
16
8
22
21
22
15
17
17
Population
(2010)
5,268,860
2,710,489
4,552,402
2,077,240
2,543,482
4,296,250
5,946,800
12,828,837
18,897,109
5,965,343
2,149,127
2,812,896
2007 4th high
(ppb)
102
92
89
83
97
93
90
105
94
102
93
94
2009 4th high
(ppb)
77
83
75
72
79
73
91
108
81
74
96
74
4.3.2 Recent Air Quality
The sections below summarize the recent air quality data input into the epidemiological
study-based risk assessment, and the exposure and controlled human exposure study-based risk
assessment. Additional details on these inputs are provided in Appendix 4A.
4.3.2.1 Exposure modeling and controlled human exposure study based risk
assessment
As discussed in more detail in Chapter 5, the HREA uses the Air Pollutants Exposure
(APEX) model (U.S. EPA, 2012a, b) to simulate exposure and to estimate lung function
decrements based on application of results of controlled human exposure studies to populations
in the 15 urban study areas. The APEX model uses spatial fields of hourly Os concentrations at
each census tract within an urban area to simulate exposure. We use Voronoi Neighbor
Averaging (VNA) (Gold, 1997; Chen et al, 2004) to estimate hourly Os concentrations at each
census tract in all 15 urban study areas, for recent measured air quality, air quality meeting the
existing standard of 75 ppb, and air quality meeting potential alternative standards.7 The VNA
fields were estimated using ambient hourly Os concentrations from monitors in each urban area,
as well as monitors within a 50 km buffer region around the boundaries of each area. Additional
details on the procedure used to generate the VNA fields, and a technical justification for the
change from nearest neighbor fields to VNA fields are included in Appendix 4A.
In the first draft REA, these hourly spatial fields were generated for four urban areas using the concentrations from
the nearest neighboring O3 monitor.
4-8
-------
Figure 4-5 shows county-level maps of the 15 urban study areas. Counties colored pink
indicate the study area boundaries used in the Zanobetti & Schwartz (2008) and/or Smith et al.
(2009b) studies,8 where applicable. Counties colored gray indicate additional counties within the
CBS A boundaries, and counties colored peach indicate any additional counties included in the
exposure and lung function risk assessments. The X's indicate locations of the Os monitors used
in the risk and exposure assessments, including those within the 50 km buffer region used to
create the VNA fields.
4.3.2.2 Epidemiology-based risk assessment
We input Os air quality concentration data for the epidemiology-based risk analyses into
the environmental Benefits Mapping and Analysis Program (BenMAP) (U.S. EPA, 2013) for
assessment. We used BenMAP to analyze four different daily Os metrics in 12 of the 15 urban
study areas, which were the basis for concentration-response relationships derived in various
epidemiology studies:
(1) Daily maximum 1-hr concentration
(2) Daily maximum 8-hr concentration
(3) Daytime 8-hr average concentration (10:OOAMto 6:OOPM)
(4) Daily 24-hr average concentration
The air quality monitoring data used in BenMAP were daily time-series of "composite
monitor" values for each of the 12 urban areas for years 2007 and 2009, which were chosen to
represent years with high and low Os concentrations, respectively. The composite monitor values
were calculated by first averaging the hourly Os concentrations for all monitors within the area-
of-interest (resulting in a single hourly time-series for each urban area), then calculating the four
daily metrics listed above. More details on the composite monitor value calculations and a
presentation of the resulting concentrations can be found in Appendices 4A and 4D, respectively.
! The Zanobetti and Schwartz (2008) and Smith et al (2009) study area boundaries were identical for 6 of the 12
urban study areas, and had at least one county in common for all 12 urban study areas. The 'Epidemiology Study
Area' labels in Figure 4-5 refer to counties included in either of these two studies.
4-9
-------
Baltimore
Philadelphia
New York
D Epidemiology
Study Area
D Additional Counties
in CBSA
D Additional Counties
in Exposure Area
x Ozone Monitor
Figure 4-5a. Maps of the 5 Eastern U.S. Urban Study Areas Including Os Monitor
Locations.
4-10
-------
Atlanta
Detroit
Cleveland
D Epidemiology
Study Area
D Additional Counties
inCBSA
D Additional Counties
in Exposure Area
x Ozone Monitor
Figure 4-5b. Maps of the 5 Central U.S. Urban Study Areas Including Os Monitor
Locations.
4-11
-------
Dallas
Los Angeles
Houston
D Epidemiology
Study Area
D Additional Counties
in CBSA
D Additional Counties
in Exposure Area
x Ozone Monitor
Figure 4-5c. Maps of the 5 Western U.S. Urban Study Areas Including
Locations.
Monitor
4-12
-------
4.3.3 Air Quality Adjustments for Just Meeting Existing and Potential Alternative Ozone
Standards
The focus of the risk and exposure assessments is the evaluation of risks and exposures
after just meeting existing and alternative standards, and the change in risk between just meeting
existing standards and just meeting alternative standards. These evaluations require estimation of
the change in hourly Os concentrations that may occur in each urban area when "just meeting"
the existing and potential alternative Os standards.
The first draft HREA and the previous Os NAAQS review used the "quadratic rollback"
method to adjust ambient Os concentrations to simulate just meeting existing and alternative
standards (U.S. EPA, 2007; Wells et al., 2012). Although the quadratic rollback method
replicates historical patterns of air quality changes better than some alternative methods (e.g.
simply shaving peak concentrations off at the NAAQS level and the proportional rollback
technique), its implementation relies on a statistical relationship instead of on a mechanistic
characterization of physical and chemical processes in the atmosphere. Because of its construct
as a statistical fit to measured Os values, the quadratic rollback technique cannot capture spatial
and temporal heterogeneity in Os response and also cannot account for nonlinear atmospheric
chemistry that causes increases in Os during some hours and in some locations as a result of
emissions reductions under some circumstances.
Photochemical grid models are better able to simulate these phenomena and therefore the
first draft HREA proposed to replace quadratic rollback with a model-based Os adjustment
methodology and presented a test case for Atlanta and Detroit using modeling for July/August
2005 (Simon et al., 2012). The section below summarizes the methodology applied in this
assessment to adjust air quality to just meet existing and alternative standards. This new
methodology applies Higher-Order Decoupled Direct Method (HDDM) capabilities in the
Community Multi-scale Air Quality (CMAQ) model to simulate the response of Os
concentrations to reductions in US anthropogenic NOx and VOC emissions. The model
incorporates anthropogenic U.S., Canadian, Mexican and other international emissions, as well
as emissions from non-anthropogenic sources. Since sources of background Os are incorporated
explicitly in the modeling, specifying U.S. background concentrations is unnecessary.
Application of this approach also addresses the recommendation by the National Research
Council of the National Academies (NRC, 2008) to explore how emissions reductions might
affect temporal and spatial variations in Os concentrations, and to include information on how
NOx versus VOC control strategies might affect risk and exposure.
4-13
-------
4.3.3.1 Methods
The EPA has developed an HDDM-adjustment methodology to estimate hourly Os
concentrations that could occur at each monitor location if urban study areas were to meet the
existing and various alternative levels of the Os standard. An early version of this methodology
was proposed in the first draft HREA (Simon et al., 2012) and published in a peer-reviewed
journal (Simon et al., 2013). The methodology and its application to hourly Os concentrations in
the urban study areas is summarized below and described in more detail in Appendix 4D.
The HDDM-adjustment methodology uses the CMAQ photochemical model to determine
monitoring site-specific response of hourly Os concentrations to reductions in US anthropogenic
NOx and VOC emissions. The CMAQ model simulation was run for 8-months (January, April-
October) in 2007 and used a 12km grid resolution covering the continental United States. The
modeled responses are then applied to ambient data to create a 5-year time-series of hourly Os
concentrations at each monitor location which is consistent with meeting various potential levels
of the Os NAAQS for the two three-year averaging periods 2006-2008 and 2008-2010. The steps
are outlined in Figure 4-6 and summarized below:
• Step 1: Run CMAQ simulation with HDDM to determine hourly Os sensitivities to NOx
emissions and VOC emissions for the grid cells containing monitoring sites in an urban
area.
• Inputs: Model-ready emissions and meteorology data
• Outputs: Ozone concentrations and sensitivities at locations of monitoring sites
for each hour in January and April-October, 2007
• Step 2: For each monitoring site, season, and hour of the day use linear regression to
relate first order sensitivities of NOx and VOC (SNOX and 5Voc)to modeled Os and second
order sensitivities to NOx and VOC (S2NOx and S2voc) to the first order sensitivities. The
relationships derived here are strictly empirical and the applicability of these model-
derived relationships to ambient data relies on the assumption that the real-world
relationship is captured by the model.
• Inputs: Step 1 outputs
• Outputs: Functions to calculate typical sensitivities based on monitor location,
Os concentration, season, and hour of the day
• Step 3: For each measured hourly Os value between 2006 and 2010, calculate the first
and second order sensitivities based on monitoring site-, season-, and hour-specific
functions derived in Step 2.
• Inputs: Step 2 outputs and hourly ambient data for 2006-2010.
• Outputs: Hourly Os observations paired with modeled sensitivities for all hours
in 2006-2010 at all monitor locations
4-14
-------
Step 4: Adjust measured hourly Os concentrations for incrementally increasing levels of
emissions reductions using assigned sensitivities and then recalculate design values until
an emissions reduction level is reached at which all monitors in an urban area are below
the existing and potential alternative levels of the standard.
• Inputs: Step 3 outputs
• Outputs: Adjusted hourly Os values for 2006-2010 at monitor locations to show
compliance with the existing and potential alternative standard levels based on the
three year average of the 4th highest daily maximum 8-hr averageOs value. For
each standard, two sets of data are created: 2006-2008 and 2008-2010. Because
the emissions reductions used to attain standards in the two time periods might be
different, adjusted 2008 Os values are different for the two sets of data.
Step 5: Process hourly monitor data to create composite monitor time-series for
epidemiology risk assessment and to create hourly VNA spatial surfaces for exposure and
lung function risk assessment as shown in Figure 4-3.
I"'-
(
i
Recent Monitored O3
(2006-2010)
Step 3:
Use Regressions and
Observed Ozone to
Predict Sensitivities
f Hourly Ozone \
/ Observations Paired with \
\ Sensitivities for 2006-2010 J
V At All Monitor Locations /
r~
I
I
Natural
Anthropogenic
Canada a
Mexico
O3 and O3 Precursor Emissions
f Unique L
/ Relationships
Sensitivities a
Location fo
\ Season
X, Hour-of-th
S
1
Step 1 a: /
CMAQ J
HDDM Modeling ^
(Jan, Apr-Oct 2007) \
near \
between \
nd Hourly \
r Each 1 Create
and /
e-Day J
\
/Select Em issions\^ Step 4:
>(/ Reductions to N, Adjust Hourly
\^ which Sensitivities / toMeetAlte
\Will Be Appliecl/ Standard
, ' Meteo
id
rology
Initial and Boundary conditions
Other Model Inputs
/" n™
/ &
Step 2: ^
Regressions " Me
V Ea
V *j£
/Adj
J™
( Sh
\ A
Steplb:
Extract Output
E
ne Concentrations N.
Sensitivities at \
Locations of
>nitoring Sites for 1
;h Modeled Hour /
n, Apr-Oct 2007) /
usted Hourly OzoneN
es for 2006-201 Oat \
h Monitor Location to )
ow Attainmentwith j
ternate Standards ^/
Figure 4-6. Flowchart of HDDM Adjustment Methodology to Inform Risk and Exposure
Assessments.
4-15
-------
The CMAQ HDDM modeling described above is used to approximate Os response to
emission changes and is most accurate for emissions changes smaller than 50%. To
accommodate increasing uncertainty at larger emission adjustments, the HDDM modeling was
performed at three distinct emissions levels (base conditions, 50% cut in NOx and/or VOC
emissions, and 90% cut in NOx and/or VOC emissions) to allow for a better characterization of
Os response over the entire range of emissions levels. A detailed description of the methodology
and evaluation for this 3-step HDDM adjustment procedure is provided in Appendix 4D.
We chose to adjust air quality for just meeting the existing and alternative standards by
decreasing U.S. anthropogenic emissions of NOx and/or VOC throughout the U.S. For the
purpose of this analysis we used the Community Multiscale Air Quality (CMAQ) model version
4.7.1 equipped with HDDM to simulate 8 months in 2007 (January and April-October). This
time period was chosen to cover the full Os season and also includes at least one month from
each season of the year. A full description of the model inputs, model set-up, and operational
model evaluation against ambient data is available in Appendix 4B. Sensitivities derived from
the 2007 model simulation were applied to the two 3-year periods of ambient data (2006-2008
and 2008-2010) described in section 4.3.1.1.
By applying across-the-board decreases in anthropogenic emissions throughout the U.S.,
we were able to estimate how Os would respond to changes in ambient NOx and/or VOC
concentrations without simulating a specific control strategy. The model was set up to track
response in hourly Os concentrations to these across-the-board changes in US anthropogenic
NOx and VOC emissions. It should be noted that although nationwide emission decreases were
imposed, different levels of emission decreases were used for different urban study areas. In
choosing to apply across the board reductions throughout the modeling domain, we recognize
that not all emissions across the domain contribute equally to nonattainment in each urban area.
However, by decreasing emissions across the domain, we allow for the possibility of
contribution from both regional and local emissions sources to nonattainment and to the overall
distribution of Os concentrations in urban areas.
The purpose of these reduction scenarios is not to evaluate the feasibility of a particular
emission control or to account for reductions associated with other rulemakings but rather to
develop internally consistent estimates of spatial and temporal variability in Os associated with
specified alternative levels of possible standards. The modeling included sources which
contribute to background Os such as biogenic emissions, wildfire emissions, and transport of Os
and its precursors from international source regions. In addition, the HDDM tool was set-up to
specifically calculate the changes in Os that would occur from changes in US anthropogenic
emissions alone, yet to account for the effects of background sources on this response.
Consequently, it is not necessary to set a "floor" background Os concentration as was done for
4-16
-------
quadratic rollback because background sources are explicitly accounted for in the model
estimates of Os response to US anthropogenic emissions.
As described in more detail in Appendix 4D, the HDDM adjustment methodology
estimates hourly Os concentrations that would be associated with attaining a targeted level of the
standard either though reductions in US anthropogenic NOx emissions alone or through
reductions of both US anthropogenic NOx and VOC emissions in equal percentages. We
explored the use of scenarios that reduced VOC alone, but found that Os response due to
reductions in anthropogenic VOC reductions alone could not adjust the air quality to just meet a
60 ppb standard in any of the 15 urban study areas evaluated. Because the combined NOx/VOC
cuts are constrained to equal percentage cuts of both precursors, this is not an optimized
NOx/VOC control scenario but rather a sensitivity analysis to characterize the range of results
that could be obtained with alternate assumptions. In most of the urban areas, although the
NOx/VOC scenario affected Os response on some days, it did not affect Os response at the
highest design value (or controlling) monitor in such a way to reduce the total required emissions
cuts. However, for the Chicago and Denver study areas, the NOx/VOC scenarios allowed for
lower percentage emissions cuts (applied to both NOx and VOC) to reach targeted standard
levels than the NOx only scenario. Table 4-3 shows the percent emissions reductions applied in
both the NOx-only and the NOx/VOC scenarios for each urban study area. Because of this, the
core analyses presented in the remainder of this chapter and in Chapters 5, 6, and 7 were based
on the NOx only assumption for all study areas except for Chicago and Denver which used the
NOx/VOC equal percentage reduction assumption. Sensitivity analyses were performed to
compare the NOx only and the NOx/VOC cases in 9 study areas: Denver, Detroit, Houston, Los
Angeles, New York, Philadelphia, and Sacramento. The effects of these sensitivity analyses on
air quality and on the epidemiology-based risk assessment are discussed in more detail in
Appendix 4D and Chapter 7, respectively.
4-17
-------
Table 4-3. NOx-only and NOX/VOC Emission Reductions Applied to 15 Urban Study
Areas. Numbers in blue bold text represent the base scenario while numbers in plain
represent the scenario used in the sensitivity analyses for each urban study area.
font
Study Area
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
(95% LB)
New York
(95% LB)
Philadelphia
Sacramento
St. Louis
Washington
D.C.
Years
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
2006-2008
2008-2010
NOx only
75
41%
N/A
54%
24%
70
54%
42%
67%
53%
65
65%
56%
76%
70%
60
75%
67%
89%
89%
NOx/VOC
75
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
51%
50%
60%
N/A
65%
40%
95%
93%
60%
41%
57%
37%
65%
65%
N/A
N/A
N/A
N/A
70
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
60%
59%
73%
53%
73%
52%
96%
95%
71%
55%
65%
52%
73%
74%
N/A
N/A
N/A
N/A
65
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
69%
68%
85%
69%
81%
65%
98%
97%
89%
86%
71%
62%
80%
81%
N/A
N/A
N/A
N/A
60
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
77%
76%
90%
85%
87%
85%
99%
98%
N/A
N/A
79%
72%
88%
88%
N/A
N/A
N/A
N/A
4-18
-------
For New York and Los Angeles it should also be noted that a somewhat different
approach was used for the HDDM-adjustment application. The HDDM adjustment methodology
produces estimates of hourly Os concentrations with standard error bounds for every potential
emission reduction scenario. Uncertainties in the application of the methodology to very large
emissions adjustments along with the highly nonlinear response of Os to NOx reductions at a
limited number of sites and hours resulted in the inability of this methodology to estimate Os
distributions in these two study areas that could just meet lower alternative standard levels (65
ppb for New York, 60 ppb for Los Angeles). This does not indicate that these two areas would
not be able to meet these lower standard levels in reality, but simply reveals the limitations of
this adjustment methodology. More specifically, at a few highly urbanized monitors during some
seasons, the Os increases during highly titrated rush-hour and nighttime hours appear to be
overestimated, leading to predictions that, when NOx was reduced, Os would peak during rush-
hour periods on the very highest Cb days. Even at these monitors, the majority of days in the Os
season did not have this issue. To address the atypical behavior on the very highest Os days in
the adjustment scenarios in New York and Los Angeles, we took the 95th percentile confidence
interval for each hourly Os prediction and used the lower bound value to determine the NOx
reductions required to meet the existing and potential alternative standards and to calculate the
hourly Os that would occur under that reduced emission scenario. During most hours and at most
monitors, the 95th percentile range was small so the use of the lower-bound of the 95th percentile
range made little difference in predicted Os concentrations. However, using the lower bound of
the 95th percent confidence interval allowed us to dampen the effect of over-predicted Os
increases during rush-hour times. Since the Os concentrations for each standard level were
created using consistent methodology, these Os datasets can be used to compare between
standards in these study areas. More details are provided in Appendix 4D. Estimates of risk for
these two study areas for these alternative standards will be significantly more uncertain,
reflecting the use of the lower bound Os predictions.
4.3.3.2 Resulting air quality
The HDDM adjustment technique tended to have several effects on the distribution of air
quality values. First, adjusted hourly Os concentrations at night and during the morning rush-
hour tended to be higher than the recent observed concentrations (additional details are provided
in Appendix 4D). The CMAQ model predicts that, in general, these times have NOx titration
conditions meaning that a reduction in NOx causes an increase in Os concentrations. The NOx
titration effect was most pronounced in urban core areas which have higher volume of mobile
source NOx emissions from vehicles than do the surrounding areas. Response of daytime
concentrations was more varied. In general, Os tended to increase on low days and decrease on
4-19
-------
high days. However, specific monitors that were either always heavily VOC limited or always
heavily NOX limited showed consistent increases and decreases respectively regardless of
whether Os concentrations were high or low on a particular day. It should be noted that locations
which were heavily VOC limited tended to have much lower observed Os concentrations than
downwind areas. The tendency of the model to predict Os increases on lower concentration days
and decreases on higher concentration days also leads to more compressed Os distributions in the
HDDM adjustment cases. The variability in predicted daily Os concentrations decreased when
meeting lower standard levels. The following paragraphs summarize a comparison of Os
distributions from application of the quadratic rollback and HDDM adjustment approach for a
case where the existing standard is estimated to be met, characterize the distribution of
composite monitor Os values at different standard levels, and provide a discussion of the spatial
distribution of Os changes in several study areas. More details and figures for other study areas
are provided in Appendix 4D.
Figure 4-7 and Figure 4-8 show a comparison of April-October composite monitor Os
distributions for recent conditions (2006-2008) and for meeting the existing standard using the
quadratic rollback technique versus the HDDM adjustment methodology. The composite monitor
values in these plots are based on the monitors included in the composite monitor from the
Zanobetti and Schwartz (2008) study which was used in the 1st draft HREA and do not include
all monitors in the CBSA as used in the main Chapter 7 analysis. In general, the Os distribution
in the HDDM adjustment case is shifted upward compared to the quadratic rollback case. The
upward shift is more pronounced in the lower parts of the Os distribution. In all study areas
displayed in Figure 4-7, the 25th percentile, median, and mean of the daily maximum 8-hr
average Os concentrations are higher in the HDDM adjustment case than the quadratic rollback.
In some study areas (Sacramento and St. Louis) the 75th percentile values appear approximately
equivalent in the two cases while in other study areas the 75th percentile values are slightly
higher in the HDDM adjustment case. In Houston, the very highest portion of the Os distribution
is lower in the HDDM adjustment case than in the quadratic rollback case but in many study
areas the upper parts of the distributions for these two cases are roughly equivalent. Similar
results are seen in the 2008-2010 time period; however there are more cases during this time
period where HDDM adjustment and quadratic rollback have similar values in the upper half of
the Os distribution. A comparison of Figure 4-7 and Figure 4-8 shows that there is some
seasonality to this effect. The two techniques appear to give very similar daily maximum 8-hr
averageOs composite monitor distributions during the summer months (June-August) and most
of the situations with higher Os levels with the HDDM adjustment come from cooler, lower Os
time periods (April, May, September, and October). Although here we discuss composite
monitor distributions based on April-October, the risk analyses in Chapter 7 are based on the
4-20
-------
required Os monitoring season, which is longer than April - October for some study areas. We
expect that the Os increases shown for spring and fall months here are also representative of the
type of response in other "cool season" months. The exceptions to this occur in Denver, Houston,
New York and Los Angeles which have higher composite monitor Os values from the HDDM
adjustment compared to quadratic rollback even in the summer time period.
Figure 4-9 and Figure 4-10 show "box-and-whisker" plots of the April-October
composite monitor daily maximum 8-hr average Os concentration distributions for the 12 urban
study areas evaluated in the epidemiology-based risk assessment; for recent air quality, and air
quality adjusted to meet the existing and potential alternative standards. Figure 4-9 shows values
from 2007, while Figure 4-10 shows values from 2009. Note that since these boxplots represent
composite monitor values (i.e. an average of values from multiple individual monitors) the
maximum composite monitor concentration may still be well below the design value which is
measured from a single monitor in some areas. Appendix 4D contains additional plots comparing
the changes in the distribution of composite monitor values in each urban area due to the air
quality adjustments across varying spatial extents, season lengths, and years. In general, the
range of the composite monitor distributions decreased (i.e. the minimum value increased, while
the maximum value decreased) in all 12 urban study areas as the air quality data were adjusted to
meet lower standard levels. However, the changes within the inter-quartile range of these
distributions (represented by the "boxes") varied in response to the model-based air quality
adjustments across the 12 urban areas. Three different types of responses are highlighted in the
boxplots for Atlanta, New York, and Houston.
The Atlanta boxplots provide an example of an urban area in which all but the lowest
composite monitor values decreased as the air quality data was adjusted to simulate compliance
with progressively lower levels of the standard. The upper tail of the distribution (represented by
the top whisker in each boxplot) decreased more quickly than the remainder of the distribution,
resulting in less total variability in the composite monitor values with each progressively lower
standard level. This type of response was also seen Sacramento and St. Louis, and to a lesser
extent in Baltimore, Denver, and Philadelphia.
In New York, the boxplots showed an initial increase in the 25th percentile and median
composite monitor values when the observed Os concentrations were adjusted to meet the
existing standard. However, the median composite monitor value decreased relative to the
existing standard as Os concentrations were adjusted to meet the 70 ppb standard, and both the
median and 25th percentile values decreased when air quality were further adjusted to meet the
65 ppb standard. When the air quality were adjusted to meet 65 ppb, the median and mean
composite monitor values were lower than under observed conditions. This type of response was
also observed in Cleveland, Detroit, and Los Angeles.
4-21
-------
Atlanta: Z & S, April-October, 2006-2008
Baltimore: Z * S, April-October, 2006-2008
Boston: Z & S, April-October, 2006-2008
Cleveland: Z & S, April-October, 2006-2008
Denver: Z & S, April-October, 2006-200B
Detroit: Z & S, April-October, 2006-200B
Houston: Z a S, April-October, Z006-Z008
LosAngeles: Z a S, April-October, 2006-2008
NewYork: Z S S, April-October, 2006-2008
75
Philadelphia: Z * S, April-October, 2006-Z008 Sacramento: Z & S, April-October, 2006-Z008
Saint-Louis: Z & S, April-October, 2006-2008
Figure 4-7. Distributions of Composite Monitor Daily Maximum 8-hr Average Os
Concentrations from Ambient Measurements (black), Quadratic Rollback (blue), and the
HDDM Adjustment Methodology (red) for Meeting the Existing Standard. Values are
based on the Zanobetti & Schwartz study areas for April-October of 2006-2008. Boxes
represent the median and quartiles, X's represent mean values, whiskers extend up to 1.5x
the inter-quartile range from the boxes, and circles represent outliers.
4-22
-------
Atlanta: Z & S, June-August. 2006-2008
Baltimore: Z & S, June-August, 2006-2008
Boston: Z & S, June-August, 2006-2008
Cleveland: Z & S, June-August, Z006-Z008
Denver: Z & S, June-August, 2006-2008
Detroit: Z & S, June-August, 2006-2008
Houston: Z & S, June-August, 2006-200B LosAngeles: Z & S, June-August, 2006-2008 NewYork: Z & S, June-August, 2006-2008
75
I
Tf
base 75
Philadelphia: Z & S, June-August, 2006-2008
Sacramento: Z & S, June-August, 2006-2008
SaintLouis: Z & S, June-August, 2006-2008
g
base
75
Figure 4-8. Distributions of Composite Monitor Daily Maximum 8-hr Average Os
Concentrations from Ambient Measurements (black), Quadratic Rollback (blue), and the
HDDM Adjustment Methodology (red) for Meeting the Existing Standard. Values are
based on the Zanobetti & Schwartz study areas for June-August of 2006-2008. Boxes
represent the median and quartiles, X's represent mean values, whiskers extend up to 1.5x
the inter-quartile range from the boxes, and circles represent outliers.
4-23
-------
Atlanta
Denver
HewYorh
I
base 75 70 65 60 base 75 70 65 60 base 75 70 65 60
Baltimore o Detroit o Philadelphia
base 75 70 65
base 75 70 65
Houston
base 75 70 65
Sacramento
base 75 70 65
Cleveland
base 75 70 65
LosAiujeles
base 75 70 65
SaintLouis
±
base 75 70 65
base 75 70 65
base 75 70 65
Figure 4-9. Distributions of Composite Monitor Daily Maximum 8-hr Average Values for
the 12 Urban Study Areas in the Epidemiology-based Risk Assessment. Plots depict values
based on ambient measurements (base), and values obtained with the HDDM adjustment
methodology when just meeting the 75, 70, 65 and 60 ppb standards. Values shown are
based on CBSAs for April-October of 2007. Note that the HDDM adjustment technique
was not able to adjust air quality to just meet a 60 ppb standard in New York, so no
boxplot is shown for that case. Boxes represent the median and quartiles, X's represent
mean values, whiskers extend up to 1.5x the inter-quartile range from the boxes, and
circles represent outliers.
4-24
-------
base 75 70 65
Baltimore
base 75 70 65
Boston
base 75 70 65 60
Cleveland
base 75 70 65
Detroit
base 75 70 65
Philadelphia
base 75 70 65
Houston
base 75 70 65
Sacramento
base 75 70 65 60 base 75 70 65 60
LosAnyeles = SaintLouis
Figure 4-10. Distributions of Composite Monitor Daily Maximum 8-hr Average Values for
the 12 Urban Study Areas in the Epidemiology-based Risk Assessment. Plots depict values
based on ambient measurements (base), and values obtained with the HDDM adjustment
methodology when just meeting the 75, 70, 65 and 60 ppb standards. Values shown are
based on CBSAs for April-October of 2009. Note that Detroit air quality was meeting 75
ppb in 2008-2010, and the HDDM adjustment technique was not able to adjust air quality
to just meet a 60 ppb standard in New York, so no boxplots are shown for those cases.
Boxes represent the median and quartiles, X's represent mean values, whiskers extend up
to 1.5x the inter-quartile range from the boxes, and circles represent outliers.
4-25
-------
In Houston, the median composite monitor value also increased between observed air
quality and air quality adjusted to meet the existing standards. However, the pattern in Houston
differed from New York and other study areas as air quality was further adjusted to reflect
meeting the potential alternative standards. The median value remained relatively constant
relative to the existing standard, while the 25th percentile values continued to increase. Thus, in
Houston, the air quality adjustments always resulted in a median composite monitor value higher
than what was seen in the observed data. The composite monitor distributions in Boston also
exhibited this type of behavior.
The exposure modeling and the clinical-based risk assessments used spatially varying
surfaces of hourly Os concentrations estimated at the centroid of each census tract within the 15
urban study areas. The maps in Figures 4-11, 4-12, and 4-13 depict the spatial distributions of the
2006-2008 average 4th highest (top) and May - September mean (bottom) daily maximum 8-hr
average (MDA8) Os concentrations for 3 of the 15 urban study areas; for observed air quality
(left), air quality adjusted to meet the existing standard (center), and air quality adjusted to meet
the 65 ppb alternative standard (right). Appendix 4A contains additional maps of the observed 4th
highest MDA8 and May - September mean MDA8 concentrations in all 15 urban study areas for
2006-2008 and 2008-2010. Appendix 4D contains maps and related figures showing the changes
in air quality that resulted from the HDDM adjustments for just meeting the existing standard,
and just meeting the potential alternative standard of 65 ppb.
These maps portray the general pattern seen in all 15 urban study areas for the 4th highest
concentrations, which decreased when observed air quality were adjusted to meet the existing
standard, and continued to decrease as the air quality were further adjusted to meet the various
alternative standards. The May-September average values also generally decreased in suburban
and rural areas surrounding the urban population center in all 15 areas. However, three different
types of general behavior which were seen in the seasonal average values near the urban
population centers, exemplified in Figures 4-11 (Atlanta), 4-12 (New York), and 4-13 (Houston).
In Atlanta, the observed May - September average were nearly constant across the entire
study area. The observed values decreased nearly uniformly across the entire study area when
observed air quality was adjusted to meet the existing standard, and continued to do so when air
quality was further adjusted to meet the alternative standard of 65 ppb. The magnitudes of these
decreases were slightly larger in suburban and rural areas than near the urban population center.
This type of behavior was also seen in Sacramento and Washington, D.C.
In New York, the observed May - September average values were lower near the urban
population center than in the surrounding suburban areas. When the observed air quality was
adjusted to meet the existing standard, the seasonal average values increased near the urban
population center and decreased in the suburban areas, so that the spatial pattern was reversed.
4-26
-------
When air quality was further adjusted to meet the 65 ppb alternative standard, large area-wide
decreases in the seasonal average values were seen relative to the existing standard. While New
York represents one of the most extreme examples, similar behavior was observed in 7 other
urban areas: Baltimore, Cleveland, Dallas, Detroit, Los Angeles, Philadelphia, and St. Louis.
The distribution of Os concentrations in Houston started out in a similar fashion as New
York. The observed May - September average concentrations were lower near the urban
population center than in the surrounding areas, and a similar pattern of increasing and
decreasing seasonal average values occurred when observed air quality was adjusted to meet the
existing standard. However, unlike New York, the seasonal average values near the Houston city
center remained nearly constant relative to the existing standard when air quality were further
adjusted to meet the 65 ppb standard. Boston, Chicago, and Denver exhibited this same type of
behavior.
Atlanta, 2006 - 2008
4th Highest MDAS- Base
4th Highest MDA8-75 ppb
4th Highest MDA8 - 65 ppb/
May - Sep mean MDA8 - Base"
May - Sep mean MDA8 - 75 ppb
May - Sep mean MDA8 - 65 ppb
Figure 4-11. Maps Showing the 4th highest (top) and May-September Average (bottom)
Daily Maximum 8-hr Average Os Concentrations in Atlanta based on 2006-2008 Ambient
Measurements (left), HDDM Adjustment to Meet the Existing Standard (center), and
HDDM Adjustment to Meet AN Alternative Standard of 65 ppb (right). Squares represent
measured values at monitor locations; circles represent VNA estimates at census tract
centroids.
4-27
-------
New York, 2006 - 2008
4th Highest M DAS-Base
4th Highest MDA8-75 ppb
4th Highest MDA8 - 65 ppb
May - Sep mean MDA8 - Base"
May - Sep mean MDA8 - 75 ppb
May - Sep mean MDA8 - 65 ppb
Figure 4-12. Maps Showing the 4th highest (top) and May-September Average (bottom)
Daily Maximum 8-hr Average Os Concentrations in New York based on 2006-2008
Ambient Measurements (left), HDDM Adjustment to Meet the Existing Standard (center),
and HDDM Adjustment to Meet the Alternative Standard of 65 ppb (right). Squares
represent measured values at monitor locations; circles represent VNA estimates at census
tract centroids.
4-28
-------
Houston, 2006 - 2008
4th Highest M DAS-Base
4th Highest MDA8-75 ppb
4th Highest MDA8 - 65 ppb
May - Sep mean MDA8 - Base
May - Sep mean MDA8 - 75 ppb
May - Sep mean MDA8 - 65 ppb
Figure 4-13. Maps of 4th Highest (top) and May-September Average (bottom) Daily
Maximum 8-hr Average Os Concentrations in Houston for 2006-2008 Ambient
Measurements (left), HDDM Adjustment to Meet the Existing Standard (center), and
HDDM Adjustment to Meet the Alternative Standard of 65 ppb (right). Squares represent
measured values at monitor locations; circles represent VNA estimates at census tract
centroids.
Figures displaying the air quality results from the NOx-only compared to the NOx/VOC
adjustment scenarios are provided in Appendix 4D. Comparison of the composite monitor and
spatial plot maps leads to several general conclusions for this sensitivity analysis. First, the
NOx/VOC reduction scenarios tended to mitigate increases that occurred in the NOx-only
scenario at the lower end of the Os concentration distribution. Second the effect on the NOx/VOC
scenario versus the NOx-only scenario was less dramatic for mid-range Os concentrations and
varied from study area to study area. The NOx/VOC scenarios lead to lower mid-range Os
concentrations than the NOx-only scenarios except in the case of New York. Third, the high end
Os concentrations at various standard levels were similar in the NOx-only and the NOx/VOC
scenarios. Finally, the VOC reductions tended to have more impact in urban core areas and
4-29
-------
relatively little impact in outlying areas. The effects of these air quality sensitivity analyses on
risk are evaluated and discussed in chapter 7.
4.4 OVERVIEW OF NATIONAL-SCALE AIR QUALITY INPUTS
The national-scale analysis, presented in Chapter 8, is focused only on evaluating the
total national burden of mortality risk associated with recent Os conditions. As such it uses a
different approach to characterize air quality conditions throughout the U.S. The national-scale
analysis employs a data fusion approach that takes advantage of the accuracy of monitor
observations and the comprehensive spatial information of the CMAQ modeling system to create
national-scale "fused" spatial surfaces of seasonal average Os concentrations. Measured Os
concentrations from 2006-2008 were fused with modeled concentrations from a 2007 CMAQ
model simulation, run for a 12 km domain covering the contiguous U.S. The spatial surfaces are
created using EPA's Downscaler software (Berrocal et al., 2012). More details on the ambient
measurements, the 2007 CMAQ model simulation, and the Downscaler fusion technique can be
found in Appendix 4C.9
Three national "fused" spatial surfaces were created for:
(1) the May-September mean of the daily maximum 8-hr average Os concentrations
(consistent with the metric used by Smith et al. 2009);
(2) the June-August mean of the daily 10am-6pm average Os concentrations (consistent
with the metric used by Zanobetti and Schwartz 2008); and
(3) the April-September mean of the daily maximum 1-hour Os concentrations
(consistent with the metric used by Jerrett et al 2009).
Figure 4-14 to Figure 4-16 show the geographic distributions of these spatial surfaces.
The spatial distributions of these three surfaces are very similar, with the highest levels occurring
in Southern California for all three surfaces.
9 In the first draft of the REA, the spatial surfaces were created using the enhanced Voronoi Neighbor Averaging
(eVNA) technique (Timin et al, 2010), using the EPA's Model Attainment Test Software (MATS; Abt Associates,
20 lOb). The technical justification for changing from eVNA to Downscaler can be found in Appendix 4C.
4-30
-------
70
80
Figure 4-14. May-September Mean of the Daily Maximum 8-hr Average Os
Concentrations in ppb, based on a Downscaler Fusion of 2006-2008 Average Monitored
Values with a 12km 2007 CMAQ Model Simulation.
4-31
-------
70
80
Figure 4-15. June-August Average 8-hr Daily 10am-6pm Mean Os Concentrations in ppb,
based on a Downscaler Fusion of 2006-2008 Average Monitored Values with a 12km 2007
CMAQ Model Simulation.
4-32
-------
70
80
90
Figure 4-16. April-September Mean of the Daily Maximum 1-hr Os Concentrations in ppb,
based on a Downscaler Fusion of 2006-2008 Average Monitored Values with a 12km 2007
CMAQ Model Simulation.
Figure 4-17 shows the frequency and cumulative distributions of these three seasonal
average Os surfaces based on all grid cells in the 12 km CMAQ modeling domain. The
minimum, median, mean, 95th percentile, and maximum values for all three surfaces are shown
in Table 4-4, and correlation coefficients between the three metrics are given in Table 4-5.
The May-September mean of the daily maximum 8-hr average Os concentrations were
most frequently in the 30-60 ppb range, while the June-August average daily 10am-6pm mean
concentrations were more evenly distributed across a range of 20-60 ppb. The April-September
mean of the daily maximum 1-hr Os concentrations were about 5 ppb higher on average than the
May-September mean daily maximum 8-hr average concentrations, and about 8 ppb higher on
average than the June-August average daily 10am-6pm mean concentrations. The correlation
coefficients between these three metrics were all very high (R > 0.97).
4-33
-------
3
IT
0.4
0.3
0.2
0.1
0
100%
0%
I May-September average 8-hr daily maximum
I June-August average 8-hr daily mean IQam-Gpm
I April-September average 1-hr daily maximum
T I I I I I I I I I I I I I I
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
0
20
40
Concentration (ppb)
60
Figure 4-17. Frequency and Cumulative Distributions of the Three Fused Seasonal
Average Os Surfaces Based on all CMAQ 12 km Grid Cells.
4-34
-------
Table 4-4. Summary Statistics based on the Three Fused Seasonal Average Os Surfaces
Based on all CMAQ 12 km Grid Cells.
Statistic
Minimum
Median
Mean
95th Percentile
Maximum
May-September mean
of the daily maximum
8-hr average
concentration (ppb)
21.8
43.6
43.2
54.3
76.1
June-August mean of the
daily 10am-6pm average
concentration (ppb)
14.9
41.7
40.9
54.8
80.1
April-September mean
of the daily maximum 1-
hr concentration (ppb)
26.2
48.8
48.2
59.0
84.2
Table 4-5. Correlation Coefficients between the Three Fused Seasonal Average
Surfaces Based on all CMAQ 12 km Grid Cells.
Seasonal metrics compared
May-September mean of the daily maximum 8-hr average
vs. June-August mean of the daily 10am-6pm average
May-September mean of the daily maximum 8-hr average
vs. April-September mean of the daily maximum 1-hr
June-August mean of the daily 10am-6pm average vs.
April-September mean of the daily maximum 1-hr
Correlation coefficient
0.974
0.995
0.972
These seasonal average metrics are not equivalent to the form of the existing standard,
which is based on the 4th highest value rather than on the seasonal mean. Thus, the values shown
in the three fused surfaces should not be directly compared to the existing standard. Figure 4-18
shows comparisons between these three metrics and the 2006-2008 Os design values based on
CMAQ 12 km grid cells containing Os monitors, and Table 4-6 presents correlation coefficients
and summary statistics based on the ratios between the design values and these three metrics.
The design values were, on average, approximately 50% higher than the seasonal average values,
with substantial spatial heterogeneity, and some variation across the seasonal average metrics.
The April-September mean of the daily maximum 1-hr Os concentration was the most strongly
correlated with the design values (R = 0.75), followed by the May-September mean of the daily
maximum 8-hr average (R = 0.71), and then the June-August mean of the daily 10am-6pm
average Os concentration (R = 0.69).
4-35
-------
80
70
-Q
Q.
a.
t/l
.2
"•M
(tl
60
§ 50
IU
DO
re
40
30
20
20
May-Sept
average 8hr
daily max
Jun-Aug
average 8hr
daily mean
(10a-6p)
Apr-Sept
average Ihr
daily max
40 60 80 100
2006-2008 Design value (ppb)
120
Figure 4-18. 2006-2008 O3 Design Values Versus 2006-2008 Fused Seasonal Average O3
Levels for the CMAQ 12km Grid Cells Containing Os Monitors.
4-36
-------
Table 4-6. Correlation Coefficients and Ratios of the 2006-2008 O3 Design Values to the
2006-2008 Fused Seasonal Average O3 Levels for the CMAQ 12km Grid Cells Containing
Os Monitors.
Statistic
Correlation
Ratios
Minimum
2.5th Percentile
Median
Mean
97.5 Percentile
Maximum
May-September mean of
the daily maximum 8-hr
average
0.71
1.1
1.3
1.5
1.6
2.0
2.4
June-August mean of
the daily 10am-6pm
average
0.69
1.1
1.3
1.5
1.6
2.2
3.0
April-September mean
of the daily maximum 1-
hr
0.75
1.0
1.2
1.4
1.4
1.6
1.9
4.5 UNCERTAINTIES IN MODELING OF RESPONSES TO EMISSION
REDUCTIONS TO JUST MEET EXISTING AND POTENTIAL ALTERNATIVE
STANDARDS
We recognize that there are sources of uncertainty in air quality measurements and the air
quality estimates for each air quality scenario. These sources of uncertainty are described below
and in Table 4-7 which discusses qualitatively the magnitude of uncertainty and potential for
directional bias.
There is inherent uncertainty in all deterministic air quality models, such as CMAQ, the
photochemical grid model which was used to develop the model-based Os adjustment
methodology. Evaluations of air quality models against observed pollutant concentrations build
confidence that the model performs with reasonable accuracy despite both structural and
parametric uncertainties. A comprehensive model performance evaluation provided in Appendix
4B shows generally acceptable model performance which is equivalent to or better than typical
state-of-the science regional modeling simulations as summarized in Simon et al. (2012). The
use of the Higher Order Decoupled Direct Method (HDDM) within CMAQ to estimate Os
response to emissions perturbations adds uncertainty to that inherent in the model itself. HDDM
allows for the approximation of Os concentrations under alternate emission scenarios without re-
running the model simulation with different inputs. This approximation becomes less accurate
for larger emissions perturbations. To accommodate increasing uncertainty at larger emissions
perturbations, the HDDM modeling was performed at three distinct emissions levels to allow for
a better characterization of Os response over the entire range of emissions levels. The accuracy
of the HDDM estimates can be quantified at distinct emissions levels by re-running the model
with modified emissions inputs and comparing the results. This method was applied to quantify
4-37
-------
the accuracy of 3-step HDDM Os estimates for 50% and 90% NOx cut conditions for each urban
study areas (as shown in Appendix 4D). At 50% NOx cut conditions, HDDM using information
from these multiple simulations predicted hourly Os concentrations with a mean bias and a mean
error less than +/- 1 ppb in all urban study areas compared to brute force model simulations. At
90% NOx cut conditions, HDDM using information from these multiple simulations predicted
hourly Os concentrations with a mean bias less than +/- 3ppb and a mean error less than +/- 4
ppb in all urban study areas. These small bias and error estimates show that uncertainty due to
the HDDM approximation method is small up to 90% emissions cuts.
In order to apply modeled Os response to ambient measurements, regressions were
developed which relate Os response to emissions perturbations with ambient Os concentrations
for every season, hour-of-the-day, and monitor location. These regressions are purely empirical
so applying Os responses to ambient data based on this modeled relationship adds uncertainty.
Preliminary work showed that the relationships developed with these regressions were generally
statistically significant for most season, hour-of-the-day, and monitor location combinations for
2005 modeling in Detroit and Charlotte (Simon et al., 2012). Statistical significance was not
evaluated for each regression in this analysis since there were over 460,000 regressions created
(322 monitors x 5 sensitivity coefficients x 3 emissions levels x 4 seasons x 24 hours = 463,680
regressions). Statistical inference can quantify the goodness of fit for the modeled relationships.
The regression model provided both a central tendency and a standard error value representing
the uncertainty in the central tendency at any at any given Os concentration. Appendix 4D
provides equations used to calculate the standard error values and propagate them through the
calculations of hourly and daily maximum 8-hr average Os values for each adjustment scenario.
Since emissions reductions increased for lower standard levels the standard errors were larger for
adjustments to lower standards. Mean (95th percentile) standard errors for the 75 ppb adjustment
case ranged from 0.13 (0.26) to 1.18 (2.87) ppb for hourly Os concentrations in the 15 urban
study areas. Mean (95th percentile) standard errors for the 65 ppb adjustment case ranged from
0.54 (1.07) to 1.39 (2.98) ppb for hourly Os concentrations. The largest standard errors occurred
in Los Angeles and New York due to the large emissions reductions applied in these cases.
Figures 4-19 and 4-20 show the range of standard errors in daily maximum 8-hr average Os
values for each of the 15 urban study areas based on the uncertainty predicting the central
tendency from the regressions. These boxplots show that standard error from this source is
generally small (less than 1 ppb) in all urban study areas. There are infrequent occurances of
larger standard errors shown by the X's denoting maximum standard error on any day in Figure
4-19. The maximum standard error on any day is still in the range of 1-3 ppb in most urban study
ares but is higher in Denver, New York and Los Angeles for the 60 ppb scenarios. Maps of mean
standard error at each monitor location are provided in Appendix 4D.
4-38
-------
The base analysis in all urban study areas except New York and Los Angeles used the
central tendency which will inherently dampen some of the variability in Os response. This may
lead to an underestimation of the variability of Os concentration decreases and increases
associated with emission reductions but is not expected to lead to a bias in estimated benefits.
However, given the small standard error values even in the urban study areas with the greatest
uncertainty (i.e. less than 1.5 ppb mean standard error), this source of uncertainty is not expected
to substantially impact exposure and risk results.
3 *
§
co
O
1
o —
D 75 ppb stnd, 2006-2008
D 75 ppb stnd, 2008-2010
n 60 ppb stnd, 2006-2008
60 ppb stnd, 2008-2010
K
•
X K
w XK, K II JL-i, T ; *
TT ^v"'' rt T-rii TTOEa T J 1, TvSn f ml, ' T LLJ olW ''il TT^ T ]
LL it • sa ^ 4^ o1 * ^ DD Hi At u *>
J^:; xx.! BjfK:-- fe*:< xjc- *£ ^ "-M KX :*x tfx >«••-• "• i^--
! I I I i i I I I i I
AIL BAL BOS ChB CLE DAL DC DEM DET HOU LA1 NY" PHI SAC STL
foes J ppb ana B .7 rob tar Hie 75 o* and fiD Kit daHads trgjertway
emus n flhcn Urine T5 and «E ffd standart *msmenU, AKE NT MI notntUHted tomwIl-icBO pfftUnd
Figure 4-19. Propagated Standard Error in Daily Maximum 8-hr Average Os
Concentrations Due to Uncertainty in Linear Regression Central Tendency for Each
Urban Study Area. Boxes show the interquartile range while whiskers extent to 1.5x the
interquartile range, X's show mina and max values.
4-39
-------
ft
-0
m
?
*
D 75 ppb stnd, 2006-2008
D 75 ppb stnd, 2008-2010
n 60 ppb stnd, 2006-2008
60 ppb stnd, 2008-2010
I I I I I I I I I ! I I 1 I I
AIL BAL BOS CH CLE DAL DC DEN DET HOU LA1 NY" PHI SAC STL
•Man mum 3E nlK *>r LA
9.3 ppb and i.7 ppb tor the 75 ppt and S3 put danlanft mpecflvety
^
Figure 4-20. Propagated Standard Error in Daily Maximum 8-hr Average Os
Concentrations Due to Uncertainty in Linear Regression Central Tendency for Each
Urban Study Area. Boxes show the interquartile range while whiskers extent to 1.5x the
interquartile range. Note that this is the same as Figure 4-19 above except with Y-axis
adjusted.
Relationships between Os response and hourly Os concentration are empirical and were
developed based on 8 months of modeling: January and April-October 2007. These relationships
were applied to ambient data from 2006-2010. This relies on the assumption that relationships
between Os response and Cb concentration, hour of day, location, and season in the real world
are well replicated in the model. Since it is not possible to directly measure Cb response in the
real-world we rely on the fact that the statistical fit was generally good for the modeled
relationships and the model performed reasonably well at predicting hourly and daily maximum
8-hr average Os concentrations in 2007. In addition to the general uncertainties introduced by
applying empirical modeled relationships to ambient Os data, uncertainty is also introduced
because these relationships are applied outside of the modeled time period. Some locations
monitor for months not included in this modeling (i.e., February, March, November, and
December) while others do not. Seasonal relationships were developed between Os response to
emissions reductions and Os concentration. Summer was the only season for which modeling
data was created for all months (June, July, August). The winter relationships were developed
4-40
-------
based on January modeling, the spring relationships were developed based on April/May
modeling, and the autumn relationships were developed based on September/October modeling.
The reduction in data points (31 or 61 instead of-90) increases uncertainty in the statistical fit
for these seasons. In addition, the modeling generally showed more Os disbenefits to NOX
decreases in cooler months. So applying April/May relationships to March and
September/October relationships to November could potentially underestimate Os increases that
would happen in those two months in the five urban study areas which measure Os during March
and/or November: Dallas, Denver, Houston, Los Angeles, and Sacramento. The eight months
that were modeled capture a variety of meteorological conditions. Applying these 2007
sensitivities to other years with potentially different meteorology and emissions adds uncertainty
to the relationship between Os response and Os concentrations. Finally, if emissions change
drastically between the modeled period and the time of the ambient data measurements this could
also change the relationship between Os response and Os concentrations. The regressions derived
from the 2007 modeling period are only applied to measurements made within 3 years of the
modeled time period. Although some emissions changes did occur over this time period, we
believe it is still reasonable to apply 2007 modeling to this relatively small window of
measurements which occurs before and after the modeling.
Os response is modeled for across-the-board reductions in U.S. anthropogenic NOX (and
VOC). These across-the-board cuts do not reflect actual emissions control strategies. The form,
locations, and timing of emissions reductions that would be undertaken to meet various levels of
the Os standard are unknown. The across-the-board emissions reductions bring levels down
uniformly across time and space to show how Os would respond to changes in ambient levels of
precursor species but do not reflect spatial and temporal heterogeneity that may occur in local
and regional emissions reductions. In cases where VOC reductions were modeled, equal
percentage NOX and VOC reductions were applied in the adjustment methodology. Regional
NOx reductions are likely to be the primary means used to reduce high Os concentrations at DV
monitors. In limited cases, VOC emissions reductions may also help lower high Os
concentrations at these locations. In actual control strategies, NOx and VOC reductions may be
applied in combination but are unlikely to be applied in equal percentages. The available
modeling constrained the NOx/VOC case to this type of control scenario. The across-the-board
cuts and the equal percentage NOx and VOC reductions scenario does not optimize the lowest
cost or least total emissions combinations as state and local agencies will likely attempt to
achieve.
4-41
-------
Table 4-7. Summary of Qualitative Uncertainty Analysis of Key Air Quality Elements in the Os NAAQS Risk Assessment.
Source
Description
Potential influence of
uncertainty on risk estimates
Direction
Magnitude
Knowledge-
Base
uncertainty*
Comments (KB: knowledge base, INF: influence of
uncertainty on risk estimates)
A. Os measurements
Ozone concentrations
measured by ambient
monitoring instruments have
inherent uncertainties
associated with them.
Additional uncertainties due to
other factors may include:
- monitoring network
locations
- Ozone monitoring
seasons
- monitor
malfunctions
-wildfire and smoke
impacts
- interpolation of
missing data
Both
Low
Low
KB: Ozone measurements are assumed to be accurate to within
V-i of the instrument's Method Detection Limit (MDL), which is 2.5
ppb for most instruments. EPA requires that routine quality
assurance checks are performed on all instruments, and that all
data reported to AQS are certified by both the monitoring agency
and the corresponding EPA regional office.
KB: Monitor malfunctions sometimes occur causing periods of
missing data or poor data quality. Monitoring data affected by
malfunctions are usually flagged by the monitoring agency and
removed from AQS. In addition, the AQS database managers
run several routines to identify suspicious data for potential
removal.
KB: There is a known tendency for smoke produced from
wildfires to cause interference in Os instruments. Measurements
collected by Os analyzers were reported to be biased high by
5.1-6.6 ppb per 100 ug/m3 of PIVh.sfrom wildfire smoke, EPA,
2007). However, smoke concentrations high enough to cause
significant interferences are infrequent and the overall impact is
believed to be minimal.
KB: Missing intervals of 1 or 2 hours in the measurement data
were interpolated, which may cause some additional uncertainty.
However, due to the short length of the interpolation periods,
and the tendency for these periods to occur at night when Os
concentrations are low, the overall impact is believed to be
minimal.
INF: EPA's current Os monitoring network requirements have an
urban focus. Rural areas where Os concentrations are lower
tend to be under-represented by the current monitoring network.
The network requirements also state that at least one monitor
within each urban area must be sited to capture the highest Os
concentrations in that area, which may cause some bias toward
higher measured concentrations.
INF: Each state has a required Os monitoring season which
varies in length from May- September to year-round. Some
4-42
-------
Source
Description
Potential influence of
uncertainty on risk estimates
Direction
Magnitude
Knowledge-
Base
uncertainty*
Comments (KB: knowledge base, INF: influence of
uncertainty on risk estimates)
states turn their Os monitors off during months outside of the
required season, while others leave them on. This can cause
discrepancies in the amount of data available, especially in
months outside of the required monitoring season. The risk
estimates attempt to minimize these impacts by focusing only on
months where 03 monitoring is required.
B. Voronoi Neighbor
Averaging (VNA)
spatial fields
VNA is a spatial interpolation
technique used to estimate Os
concentrations in unmonitored
areas, which has inherent
uncertainty
Both
Low-Medium
Low-Medium
KB: VNA interpolates monitored hourly Os concentrations to
provide estimates of Os exposure at each census tract in the 15
urban areas. The VNA estimates are weighted based on
distance from neighboring monitoring sites, thus the amount of
uncertainty tends to increase with distance from the monitoring
sites.
KB: The 15 urban areas each had fairly dense monitoring
networks which were generally sufficient to capture spatial
gradients in Os concentrations. The use of hourly data to create
the VNA fields instead of daily or other aggregates also served
to reduce uncertainty by better capturing relationships in the
diurnal patterns between Os monitors.
C.CMAQ modeling
Model predictions from CMAQ,
like all deterministic
photochemical models, have
both parametric and structural
uncertainty associated with
them
Both
Medium
Medium
KB: Structural uncertainties are uncertainties in the
representation of physical and chemical processes in the model.
These include: choice of chemical mechanism used to
characterize reactions in the atmosphere, choice of land surface
model and choice of planetary boundary layer model.
KB: Parametric uncertainties include uncertainties in model
inputs (hourly meteorological fields, hourly 3-D gridded
emissions, initial conditions, and boundary conditions)
KB: Uncertainties due to initial conditions are minimized by using
a 10 day ramp-up period from which model results are not used.
KB: Evaluations of models against observed pollutant
concentrations build confidence that the model performs with
reasonable accuracy despite the uncertainties listed above. A
comprehensive model evaluation provided in Appendix 4B
shows generally acceptable model performance which is
equivalent or better than typical state-of-the science regional
modeling simulations as summarized in Simon et al (2012).
However, both under-estimations and over-estimations do occur
at some times and locations. Generally the largest mean biases
4-43
-------
Source
Description
Potential influence of
uncertainty on risk estimates
Direction
Magnitude
Knowledge-
Base
uncertainty*
Comments (KB: knowledge base, INF: influence of
uncertainty on risk estimates)
occur on low Os days during the summer season. In addition, the
model did not fully capture rare wintertime high Os events
occurring in the Western U.S.
D. Higher Order
Decoupled Direct
Method (HDDM)
HDDM allows for the
approximation of Os
concentrations under alternate
emissions scenarios without
re-running the model
simulation multiple times using
different emissions inputs. This
approximation becomes less
accurate for larger emissions
perturbations especially under
nonlinear chemistry conditions.
Both
Medium
Medium
KB: To accommodate increasing uncertainty at larger emissions
perturbations, the HDDM modeling was performed at three
distinct emissions levels to allow for a better characterization of
Os response over the entire range of emissions levels. The
replication of brute force hourly Os concentration model results
by the HDDM approximation was quantified for 50% and 90%
NOx cut conditions for each urban study areas (as shown in
Appendix 4D). At 50% NOX cut conditions, HDDM using
information from these multiple simulations predicted hourly Os
concentrations with a mean bias and a mean error less than +/-
1 ppb in all urban study areas compared to brute force model
simulations. At 90% NOX cut conditions, HDDM using
information from these multiple simulations predicted hourly Os
concentrations with a mean bias less than +/- 3ppb and a mean
error less than +/- 4 ppb in all urban study areas.
E. Application of
HDDM sensitivities
to ambient data
In order to apply
modeled sensitivities to
ambient measurements,
regressions were developed
which relate Os response to
emissions perturbations with
ambient Os concentrations for
every season, hour-of-the-day
and monitor location. Applying
Os responses based on this
relationship adds uncertainty.
Both
Medium
Medium
KB: Preliminary work showed that the relationships developed
with these regressions were generally statistically significant for
most season, hour-of-the-day, and monitor location
combinations for 2005 modeling in Detroit and Charlotte.
Statistical significance was not evaluated for each regression in
this analysis since there were over 460,000 regressions created
(322 monitors * 5 sensitivity coefficients * 3 emissions levels * 4
seasons * 24 hours = 463,680 regressions). Statistical inference
can quantify the goodness of fit for the modeled relationships.
However it is not possible to quantify the applicability of this
modeled relationship to the actual atmosphere.
KB: The regression model provided both a central tendency
estimate and a standard error estimate for Os response at each
measured hourly Os concentration. The base analysis used the
central tendency which will inherently dampen some of the
variability in Os response. The standard error of each sensitivity
coefficient was propagated through the calculation of predicted
hourly and daily maximum 8-hr average Os concentrations at
various standard levels. These standard errors reflect
uncertainty associated with estimating the central tendency.
Since emissions reductions increased for lower standard levels
the standard errors were larger for adjustments to lower
4-44
-------
Source
Description
Potential influence of
uncertainty on risk estimates
Direction
Magnitude
Knowledge-
Base
uncertainty*
Comments (KB: knowledge base, INF: influence of
uncertainty on risk estimates)
standards. Mean (95th percentile) standard errors in hourly Os for
the 75 ppb adjustment case ranged from 0.13 (0.26) to 1.18
(2.87) ppb in the 15 urban study areas. Mean (95th percentile)
standard errors in hourly Os for the 65 ppb adjustment case
ranged from 0.54 (1.07) to 1.39 (2.98) ppb. The largest standard
errors occurred in Los Angeles and New York.
F. Applying modeled
sensitivities to un-
modeled time
periods
Relationships between Os
response and hourly Os
concentration were developed
based on 8 months of
modeling: January and April-
October 2007. These
relationships were applied to
ambient data from 2006-2010.
Some locations monitor for
months not included in this
modeling (February, March,
November, and December)
while others do not.
Both
Medium
Medium
KB: The eight months that were modeled capture a variety of
meteorological and emissions conditions. Applying these 2007
sensitivities to other years with potentially different meteorology
and emissions adds uncertainty to the relationship between Os
response and Os concentrations. The regressions derived from
the 2007 modeling period are only applied to measurements
made within 3 years of the modeled time period. Although some
emissions changes did occur over this time period, we believe it
is still reasonable to apply 2007 modeling to this relatively small
window of measurements which occurs before and after the
modeling.
INF: Seasonal relationships were developed between Os
response to emissions reductions and Os concentration.
Summer was the only season for which modeling data was
created for all months (June, July, August). The winter
relationships were developed based on January modeling, the
spring relationships were developed based on April/May
modeling, and the autumn relationships were developed based
on September/October modeling. The reduction in data points
(31 or 61 instead of-90) increases uncertainty in the statistical
fit for these months. In addition, the modeling generally showed
more Os disbenefits to NOX decreases in cooler months. So
applying April/May relationships to March and
September/October relationships to November could potentially
underestimate Os increases that would happen in those two
months in the five urban study areas which measure Os during
March and/or November: Dallas, Denver, Houston, Los Angeles,
and Sacramento.
G. Assumptions of
across-the-board
emissions
reductions
Ozone response is modeled
for across-the-board
reductions in U.S.
anthropogenic NOX (and VOC).
Both
Low-Medium
Low-Medium
KB: The form, locations, and timing of emissions reductions that
would be undertaken to meet various levels of the Os standard
are unknown. The across-the-board emissions reductions bring
levels down uniformly across time and space to show how Os
4-45
-------
Source
Description
Potential influence of
uncertainty on risk estimates
Direction
Magnitude
Knowledge-
Base
uncertainty*
Comments (KB: knowledge base, INF: influence of
uncertainty on risk estimates)
These across-the-board cuts
do not reflect actual emissions
control strategies.
would respond to changes in ambient levels of precursor species
but do not reflect spatial and temporal heterogeneity that may
occur in local and regional emissions reductions.
H. Assumption of
equal percentage
NOx and VOC
reductions
In cases where VOC
reductions were modeled,
equal percentage NOX and
VOC reductions were applied
in the adjustment
methodology.
Both
Low-Medium
Medium
KB: NOx reductions are likely to be the primary means used to
reduce high Os concentrations at DV monitors. In limited cases,
VOC emissions reductions may also help lower high Os
concentrations at these locations. NOX and VOC reductions may
be applied in combination but are unlikely to be applied in equal
percentages. The available modeling constrained the NOX/VOC
case to this unrealistic scenario. The equal percentage NOX and
VOC reductions scenario does not optimize the lowest cost or
least total emissions combinations as state and local agencies
will likely attempt to achieve.
I. Downscaler
Downscaler combines
monitored and modeled
concentrations to produce a
"fused" air quality surface.
Uncertainties may occur in
sparsely monitored regions, or
in urban areas with dense
monitoring networks and large
spatial gradients.
Both
Low-Medium
Low-Medium
KB: Downscaler combines modeled and monitored
concentrations to provide estimates of Os concentrations in
unmonitored areas while correcting model biases near monitors.
The cross-validation analysis in Appendix 4A shows that
Downscaler generally gives more accurate estimates of air
quality in monitored locations than either the monitored or
modeled values alone. However, it is not possible to quantify the
uncertainty associated with the estimates in unmonitored
locations.
KB: The air quality surfaces modeled by Downscaler for the
national-scale risk assessment were seasonal average
concentrations, which tend to have smaller spatial gradients than
other metrics such as peak concentrations, and thus less
uncertainty.
INF: The cross-validation analysis in Appendix 4A also shows
that Downscaler tends to over-estimate low concentrations and
under-estimate high concentrations. The mean bias in the
estimates in monitored locations is nearly zero, but monitor
locations are often chosen to capture the highest concentrations,
thus there might be some bias towards higher concentrations in
unmonitored areas.
4-46
-------
4.6 REFERENCES
Abt Associates, Inc. 2010a. Environmental Benefits and Mapping Program (Version 4.0).
Bethesda, MD. Prepared for U.S. Environmental Protection Agency Office of Air Quality
Planning and Standards. Research Triangle Park, NC. .
Abt Associates, Inc. 201 Ob. Model Attainment Test Software (Version 2). Bethesda, MD.
Prepared for the U.S. Environmental Protection Agency Office of Air Quality Planning
and Standards. Research Triangle Park, NC.
Bell, M.L.; A. McDermott; S.L. Zeger; J.M. Samet and F. Dominici. 2004. Ozone and short-term
mortality in 95 U.S. urban communities, 1987-2000. JAMA. 292:2372-2378.
Berrocal, V.J.; A.E. Gelfand and D.M. Holland. 2012. Space-time data fusion under error in
computer model output: an application to modeling air quality. Biometrics. 68(3):837-
848.
Chen, J.R.; Zhao; Z. Li. 2004. Voronoi-based k-order neighbor relations for spatial analysis.
ISPRS J Photogrammetry Remote Sensing. 5 9( 1 -2): 60-72.
Duff, M.; R. L. Horst; T.R. Johnson, 1998. Quadratic Rollback: A Technique to Model Ambient
Concentrations Due to Undefined Emission Controls. San Diego, CA: Presented at the
Air and Waste Management Annual Meeting, June 14-18, 1998.
Fann, N.; A.D. Lamson; S.C. Anenberg; K. Wesson; D. Risley; B.J. Hubbell. 2012. Estimating
the national public health burden associated with exposure to ambient PM2.5 and ozone.
Risk Analysis. 32:81-95.
Gold, C. 1997. Voronoi Methods in GIS, Vol. 1340. In Algorithmic Foundation of Geographic
Information Systems (Kereveld M., J. Nievergelt, T. Roos, P. Widmayer, eds.). Lecture
notes in Computer Science, Berlin: Springer-Verlag, 21-35.
Hall, E.; A. Eyth; S. Phillips. 2012. HierarchicalBayesianModel (HBM)-DerivedEstimates of
Air Quality for 2007: Annual Report. (EPA document number EPA/600/R-12/538).
< http://www.epa.gov/heasd/sources/projects/CDC/AnnualReports/2007_HBM.pdf>.
Jerrett, M.; R.T. Burnett; C.A. Pope, III; K. Ito; G. Thurston; D. Krewski; Y. Shi, E. Calle and
M. Thun. 2009. Long-term ozone exposure and mortality. New England Journal of
Medicine. 360:1085-1095.
Johnson, T. 2002. A Guide to Selected Algorithms, Distributions, and Databases Used in
Exposure Models Developed by the Office of Air Quality Planning and Standards.
Prepared by TRJ Environmental, Inc. for the U.S. Environmental Protection Agency.
Research Triangle Park, NC: Office of Research and Development.
National Research Council of the National Academies. 2008. Estimating Mortality Risk
Reduction and Economic Benefits from Controlling Ozone Air Pollution. Washington,
DC: The National Academies Press.
Rizzo, M. 2005. A Comparison of Different Rollback Methodologies Applied to Ozone
Concentrations. Posted on November 7, 2005.
4-47
-------
Rizzo, M. 2006. A Distributional Comparison between Different Rollback Methodologies
Applied to Ambient Ozone Concentrations. Posted on May 31, 2006.
Simon, H.; K. Baker; N. Possiel; F. Akhtar; S. Napelenok; B. Timin; B. Wells. 2012. Model-
based Rollback Using the Higher Order Direct Decoupled Method (HDDM).
.
Simon, H.; K. R. Baker; F. Akhtar; S.L. Napelenok; N. Possiel; B. Wells and B. Timin. 2013. A
direct sensitivity approach to predict hourly ozone resulting from compliance with the
National Ambient Air Quality Standard. Environmental Science and Technology.
47:2304-2313.
Smith, R.L., B. Xu, P. Switzer. 2009b. Reassessing the relationship between ozone and short-
term mortality in U.S. urban communities. Inhale Toxicol. 21:37-61.
Timin, B.; K. Wesson and J. Thurman. 2010. Application of Model and Ambient Data Fusion
Techniques to Predict Current and Future Year PM2.5 Concentrations in Unmonitored
Areas, in D.G. Steyn and S.T. Rao (eds.), Air Pollution Modeling and Its Application XX,
Netherlands: Springer, pp. 175-179.
U.S. Environmental Protection Agency. 2007. Review of the National Ambient Air Quality
Standards for Ozone: Policy Assessment of Scientific and Technical Information OAQPS
Staff Paper. Washington, DC: EPA Office of Air and Radiation. (EPA document number
U. S. EPA. 2012a. Integrated Science Assessment for Ozone and Related Photochemical
Oxidants: Third External Review Draft. Research Triangle Park, NC: EPA Office of Air
Quality Planning and Standards. (EPA document number EPA-452/R-07-007;
EPA/600/R-10/076C).
U.S. EPA. 2012b. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
Documentation (TRIM. Expo / APEX, Version 4.4) Volume I: User's Guide. Research
Triangle Park, NC: Office of Air Quality Planning and Standards. (EPA document
number EPA-452/B-12-001a). .
U.S. EPA. 2012c. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
Documentation (TRIM. Expo / APEX, Version 4.4) Volume II: Technical Support
Document. Research Triangle Park, NC: Office of Air Quality Planning and Standards.
(EPA document number EPA-452/B-12-001b).
.
Wells, B.; K. Wesson and S. Jenkins. 2012. Analysis of Recent U.S. Ozone Air Quality Data to
Support the O3 NAAQS Review and Quadratic Rollback Simulations to Support the First
Draft of the Risk and Exposure Assessment.
.
Zhang, L.; DJ. Jacob; N.V. Smith-Downey; D.A. Wood; D. Blewitt; C.C. Carouge; A. van
Donkelaar; D.B.A. Jones; L.T. Murray and Y. Wang. 2011. Improved estimate of the
policy-relevant background ozone in the United States using the GEOS-Chem Global
Model with l/2°x2/3° horizontal resolution over North America." Atmospheric
Environment. 45:6769-6776.
Zanobetti, A. and J. Schwartz. 2008. Mortality displacement in the association of ozone with
mortality: an analysis of 48 cities in the United States. American Journal of Respiratory
and Critical Care Medicine. 177:184-189.
4-48
-------
5 CHARACTERIZATION OF URBAN-SCALE HUMAN EXPOSURE
As part of the previous 2007 ozone (Os) NAAQS review, EPA staff conducted exposure
analyses for the general population, all school-age children (ages 5-18), all active school-age
children,1 and asthmatic school-age children (U.S. EPA, 2007a,b). Population-based exposures
were simulated for these study groups residing in 12 urban study areas2 considering recent years
of air quality and for just meeting the existing 8-hour (8-hr) standard and several potential
alternative 8-hr standards. In each of these 12 urban study areas, estimated daily maximum 8-hr
average Os exposures concomitant with moderate or greater exertion levels were compared with
benchmark exposures concentrations, i.e., concentration levels at which adverse health responses
were observed in controlled human exposure studies while exercising. EPA also estimated risk of
impaired lung function for two study groups - all school-age children and asthmatic school-age
children - in these same 12 study areas using exposure-response (E-R) relationships developed
from controlled human exposure studies combined with the population-based exposures.
The exposure analysis conducted for this current NAAQS review builds upon the
methodology and lessons learned from the exposure analyses conducted in previous Cb reviews
(U.S. EPA, 1996a, 2007a,b) and information provided in the Os ISA (U.S. EPA, 2013). Here, we
estimate exposures for people residing in 15 urban study areas in the U.S.3 The population-based
exposures to ambient Os concentrations were modeled using EPA's Air Pollutants Exposure
(APEX) (US EPA, 2012a,b). Exposures were calculated considering ambient Os concentrations
in recent years, using 2006 to 2010 spatially interpolated ambient monitoring data. Exposures
were also estimated considering alternative air quality scenarios, that is, where Os concentrations
just meet the existing 8-hr Os NAAQS and at several other standard levels considering the same
indicator, form, and averaging time, based on adjusting data as described in Chapter 4.
Exposures were modeled for (1) all school-age children (ages 5-18), (2) asthmatic school-age
children (ages 5-18), (3) asthmatic adults (ages 19-95), and (4) all older adults (ages 65-95), each
1 In the previous 2007 exposure assessment, a study group of active school-age children was identified as children
having their median daily physical activity index (PAI) over the exposure period > 1.75, an activity level
characterized by exercise physiologists as being "moderately active" or "active" (McCurdy, 2000).
2 The twelve study areas evaluated in the 2007 exposure assessment were Atlanta, Boston, Chicago, Cleveland,
Detroit, Houston, Los Angeles, New York, Philadelphia, Sacramento, St. Louis, Washington D.C. (an area which
at that time was modeled to include Baltimore as part of the Baltimore-Northern Virginia MSA).
3 In addition to the twelve study areas identified in the 2007 exposure assessment, staff has added Dallas and
Denver, while also separately modeling Baltimore (from Washington D.C.) in this current assessment. Inclusion
of Seattle, WA was considered but not included due to a lack of appropriate monitoring data.
5-1
-------
while at moderate or greater exertion level at the time of exposure.4 The strong emphasis on
children, asthmatics, and older adults reflects the finding of the last Os NAAQS review (U.S.
EPA, 2007a) and the Os ISA (U.S. EPA, 2013, Chapter 8) that these are important at-risk groups.
As was done in the previous review, estimated daily maximum 8-hr average exposures are
compared with exposure benchmark levels based on adverse health effects observed in controlled
human exposure studies. Specific exposure model output of interest for this HREA chapter are
the percent (and number) of people exposed one time (or multiple occurrences) at or above these
8-hr average Os concentrations of concern, all while at moderate or greater exertion levels. These
exposure modeling results provide context on exposure across each of the urban study areas
given the existing and alternative ambient air quality standard scenarios (i.e., who is likely
exposed, how many in people in the exposure study group of interest, how often per year, etc.)
though not directly providing an estimate of human health risk per se. However, the complete
time series of individual exposures estimated by APEX in this assessment (not simply the daily
maximum 8-hr average concentrations) then serve as input to a lung function risk module that
estimates health risk based on information developed from controlled human exposure studies
(Chapter 6).
This chapter first provides a brief overview of human exposure and exposure modeling
using APEX (section 5.1), the scope of this Os exposure assessment and key inputs used to
model exposure in the 15 U.S. study areas selected (section 5.2), and followed by the main body
exposure results (section 5.3). Then, section 5.4 presents an assemblage of targeted analyses
designed to provide additional insight to the main body of exposure results by focusing on
important data inputs, additional at-risk populations, lifestages, or scenarios, influential attributes
in estimating exposures, and performance evaluations. The results of these and other exposure
model targeted analyses are integrated in a variability and uncertainty characterization section
(section 5.5) along with a final section summarizing the key observations for this chapter
(section 5.6).
5.1 SYNOPSIS OF OZONE EXPOSURE AND EXPOSURE MODELING
5.1.1 Human Exposure
Human exposure to a contaminant is defined as "contact at a boundary between a human
and the environment at a specific contaminant concentration for a specific interval of time," and
4 The "all school-age children" study group includes both asthmatic and non-asthmatic children ages 5 to 18. The
"all older adults" includes both asthmatic and non-asthmatic older adults ages 65 to 95. Note also that the 8-hr
average exposure of interest in both this and the previous assessment was concomitant with moderate or greater
exertion for all study groups.
5-2
-------
has units of concentration times duration (National Research Council, 1991). For air pollutants
the contact boundary is nasal and oral openings in the body, and personal exposure of an
individual to a chemical in the air for a discrete time period is fundamentally quantified as (Lioy,
1990; National Research Council, 1991):
Equation (5-1)
where E[ft/2] is the personal exposure or exposure concentration during the time period
from t\ to t2, and C(f) is the concentration at time t in the breathing zone. The breathing rate at
the time of exposure will influence the dose received by the individual. While we do not directly
estimate dose in this assessment, intake is the total Os inhaled (i.e., exposure concentration,
duration, and ventilation combined).5
5.1.2 Estimating Ozone Exposure
Exposure to Os can be directly estimated by monitoring the concentration of Os in a
person's breathing zone (close to the nose/mouth) using a personal exposure monitor. Studies
employing this measurement approach have been reviewed in the Cb ISA and EPA Os Air
Quality Criteria Documents (U.S. EPA, 1986, 1996b, 2006, 2013). Personal exposure
measurements from these studies are useful in describing a general range of exposure
concentrations (among other reported measurement data) and in identifying factors that may
influence varying exposure levels. However, these measurement studies are largely limited by
the disparity between sample measurement duration and exposure concentration averaging-times
of interest and in appropriately capturing variability in population exposure occurring over large
geographic areas , particularly when considering both concentration (e.g., spatial variability) and
population (e.g., age, sex) attributes that influence exposure.
Ozone exposure for individuals, small groups of individuals or large populations can be
calculated indirectly (or modeled) using Equation (5-1). When employing such an approach in a
population-based exposure assessment, two basic types of input data are needed; a time-series of
Os concentrations that appropriately represents spatial heterogeneity in Os concentrations and a
corresponding time-series of locations visited by the people exposed. When considering air
pollutant concentrations, population exposure models are commonly driven by ambient
concentrations. These ambient concentrations may be provided by monitoring data, by air quality
model estimates, or perhaps by a combination of these two data sources. Then, an understanding
5 In Chapter 6, the estimation of risk combines the time series of both the personal exposure concentrations and
ventilation rate, among other variables in essentially calculating a dose, though not explicitly output from the
model.
5-3
-------
of the relationships between ambient pollutants and the locations people occupy is needed. This
is because human exposure, regardless of the pollutant or whether one is interested in individual
or population exposure, depends on where an individual is located, how long they occupy that
location, and what the pollutant concentration at the point of contact is. Furthermore, if interested
in air pollutant intake rate or dose, one needs to know what activity the person is performing
while exposed.
Thus, the types of measurement and modeling studies that provide information for more
realistically estimating exposure to Os can be augmented from the above list to include studies
of: (1) Os formation, deposition, and decay, (2) people's locations visited and activities
performed, (3) human physiology, and (4) local scale meteorological measurements and/or
modeling. Useful data derived from these varied studies are Os concentrations (i.e., fixed site,
personal exposure, indoor and outdoor locations), built environment physical factors (i.e., air
exchange rates (AERs), infiltration rates, decay and deposition rates), human time-location-
activity patterns (minute-by-minute, hourly, daily, and longer-term), time-averaged or activity-
specific breathing rates among varying sexes and/or lifestages, and hourly ambient temperatures.
When integrating these varied data (among others such as population demographics and
disease prevalence) and understanding factors affecting exposure, exposure models can extend
beyond the limited information given by measurement data alone. For example, an exposure
model can reasonably estimate exposures for any perceivable at-risk population (e.g., asthmatics
living in a large urban area) and considering any number of hypothetical air quality conditions
(e.g., just meeting a daily maximum 8-hr average concentration of 70 ppb). Exposure models that
account for variability in human physiology can also realistically estimate pollutant intake by
using activity-specific ventilation rates. These types of measurements cannot realistically be
performed for a study group or population of interest, particularly when considering time, cost,
and other constraints. The following section provides an overview of how such exposure
modeling can be done using APEX, the model developed by EPA to perform such calculations
and used to estimate Os exposures in this FtREA.
5.1.3 Modeling Ozone Exposure Using APEX
EPA has developed the APEX model for estimating human population exposure to
criteria and air toxic pollutants, used most recently in estimating exposures for the Os (U.S. EPA,
2007b), nitrogen dioxide (U.S. EPA, 2008), sulfur dioxide (U.S. EPA, 2009a), and carbon
monoxide (U.S. EPA, 2010) NAAQS reviews. APEX is a probabilistic model designed to
account for the numerous sources of variability that affect people's exposures. An overview of
the approaches used by APEX to estimate exposure concentrations is found in Appendix 5A with
details provided in U.S. EPA (2012a,b).
5-4
-------
Briefly, APEX simulates the movement of individuals through time and space and
estimates their exposure to a given pollutant while occupying indoor, outdoor, and in-vehicle
locations. The model stochastically generates simulated individuals in selected study areas using
census-derived probability distributions for demographic characteristics. Population
demographics are drawn from the 2000 Census data6 at a tract level, and a national commuting
database based on 2000 Census data provides home-to-work commuting flows between tracts.7
Any number of individuals can be simulated, and collectively they approximate a random
sampling of people residing in a particular study area.
Daily activity patterns for individuals in a study area, an input to APEX, are obtained
from detailed daily time-location-activity pattern survey data that are compiled in the
Consolidated Human Activity Database (CHAD) (McCurdy et al., 2000; U.S. EPA, 2002). These
daily diaries are used to construct a sequence of locations visited and activities performed for
APEX simulated individuals consistent with their demographic characteristics, day-type (e.g.,
weekend or weekday), and season of the year, as defined by ambient temperature regimes
(Graham and McCurdy, 2004). The time-location-activity data input to APEX are linked with
personal attributes of the surveyed individuals, such as age, sex, employment status, day-of-week
surveyed, and daily maximum and daily mean temperature. These specific personal attribute data
are then used by APEX to best match the daily diary with the simulated study group of interest,
using the same variables as first-order diary selection characteristics. The approach is designed to
capture the important attributes contributing to an individuals' time-location-activity pattern, and
of particular relevance here, time spent outdoors (Graham and McCurdy, 2004). In using a
diverse collection of time-location-activity diaries that capture the duration and frequency of
occurrence of visitations/activities performed, APEX can simulate expected variability in human
behavior, both within and between individuals. This, combined with exposure concentrations,
allows for the reasonable estimation of the magnitude, frequency, pattern, and duration of
exposures an individual experiences.
A key concept in modeling exposure using APEX is the microenvironment, a term that
refers to the immediate surroundings of an individual at a particular time. APEX has a flexible
approach for modeling micro-environmental concentrations whereas the model user defines the
type, number and characteristics of the microenvironments to be modeled. Typical
microenvironments include indoors at home, indoors at school, near roadways, inside cars, and
outside home. In this exposure assessment, all microenvironmental Os concentrations are derived
from ambient Os concentrations input to APEX and are estimated using either a mass-balance or
6 Due to resource limitations and data availability, the 2010 Census data were not processed in time for use in this
HREA, except for a limited sensitivity evaluation performed in section 5.4.4.
7 There are approximately 65,400 census tracts in the -3,200 counties in the U.S.
5-5
-------
transfer factors approach, selected by the user. The mass balance approach assumes that the air
in an enclosed microenvironment is well-mixed and that the air concentration is spatially
uniform at a given time within the microenvironment. The approach employs indoor-to-outdoor
AERs (i.e., number of complete air exchanges per hour) and considers removal mechanisms such
as deposition to building surfaces and chemical decay rates. The transfer factors model is simpler
than the mass balance model, and employs two variables, a proximity factor, used to account for
proximity of the microenvironment to sources or sinks of pollution, or other systematic
differences between concentrations just outside the microenvironment and the ambient
concentrations, and & penetration factor, which quantifies the degree to which the outdoor air
penetrates into the microenvironment.
Activity-specific simulated breathing rates of individuals are used in APEX to
characterize intake received from an exposure. This is done because controlled human exposure
studies have shown adverse health outcomes are associated with both elevated concentrations
and study participant exertion levels. The breathing rates calculated by APEX are derived from
the energy expenditure associated with each simulated individual's activity performed, adjusted
for age- and sex-specific physiological parameters (Graham and McCurdy, 2005). The energy
expenditure estimates themselves are derived from distributions of METS8 (or metabolic
equivalents of work) associated with every activity performed (McCurdy et al., 2000, using
Ainsworth et al., 1993).
An important feature of APEX is the ability to account for variability in exposure by
representing input variables as statistical distributions along with conditional variables, where
appropriate. For example, the distribution of AERs in a home, office, or motor vehicle can
depend on the type of heating and air conditioning present, which are also stochastic inputs to the
model, as well as the ambient temperature on a given day. The user can choose to keep the value
of a stochastic parameter constant for the entire simulation (appropriate for the volume of a
house), or can specify that a new value shall be drawn hourly, daily, or seasonally from specified
distributions.
Finally, APEX calculates a unique time-series of exposure concentrations on the order of
minutes or smallest diary event duration that each simulated individual may experience during
the modeled time period, based on that individual's estimated microenvironmental
concentrations and the time spent in each of sequence of microenvironments visited according to
the time-location-activity diary of each individual. Then, hourly average exposures of each
! METS are a dimensionless ratio of the activity-specific energy expenditure rate to the basal or resting energy
expenditure rate. The metric is used by exercise physiologists and clinical nutritionists to estimate work
undertaken by individuals as they go through their daily activities (Montoye et al., 1996).
5-6
-------
simulated individual are estimated using time-weighted averages of the within-hour exposures.
From hourly exposures, APEX calculates other time averaged exposures of interest (e.g., 8-hr or
daily average) that a simulated individual experiences during the modeled period.
5.2 SCOPE OF THE EXPOSURE ASSESSMENT
This section broadly presents the scope of the exposure assessment including descriptions
of the modeling domains, ambient concentrations used, time periods and populations modeled, as
well as identifying key approaches, inputs and outputs used by APEX in estimating population
Os exposures. Detailed descriptions regarding APEX modeling, model inputs and other
supporting information are provided in Appendices 5A-E and the APEX user's guide and
technical support documents (U.S. EPA 2012a,b). Figure 5-1 illustrates the general conceptual
framework for generating our population exposure concentrations, including the time-series of
exposure and ventilation rate output generated for input to the lung function risk calculations in
Chapter 6.
5.2.1 Urban Areas Selected
The selection of urban areas to include in the exposure assessment considered the
location of Os epidemiological studies, the availability of ambient Os monitoring data, and the
desire to represent a range of geographic areas, encompassing variability in climate and
population demographics. Specifically, the criteria included the following:
• The overall set of urban locations should represent a range of geographic areas, urban
population demographics, and climate, beginning with study areas selected in the 2007
Os NAAQS review.
• The locations should be focused on areas that do not meet or are close to not meeting the
existing 8-hr Os NAAQS and should include areas having Os non-attainment
designations.
• There must be sufficient Os ambient air quality data for the recent 2006-2010 period.
• The study areas should include the 12 urban study areas modeled in the epidemiologic-
based risk assessment (Chapter 7).
5-7
-------
Population Information
US Census Tract
Population
Distributions
US Census Tract
Home-to-Work
Commuting
X
\
Air Quality
(Chiptor4)
Census Tract Hourly
03 Concentrations:
recent conditions and
adjusted to jjst meet
existing and
alternative standards
Meteorology
Urban Area Hourly
Temperatures (ISH)
US Census Tract
Asthma Prevalence
US Census Tract
Employment
Distributions
Individual Information
Daily Human Time
Location Activity
Patterns (CHAD)
Metabolic Equivalents
of Work (METS)
Distributions
1
Microanvirorinfwntu Informnon
APEX
In-Vehicle and Near-
Road Proximity and
Penetration Factor
Distributions
Urban Area Vehicle
Miles Traveled
(US DOT)
Indoor Air Exchange
Rate (AER) and O3
Decay Distributions
T
Urban Area Air
Conditioning
Prevalence (AHS)
Calculate 03 Exposure
Concentrations for
Individuals Comprising
Population
T
Number and percent of
persons with 8-hour
average exposures at or
above selected
benchmark level per
year while at moderate
or greater exertion
Distributions of
Physiological
Attributes &. Ventilation
V Equations
Number of persons with
multiple 8-hour average
exposures at or above
selected benchmark
level per year while at
moderate or greater
exertion
C Time-series of 03 exposure
oncentrations and ventilation rate for
individuals
I
Output ID FEV1 Modelng
None
AHS: AiMriun Hawing Survty
anHUm
CHAD: US ERACa
d Humni
EH Nrtand ClirMfc Dm Ctrtof
US DOT: US Dtpir
F«tard Hghwty MntaWnflon
Figure 5-1. Conceptual Framework Used for Estimating Study Area Population Os Exposure Concentrations.
5-8
-------
Based on these selection criteria, we chose the 15 study areas listed in Table 5-1 to
develop our population exposure estimates. We then defined an air quality domain for each
study area, broadly bounding the ambient concentration field where exposures were to be
estimated. To do this, we evaluated (1) counties modeled in the previous 2007 Os NAAQS
review common to current study areas, (2) political/statistical county aggregations (e.g., whether
in a metropolitan statistical areas or MS As), and (3) if the study area was designated as a non-
attainment area (NAA), the counties that were part of the NAA list. We identified a final list of
215 counties9 to comprise the air quality domain for the 15 study areas, the names of which are
provided in Appendix 5B.
5.2.2 Time Periods Simulated
The exposure periods modeled are the Os seasons for which routine hourly Os monitoring
data were available for years 2006 to 2010 (Table 5-1), and defined by 40 CFR part 58,
Appendix D, Table D-3. These periods are designed to reasonably capture year-to-year
variability in ambient concentrations and meteorology and include most of the high
concentration events occurring in each area. Having this wide range of air quality data across
multiple years allows us to more realistically estimate a range of exposures, rather than using a
single year of air quality data. While the number of Os monitors in operation may vary slightly
from year to year, we assumed consistent representation by the complete set of all monitors
within a study area. Thus the spatial representation of each study area using the statistically
interpolated data remained consistent for each year over the simulation period (see section 5.2.3).
5.2.3 Ambient Concentrations Used
We used the available hourly ambient monitor concentration data within and around each
study area along with a statistical interpolation technique (Chapter 4) to estimate hourly census
tract concentrations within the counties comprising each study area. These concentrations served
as the 'base' air quality input for each study area year. Ambient concentrations were also
adjusted to just meet the existing standard (75 ppb, 4th highest 8-hr average, averaged over a 3-
year period) and alternative standard levels (70, 65, 60, and 55 ppb) using an air quality model
and the statistical interpolation technique (Chapter 4).
These estimated hourly census tract Os concentrations served as the APEX air districts.,
the basic ambient concentrations from which each simulated individual's microenvironmental
concentrations are estimated. Having these temporally and spatially resolved air districts in each
study area allows for better utilization of APEX spatial and temporal capabilities in estimating
' Of the 215 counties defining the air quality domain, 207 remained in the exposure model domain.
5-9
-------
exposure. Because APEX simulates where individuals are located at specific times of the day,
more realistic exposure estimates are obtained in simulating the contact of individuals with these
spatially and temporally diverse concentrations.
Table 5-1. General Characteristics of the Population Exposure Modeling Domain
Comprising Each Study Area.
Study Area
(Abbreviation)
Atlanta (ATL)
Baltimore (BAL)
Boston (BOS)
Chicago (CHI)
Cleveland (CLE)
Dallas (DAL)
Denver (DEN)
Detroit (DET)
Houston (HOU)
Los Angeles (LA)
New York (NY)
Philadelphia (PHI)
Sacramento (SAC)
St. Louis (STL)
Wash., DC (WAS)
All Study Areas
Os Season1
Mar1-Oct31
Apr1-Oct31
Apr 1-Sep30
Apr1-Oct31
Apr1-Oct31
Mar1-Oct31
Mar1-Sep30
Apr 1-Sep30
Jan 1-Dec31
Jan 1-Dec31
Apr1-Oct31
Apr1-Oct31
Jan 1-Dec31
Apr1-Oct31
Apr1-Oct31
-
Study Area Number of:
Counties
32
7
7
16
8
11
12
9
10
5
27
15
7
15
26
207
Ambient
Monitors
14
12
13
28
16
21
25
13
19
50
32
19
18
16
28
324
APEX Air
Districts
664
603
1,005
1,882
802
1,012
655
1,419
779
2,000
1,900
1,452
447
494
1,013
16,127
US Census
Tracts
678
618
1,028
2,055
879
1,036
675
1,454
802
3,352
4,889
1,555
461
518
1,037
21,037
People
All School
Age
Children
(age 5-1 8)
860,649
505,140
905,208
1,899,073
578,733
1,097,004
560,137
1,016,896
970,528
3,620,972
3,843,450
1,231,052
466,169
527,755
966,791
19,049,557
All Ages
(age 5-95)
3,850,951
2,209,226
4,449,291
8,345,373
2,692,846
4,698,392
2,626,239
4,572,479
3,925,054
14,950,340
18,520,868
5,506,954
1,926,598
2,340,325
4,498,374
85,113,310
Even though we estimated Os ambient concentrations occurring at all census tracts in
each county-level study area using the combined ambient monitor data, VNA interpolation, and
CMAQ-HDDM simulation approach, the urban study area exposure modeling domain was
defined as a subset of these census tracts and were selected by using an overlay of the ambient
monitor locations within the urban core of each study area's air quality domain and a 30 km
radius of influence. We used a 30 km zone of influence to remain consistent with what was done
in the 1st draft Os HREA, though in that exposure assessment, only the ambient monitoring data
were used to represent the APEX air districts, hence concentrations measured at a particular
monitoring site would be directly extrapolated outwards to all census tracts within 30 km of that
5-10
-------
site. In contrast, by incorporating the VNA estimated concentrations at all census tracts and
retaining the same 30 km radius of influence, we stress the significance of the monitor
information in defining the urban core air quality while also reasonably estimating concentration
gradients (where such gradients exist) with increasing distance from monitoring locations.
Thus, all air districts10 and census tracts (and their spatially varying concentrations) that
fall within the 30 km radius of each ambient monitor location were used to estimate the
exposures, defining the final exposure modeling domain in each study area (Table 5-1). The
monitor IDs used to select the census tracts to be modeled are provided in Appendix 5B, while
the complete list of census tract IDs where exposures were modeled are within the APEX control
files for each study area (and are the same for each simulation year).
5.2.4 Meteorological Data Used
APEX uses study area temperature data to select representative diaries for a particular
day and in selecting an appropriate air exchange rate used to calculate indoor residential
microenvironmental concentrations. APEX uses the data from the closest weather station to each
Census tract. To ensure reasonable coverage for each study area, a few to several meteorological
stations recording hourly surface temperature measurements were identified using data obtained
from the National Weather Service ISH data files.11 Details regarding the meteorological stations
selected and data processing are given in Appendix 5B, section 5B-8. Briefly, APEX requires the
temperature input data to be 100% complete. In general, any missing values were filled using a
linear interpolation or regression approach that employs information from proximal
meteorological stations. Details regarding the number of missing values in any meteorological
station by year are provided in Appendix 5B Table 5B-10.
5.2.5 Populations Simulated
Exposure was estimated for four at-risk study groups residing in each study area: all
school-age children (ages 5-18), asthmatic school-age children, asthmatic adults (ages 19-95),
and all older adults (ages 65-95). Due to the increased amount of time spent outdoors engaged in
relatively high levels of physical activity (which increases intake), school-age children as a group
are particularly at risk for experiencing Os-related health effects (U.S. EPA, 2013, Chapter 8).
We report results for all school-age children down to age five, recognizing an increasing trend
for younger children to attend school. Some U.S. states allow 4-year-olds to attend kindergarten,
and most states have preschool programs for children younger than five. In 2000, six percent of
10 The original number of air quality districts for New York and Los Angeles needed to be reduced by about half due
to exceeding personal computer memory capacity when APEX used > 2,000 air districts. See Appendix 5B for
details.
11 http://www.ncdc.noaa.gov/oa/climate/surfaceinventories.html
5-11
-------
U.S. children ages 3 to 19 who attend school were younger than five years old (2000 Census
Summary File 3, Table QT-P19: School Enrollment). Currently we do not estimate exposure for
these younger children due to a lack of information that would let us confidently characterize
these younger aged children. While EPA guidance recommends, for certain instances, an upper
age group of children ages 16 through 21 (U.S. EPA, 2005), we restricted our upper age
classification of children through age 18. In considering the expected variability in activity
patterns over the span of ages 16 through 21 (e.g., time spent outdoors, time in school, each in
contrast to time spent working) and the relatively small difference in respiratory physiology over
that same age span compared with that of adults (e.g., Figure 5-22), factors critical for high Os
exposure and dose, we assumed simulated people age 19 to 21 would be best included in our
adult study group. The number of people represented in each of the 15 study areas is given in
Table 5-1 and, considering all study areas together, captures approximately 32.8 % of all
children ages 5 to!8 and 32.0 % of the total U.S. population ages 5 to 95.
The number of asthmatic school-age children and asthmatic adults in each study area was
estimated using asthma prevalence from the Center for Disease Control (CDC) and Prevention's
National Health Interview Survey (NHIS).12 Briefly, years 2006-2010 NHIS survey data were
combined to calculate asthma prevalence, defined as the probability of a "Yes" response to the
question: "do you still have asthma?" among those that responded "Yes" to the question "has a
doctor ever diagnosed you with asthma?". The asthma prevalence was first stratified by NHIS
defined regions (Midwest, Northeast, South, and West), sex, age (single years for ages 0-17) or
age groups (ages > 18), and by a family income/poverty ratio.13 These new asthma prevalence
estimates were then linked to U.S. census tract level poverty ratio probabilities (U.S. Census
Bureau, 2007), also stratified by age and age groups, to generate a final database consisting of
census tract level asthma prevalence for the entire U.S. A detailed description of how the data
base was developed is presented in Appendix 5C, while the estimated asthma prevalence used
for each census tract is provided in the APEX asthma prevalence input file. A summary of the
asthma prevalence approximated for each urban study area, as weighted by the age, sex, and
income/poverty level distributions comprising each urban study area, is provided here in Table
5-2.
All simulated individuals (either asthmatic or non-asthmatic) used time-location-activity
data from CHAD, the most complete, high quality source of human activity data for use in
exposure modeling. The current CHAD database contains over 53,000 individual daily diaries
including time-location-activity patterns for individuals of both sexes across a wide range of ages
12 See http://www.cdc.gov/nchs/nhis.htm (accessed October 4, 2011).
13 The income/poverty ratio threshold used was 1.5, that is the surveyed person's family income was considered
either < or > than a factor of 1.5 of the U.S. Census estimate of poverty level for the given year.
5-12
-------
(<1 to 94). The database is geographically diverse, containing activity pattern data from
individuals residing in several major cities, suburban, and rural areas across the U.S. Time spent
performing activities within particular locations can be on a minute-by minute basis, thus
avoiding the smoothing of potential peak exposures longer event durations would yield.
Table 5-3 summarizes the studies and number of diaries in CHAD used in this
assessment, noting that the total CHAD diaries used by APEX is restricted to just over 41,000
given our simulation age range (5-95) and additionally selected usability requirements.14
Additional context regarding the representativeness of the CHAD data in estimating exposure is
provided in section 5.3.1 and Appendix 5G.
Table 5-2. Asthma Prevalence for Children and Adults Estimated by APEX in Each
Simulated Study Area.
Study Area
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington D.C.
All Areas
Asthma Prevalence (%)
Children (5-1 8)
9.6
9.7
11.4
10.7
10.9
9.9
8.9
11.1
10.1
9.0
12.2
11.3
9.0
11
9.5
10.5
Adults (18-95)
6.5
6.6
7.9
7.8
7.7
6.5
7.7
7.7
6.5
7.7
8.1
7.9
7.8
7.6
6.4
7.6
All People (5-95)
7.2
7.3
8.6
8.4
8.4
7.3
7.9
8.5
7.4
8.0
9.0
8.7
8.1
8.4
7.1
8.2
14 In this assessment, the CHAD diaries must be from persons having a known age, sex, day-of-week, and daily
temperature. In addition, diaries must have no more than 3 hours total of missing location and/or activity data.
5-13
-------
Table 5-3. Consolidated Human Activity Database (CHAD) Study Information and Diary-
days Used by APEX.
Activity Pattern Study
(Abbrev.)
Baltimore Retirement Home
(BAL)
California Youth (CAY)
California Adults (CAA)
California Children (CAC)
Cincinnati (CIN)
Detroit Exposure and
Aerosol Research (DEA)1 2
Denver CO Personal
Exposure (DEN)
EPA Longitudinal (EPA)1 2
LA Ozone Exposure:
Elementary School (LAE)
LA Ozone Exposure: High
School (LAH)
National Human Activity
Pattern Study: Air (NHA)
National Human Activity
Pattern Study: Water
(NHW)
National-Scale Activity
Survey (NSA)
Population Study of Income
Dynamics I (ISR)1
Population Study of Income
Dynamics II (ISR)1
Population Study of Income
Dynamics III (ISR)1 2
RTI Ozone Averting
Behavior (OAB)1
RTP Panel (RTP)1
Seattle (SEA)1
Study of Use of Products
and Exposure Related
Behavior (SUP) 12
Washington D.C. (WAS)
Totals
General
Study Area
Baltimore,
MD
California
California
California
Cincinnati,
OH
Detroit, Ml
Denver, CO
RTP, NC
Los Angeles,
CA
Los Angeles,
CA
National
National
7 US metro.
areas
National
National
National
35 US
metro, areas
RTP, NC
Seattle, WA
Sac/San
Fran, CA
Counties
Wash., DC
Study
Years
1997-98
1987-88
1987-88
1989-90
1985
2005-06
1982-83
1999-2000,
2002, 06-08
1989
1990
1992-94
1992-94
2009
1997
2002-03
2007-08
2002-03
2000-01
1999-2002
2006-10
1982-83
1982-2010
Subject
Ages
72-93
12-17
18-94
<1 -11
<1 -86
18-74
18-70
<1 -60
10-12
13-17
<1 -93
<1 -93
35-92
<1 -13
5-19
10-19
2- 12
55-85
6-91
1 -88
18-71
<1 -94
Subjects
(ages 4-94)
26
182
1,555
771
807
74
410
10
17
19
4,129
4,099
1,192
1,904
2,469
1,351
575
37
178
696
686
21,187
Diary-days
(ages 4-94)
304
182
1,555
771
2,259
331
714
1,386
50
42
4,129
4,099
6,825
3,507
4,800
2,619
2,187
871
1,624
2,533
686
41,474
Diary-days
(ages 4-1 8)
0
182
36
771
727
5
7
0
50
42
693
745
0
3,507
4,793
2614
2,187
0
317
994
10
17,680
Study data added after 2007 O3
: Study data added after 2012 1st
NAAQS review.
Draft O3 HREA.
5-14
-------
APEX creates a sequence of daily diaries across the entire Os season for each simulated
individual using a method designed to capture the tendency of individuals to repeat activities,
based on reproducing realistic variation in a key diary variable (Glen et al., 2008). For this Cb
analysis, the key variable selected is the amount of time an individual spends outdoors each day,
one of the most important determinants of exposure to high levels of Os (see section 5.3.2). The
longitudinal method targets two statistics, a population diversity statistic (D) and a within-person
autocorrelation statistic (A). Values of 0.2 for D and 0.2 for^4 were initially developed based on
analyses by Geyh et al. (2000) and Xue et al. (2004), with both studies evaluating groups of
children ages 7 to 12 in a single study area. We adjusted values for D upwards to 0.5 to reflect a
broader range of ages and to better estimate repeated activities.15 Further details regarding the
development of the longitudinal methodology can be found in U.S. EPA (2012a, b).
5.2.6 Key Physiological Processes and Personal Attributes Modeled
The modeling of physiological processes relevant to the Os exposure and intake is
complex, particularly when representing inter- and intra-personal variability in energy
expenditure (EE) and ventilation rates (VE). APEX has a module capable of estimating several
variables associated with every activity performed by simulated individuals. Briefly, the module
links the diary indicated activities to specific energy expended, the rate of oxygen consumed
(VCh) and the associated ventilation rate, all considering the unique sequence of events
individuals go through each simulated day. The activity-specific time-series of VE estimates
ultimately serve as an important variable used in estimating Os intake as well as in identifying
when simulated individuals performing activities at moderate or greater exertion. In addition,
age, sex, and body mass related physiological differences are specifically taken into account by
the ventilation algorithm, derived using ventilation data obtained from several human studies
(see Graham and McCurdy, 2005):
\n(VEIBM )=b0 +£>! \n(V 02! BM ) + b2 \n(\ + age) + b3 sex + eb+ew Equation (5-2)
where,
In = natural logarithm of variable
•
VE/ BM = activity specific ventilation rate, body mass normalized (L air/kg)
hi = see Table 5-4
15 A small D means that the overall variability between people in the key diary statistic is smaller than the variability
observed over days within the same person. A D closer to 1 means that each person shows little variation over
time relative to the variability between persons.
5-15
-------
V'oil BM = activity specific oxygen consumption rate, body mass normalized (L
02/kg)
age = age of the individual (years)
sex = sex (-1 for males, +1 for females)
eb = randomly sampled error term for between-person variability N{0, se},
(L air/kg)
ew = randomly sampled error term for within-person variability N{0, se}, (L
air/kg)
As indicated by Equation (5-2), the random error (e) is allocated to two variance
components used to estimate the between-person (inter-individual variability) residuals
distribution (eb) and within-person (intra-individual variability) residuals distribution (ew~). The
regression parameters bo, hi, b2, and bs are assumed constant over time for all simulated
individuals, eb is sampled once per person by APEX, whereas ew varies from event to event.
Point estimates of the regression coefficients and standard errors of the residuals distributions are
given in Table 5-4. See Appendix 5A, Isaacs et al. (2008), and Chapter 7 of the APEX TSD (US
EPA, 2012b) for further discussion of this module. See also section 5.4.4 for a limited
performance evaluation of this module in estimating ventilation rates.
Table 5-4. Ventilation Equation Coefficient Estimates (bj) and Residuals Distributions (e\).
Age
group
<20
20-<34
34-<61
en-
Ventilation Equation Coefficients1
bo
4.3675
3.7603
3.2440
2.5826
bi
1.0751
1.2491
1.1464
1.0840
b2
-0.2714
0.1416
0.1856
0.2766
b3
0.0479
0.0533
0.0380
-0.0208
Random Error1
6b
0.0955
0.1217
0.1260
0.1064
ew
0.1117
0.1296
0.1152
0.0676
1 These are values of the coefficients and residuals distributions described by Equation (5-2) and described
in Graham and McCurdy (2005).
Two key personal attributes determined for each simulated individual in this assessment
are body mass (BM) and body surface area (BSA). Each simulated individual's body mass is
randomly sampled from age- and sex-specific body mass distributions generated from National
5-16
-------
Health and Nutrition Examination Survey (NHANES) data for the years 1999-2004.16 Details in
their development and the parameter values are provided by Isaacs and Smith (2005). Then age-
and sex-specific body surface area can be estimated for each simulated individual based on
logarithmic relationships developed by Burmaster (1998) using body mass as an independent
variable as follows:
BSA=e-2'2/«i BM^Zi Equation (5-3)
5.2.7 Microenvironments Modeled
APEX is designed to estimate human exposure by using algorithms that attempt to
capture the full range of Os concentrations expected within several microenvironments. Broadly
aggregated, these can be either indoor, inside a motor vehicle, near road, or outdoor locations.
The two methods available in APEX for calculating pollutant concentrations within
microenvironments are a mass balance model and a transfer factor approach. Table 5-5 lists the
28 microenvironments selected for this analysis and the exposure calculation method used for
each. Additional information on the linking of CHAD codes with APEX microenvironments can
be found in Appendix 5B, section 5B-6. Further, information on what microenvironments were
included as a location for each CHAD study is provided in Appendix 5B, Table 5B-2.
The importance of modeling indoor microenvironments (e.g., homes, offices, schools) is
underscored by research indicating that personal exposure measurements of Os may not be well-
correlated with ambient measurements and indoor concentrations are usually much lower than
ambient concentrations (U.S. EPA, 2013, Section 4.3.3). We used mass balance modeling to
estimate Os concentrations in all indoor microenvironments, considering probabilistic
distributions of temperature-dependent (where data were available) building air exchange and
chemical decay rates. Parameter settings for each of these variables are provided in Appendix
5B, while additional discussion regarding updates made to air exchange rates using more recent
study data is given in Appendix 5E.
The remaining microenvironments were modeled using a transfer factors approach.
Outdoor microenvironmental concentrations were assumed equivalent to ambient concentrations,
near-road concentrations were adjusted considering whether or not Os concentrations were
reduced by atmospheric reactions (e.g., scavenging by NOX) or other processes, and vehicular
microenvironments considered both the outdoor concentration attenuation and
16 Demographic (Demo) and body measurement (BMX) datasets for each of the NHANES studies were obtained
from http://www.cdc.gov/nchs/nhanes/nhanes_questionnaires. htm.
5-17
-------
infiltration/removal in the concentration estimation. Specific parameter settings for each of these
variables are provided in Appendix 5B.
Table 5-5. Microenvironments Modeled, Calculation Method Used, and Variables
Included.
Microenvironment
Indoor: Residence, Community Center or Auditorium, Restaurant,
Hotel/Motel, Office building/Bank/Post Office, Bar/Night Club/Cafe, School,
Shopping Mall/Non-Grocery Store, Grocery Store/Convenience Store,
Metro-Subway-Train Station, Hospital/Medical/Care Facility, Industrial
Factory/Warehouse, Other Indoor
Outdoor: Residential, Park/Golf Course, Restaurant/Cafe, School
Grounds, Boat, Other Outdoor Non-Residential
Near-road: Metro-Subway-Train Stop, Within 10 Yards of street, Parking
Garage (covered or below ground), Parking lot (open)/Street parking,
Service Station
Vehicle: Cars/Light Duty Trucks, Heavy Duty Trucks, Bus, Train/Subway
Calculation
Method
Mass
balance
Factors
Factors
Factors
Variables
Building air
exchange &
chemical
decay rates
None
Proximity
factors
Proximity &
penetration
factors
5.2.8 Model Output
APEX estimates the complete time series of exposure concentrations for every simulated
individual and can summarize data using standardized time metrics (e.g., hourly or daily average,
daily maximum 8-hr average) or can output the minute-by-minute exposure concentrations (as is
needed for the risk estimation in Chapter 6). As an indicator of exposure to Os air pollution, we
selected the daily maximum 8-hr average Os exposure17 for every simulated individual and
stratified these exposures by exertion level at the time of exposure. This indicator was selected
based on controlled human exposure studies where reported adverse health responses were
associated with exposure to Os and while the study subject was exercising.18 Factors important in
calculating this indicator includes the magnitude, duration, frequency of exposures, and the
breathing rate of individuals at the time of exposure. As a reminder, the calculated daily
maximum 8-hr average exposure concentrations are distinct from that of daily maximum 8-hr
average ambient concentrations by accounting for simulated individual's time-location-activity
patterns and Os concentration decay/variation occurring within the occupied microenvironments.
17 It is important to stress here that only the maximum 8-hr average Os exposure concentration is retained for each
day simulated, per person. While every day has twenty-four unique 8-hr averages and that it is entirely possible
multiple benchmark exceedances could occur for an individual on certain high ambient Os concentration days,
staff judge this is not a practical output for the purposes of this assessment.
18 It is worth noting that the adverse health responses in the human clinical studies are generally based on 6.6 hour
exposure to O3. Therefore, it is possible that the effects observed at benchmark levels identified using a 6.6 hour
exposure could correspond to lower 8-hr exposures.
5-18
-------
Benchmark levels used in this assessment include 8-hr average Os exposure
concentrations of 60, 70 and 80 ppb; the same benchmark levels used for the 2007 Os exposure
assessment (U.S. EPA, 2007b). Estimating exposures to ambient Cb concentrations at and above
these benchmark levels is intended to provide perspective on the public health impacts of Cb-
related health effects observed in human clinical and toxicological studies, but that cannot
currently be evaluated in quantitative risk assessments (e.g., lung inflammation, increased airway
responsiveness, and decreased resistance to infection). The 80 ppb benchmark level represents an
exposure level where there is substantial clinical evidence demonstrating a range of Os-related
effects including lung inflammation and airway responsiveness in healthy adults. The 70 ppb
benchmark level reflects evidence that asthmatics have larger and more serious effects than
healthy people as well as a substantial epidemiological evidence of adverse effects associated
with Os levels that extend below 80 ppb. The 60 ppb benchmark level represents the lowest
exposure level at which Os-related effects have been observed in clinical studies of healthy
individuals. See ISA section 6.2.1 for further discussions regarding the body of evidence
supporting the selection of these benchmark levels.
The level of exertion of achieved by all simulated individuals is approximated by an
equivalent ventilation rate (EVR), that is, a time-averaged ventilation rate normalized by body
surface area (BSA, in m2) and is calculated as VE/BSA, where VE is the ventilation rate in
liters/minute. For identifying moderate or greater exertion occurring during any 8-hr average
exposure period in this assessment, we used the lower bound EVR value of 13 (liters/min-m2)
based on a range of EVRs used by Whitfield et al. (1996) to categorize people engaged in
moderate exertion activities for an 8-hr period. Whitfield et al. (1996) developed this range from
EVR data reported in a 6.6-hr controlled human exposure study conducted by McDonnell et al.
(1991), whereas, adult study participants were required to maintain a particular ventilation rate
(generally between 35-45 L/minute) while exposed over the duration of the experiment. In
APEX, EVR is calculated using an 8-hr averaging time as is done in calculating 8-hr exposures,
though of interest in this exposure assessment are instances when the simulated individual
experiences an 8-hr average exposure at or above the selected benchmark levels that
simultaneously occurs while at a moderate or greater exertion level (i.e., 8-hr average EVR > 13
L/min-m2).
APEX then calculates two general types of exposure estimates for the population of
interest: the estimated number of people exposed to a specified Os concentration level and, the
number of days per Os season that they are so exposed; the latter metric is expressed in terms of
person-days. The former highlights the number of individuals exposed one or more times per Os
season at or above a selected benchmark level. The person-days measure estimates the number of
times per season the simulated individuals are exposed at or above a selected benchmark level
5-19
-------
and summed across individuals comprising the population. We note that a person-days metric
conflates people and occurrences: one occurrence for each of 10 people would be counted the
same as 10 occurrences for one person (i.e., 10 person-days at or above benchmark level). In this
assessment we are more interested in reporting multiday exposures rather than total person-days,
that is, the number of times an individual experiences multiple exposures at or above a
benchmark level during an Os season. Given the complexities of the exposure modeling, the four
study groups considered, the 15 study areas, the 5 years of ambient air quality, the multiple air
quality scenarios simulated, and ultimately the output data generated, including both single and
multiday exposures for simulated individuals, the consolidation of the results and the related
graphic depictions used in this assessment requires additional discussion.
To begin, a simple example of exposure results is the estimated percent of asthmatic
school-age children experiencing exposures at or above a single 8-hr benchmark level when
considering air quality adjusted to just meet the existing 8-hr Cb standard (4th highest daily
maximum 8-hr average Os concentration averaged over a three year period) in Atlanta (Figure 5-
2). This presentation largely depicts the variability in Os exposure across Atlanta during the five
years of air quality evaluated (2006-2010, given by the x-axis) for each of the benchmark levels
of interest. A general finding regarding temporal variability extracted from this graph would be
that fewer asthmatic school-age children exceed daily maximum 8-hr average exposures of 60
ppb (red symbol/lines) considering 2009 air quality adjusted to just meet the existing standard
(the top panel heading of "75") when compared with other simulation years (approximately 12%
of asthmatic school-age children experienced such an exposure compared with upwards to 20%).
Also worthy of mention is the two exposure results reported for year 2008 considering the 60
ppb exposure benchmark of approximately 6% and 20%. Two values were generated for 2008
due to the two 3-year averaging periods used to simulate just meeting the air quality standards
(i.e., 2006-2008 and 2008-2010, respectively). The percent of study subjects exposed to higher
benchmark levels is also presented in similar fashion (e.g., the percent of people experiencing 8-
hr exposures at or above 70 ppb is illustrated with a green symbol/line), though by definition,
will always be at or below that estimated for the percent of people exceeding the 60 ppb
benchmark in a given year. For example, only 5% of asthmatic school-age children are
estimated to experience at least one daily maximum 8-hr exposure at or above 70 ppb
considering 2006 air quality, while about 20% are estimated to exceed the 60 ppb benchmark
level for that same year. Further, exposures at or above the highest exposure benchmark
considered in this exposure assessment are indicated by a third colored symbol and continuous
line (80 ppb, colored blue), and again by definition, the percent of study subjects exposed would
be at or below that estimated for the 70 ppb exposure benchmark.
5-20
-------
Percent of
Study Group
Exposed in
Urban Area
Air Quality
Standard
Level (ppb)
Results for
a 3-year
Averaging
Period
Exposure
Benchmark
(ppb)
60
Individual
Years of
Air Quality
Figure 5-2. Percent of Asthmatic School-age Children in Atlanta with at least One Daily
Maximum 8-hr Average Os Exposure while at Moderate or Greater Exertion, 2006-2010
Air Quality Adjusted to Just Meet the Existing Standard. This single panel display illustrates
the exposure results by year (x-axes) including the two standard averaging periods (the sets of
continuous lines each for 2006-2008 and 2008-2010), the three exposure benchmarks (60, 70, 80
ppb - red, green, and blue lines) considering the adjusted air quality standard level (75 ppb,
labelled top shaded area) in Atlanta (labelled right shaded area).
Mindful of the further complexity when considering the four air quality standard
scenarios (existing standard and three alternative standard levels) and 15 urban study areas, we
elected to use a multi-panel graphing approach to succinctly summarize the exposure output data,
while also retaining as much information as possible in a single page format to allow for visual
analysis of trends and patterns. Figure 5-3 exhibits all of the dimensions of the exposure results
mentioned above (i.e., year, benchmark level, and study area) along with distinguishing between
the existing (75 ppb) and alternative standard levels (60, 65, 70 ppb). We elected to use the
multi-panel display, going from left to right, to illustrate the impact to estimated exposures
resulting from progressively increasing the stringency of the air quality standard. For example,
the left most panel heading of "75" contains the exposure results when air quality was adjusted to
just meet the existing 8-hr standard level of 75 ppb, i.e., the same results presented in Figure 5-2
and described above. Also included in going to the right of this panel in Figure 5-3 are the
exposure results considering a 70 ppb standard (panel heading of "70"), a 65 ppb standard (panel
heading of "65"), and a 60 ppb standard (panel heading of "60").
5-21
-------
65 ppb Air
Quality
Standard
60 ppb Air
Quality
Standard
75ppbAir
Quality
Standard
70 ppbAir
Quality
Standard
2, >4, and >6
per Os season), considering the 2006-2010 air quality adjusted to just meet the existing and
alternative standards levels. When considering 2006 air quality adjusted to just meet the existing
standard, approximately 10% of asthmatic school-age children experienced at least two days
where their daily maximum 8-hr average exposure was at or above 60 ppb, though fewer than
5-22
-------
5% experienced such exposures in 2009. When collectively considering all simulated air quality
scenarios and years, fewer than 3% of asthmatic school-age children experienced at least four
exposures at or above 60 ppb and virtually no asthmatic school-age children experienced six or
more such exposures over the Os season.
Percent of Study
Group Exposed to
Selected Exposure
Benchmark
Number of Exceedances
of Selected Exposure
1 Benchmark per O3 Season
Figure 5-4. Percent of Asthmatic School-age Children in Atlanta with Multiple (> 2, > 4, >
6 per Os season - red, green, and blue lines) Daily Maximum 8-hr Average Os Exposures at
or above 60 ppb while at Moderate or Greater Exertion, 2006-2010 Air Quality (x-axes)
Adjusted to Just Meet 8-hr Os Standard Levels of 75, 70, 65, 60 ppb (panels left to right).
Also worth discussing is the appearance of a similar pattern between the benchmark level
results (Figure 5-3) and the number of exceedances of a single benchmark (Figure 5-4). Because
the ambient concentration is an important determinant in exposure concentrations, it is not
surprising to see that the trend over years for people having at least one exposure at or above a
particular benchmark level (e.g., 60 ppb) is similar to those experiencing at least two daily
maximum 8-hr average exposures above 60 ppb (though a smaller percentage of the exposure
study group). This is because years having the highest peak concentrations will yield the greatest
percent of people above benchmark levels, and when one year has a day with the highest
concentration, it is likely that year also has a second day with a similarly and relatively high
concentration, and so on. Using the same logic, one might also conclude that there could be a
pattern between the percent of people experiencing a single daily maximum 8-hr exposure of 70
ppb and multiple exceedances (e.g., four) of 60 ppb also driven by the overall ambient
concentration distribution. However, given that very few people experience multi-day
benchmark exceedances, determining the relationship between the two (if present) may not be of
practical significance. For brevity, the complete multiday exposure results for all APEX
simulations are presented in Appendix 5F, with results presented for one study group (e.g., all
school-age children) in the main body of this FtREA.
5-23
-------
5.3 EXPOSURE ASSESSMENT RESULTS
5.3.1 Overview
The results of the exposure analysis are presented as a series of figures focusing on the
defined range of benchmark levels (i.e., people experiencing daily maximum 8-hr average Os
exposure concentrations > 60, 70, and 80 ppb), noted as being of particular health concern
(Section 5.2.8). A range of concentrations in the air quality data over the five year period (2006-
2010) were used in the exposure model, providing a range of estimated exposures output by
APEX. The adjusted air quality was developed using two distinct 3-year period design values
(2006-2008 and 2008-2010), as described in Chapter 4.19 Exposures were estimated for four
study groups of interest (i.e., all school-age children (5-18), asthmatic school-age children,
asthmatic adults (19-95), and older adults (65-95)) in each of the 15 study areas.
In this exposure assessment, we are primarily interested in Cb exposures associated with
the ambient air quality adjusted to just meet the existing and potential alternative Os standards.
Thus, most of the exposure results presented and discussed are for where ambient air quality was
adjusted to just meet these particular scenarios. While understanding exposures and health risks
associated with historical and existing air quality is important, the primary goal of this and any
HREA is to evaluate to what extent the existing NAAQS, and its associated air quality, protect
health and to what extent alternative NAAQS protect health. Exposure results associated with
recent (base) air quality are briefly discussed here first, though largely reported in Appendix 5F.
5.3.2 Exposure Modeling Results for Base Air Quality
The exposure results for the base (unadjusted) air quality are distinguished from the other
air quality scenario results primarily due to the wide ranging variability in estimated exposures
across the study areas and years. The variability in exposures are the result of the wide ranging
variability in ambient concentration levels, with some base year air quality in some study areas
exhibiting air quality at or near that just meeting the existing 8-hr standard, while other study
areas and years exhibiting base air quality levels much higher than the existing 8-hr standard.
These exposures are informative in describing the existing or recent health risks associated with
a unique air quality scenario, but because they variably diverge from a set concentration level of
interest (such as the existing 8-hr standard), they are of limited relevance in evaluating the
adequacy of either the existing NAAQS as well as potential alternative air quality standards.
That said, detailed tabular and graphic presentations of exposure results associated with the base
19 Thus, the year 2008 will have two sets of estimated exposures, one from each of the two sets of design values.
Where mean results are presented in subsequent results sections, the two values given for year 2008 were first
averaged to give a single exposure value for 2008 before averaging across all years.
5-24
-------
air quality (years 2006-2010) are provided in Appendix 5F, with only key findings summarized
in the following discussion.
Consistent with observations regarding year-to-year variability in ambient concentrations
(Chapter 4), most study areas have the greatest percent of all school-age children experiencing
concentrations at or above the three benchmark levels during 2006 or 2007 along with having the
lowest percent of all school-age children exposed during 2009 considering the base air quality. In
general, between 20-40% of all school-age children experience at least one daily maximum 8-hr
Os exposure > 60 ppb, 10-20% experience at least one Os exposure > 70 ppb, and 0-10%
experience at least one Os exposure > 80 ppb, all while at moderate or greater exertion (i.e., an 8-
hr EVR > 13 L/min-m2) and. Year-to-year variability and percent of study subjects exposed were
similar for all school-age children and asthmatic school-age children, largely a function of
having both simulated study groups using an identical activity diary pool to construct each
simulated individual's time series of activities performed and locations visited.20
The overall year-to-year pattern of exposure for asthmatic adults is similar to that
observed for all school-age children, though the percent of the asthmatic adult study group
exposed is lower by a factor of about three or more. Having a lower percent of asthmatic adults
exposed is expected given that outdoor time expenditure is an important determinant of Os
exposure (section 5.4.2) and that adults spend less time outdoors than children (section 5.4.1), as
well as adults having lower participation in outdoor events. The percent of all older adults
experiencing exposures at or above the selected benchmark levels is lower by a fewer percentage
points when compared with the results for asthmatic adults. Again, older adults, on average,
would tend to spend less time outdoors and do so with less frequency when compared with both
adults and children (section 5.4.1), in addition to fewer older adults performing activities at
moderate or greater exertion for extended periods of time, thus leading to fewer people exposed
to Os concentrations of concern.
The year-to-year patterns of the single and multiple exposure occurrences considering
base air quality (2006-2010) were similar among the four exposure study groups, therefore only
results for all school-age children will be summarized here. Depending on the year and study
area, about 10-25% of all school-age children could experience at least two exposures above the
60 ppb benchmark during the Os season, while about 5-10% school-age children could
experience at least four. Most study areas and years are estimated to have fewer than 5% of all
school-age children experience six or more daily maximum 8-hr exposures > 60 ppb considering
the base air quality. When considering the multi-day exposures for all school-age children at or
20 Using the same diary pool to simulate asthmatic and non-asthmatic children's activity patterns is supported by
similarities in time spent outdoors and exertion levels (HREA section 5.4.1.5).
5-25
-------
above the 70 ppb benchmark, about 2-10% of all school-age children could experience at least
two exposures during the Os season, while experiencing four or more exposures were generally
limited to fewer than 4% of all school-age children. Almost half of the study area-year
combinations had no school-age children experiencing two or more exposures > 80 ppb
benchmark, with the other half estimated to have about 1% of all school-age children
experiencing two or more exposures at or above the 80 ppb benchmark. School-age children
having four or more 80 ppb benchmark exceedances were limited to only a few study area years
and, where a non-zero value was estimated, were limited to < 0.5% of the study group.
5.3.3 Exposure Modeling Results for Simulations of Just Meeting the Existing and
Alternative 8-hour Ozone Standards
In this section, we present the exposures estimated when considering the air quality
adjusted to just meeting the existing Os NAAQS standard, as well as when considering potential
alternative 8-hr standard levels (55, 60, 65, 70 ppb) of the existing standard. Comprehensive
multi-panel displays of exposure results are presented for each of the study groups of interest,
i.e., all school-age children (5-18), asthmatic school-age children, asthmatic adults (19-95), and
all older adults (ages 65-95; Figure 5-5 to Figure 5-9, respectively). Included in each display are
the three exposure benchmarks (60, 70, and 80 ppb), the five years of air quality (2006-2010), for
the 15 urban study areas. A single multi-panel display is used to present the results for each of
the four study groups, beginning with the estimated percent of people exposed at least one time
at or above the selected benchmark levels. Modeled exposures in the 15 study areas and
considering each benchmark level are presented on the same scale to allow for direct
comparisons across the multi-panel display. The most notable patterns in the exposure results are
described here using one study group (i.e., all school-age children), as there is a general
consistency in the year-to-year variability within each study area across all four study groups.
Any deviation from the observed pattern will be discussed for the subsequent study group.
We note that after adjusting to just meet a potential 8-hr ambient standard level of 55
ppb, there were nearly no people exposed at or above any of the selected benchmark levels, thus
these data, while modeled, are not presented in detail here. In addition, in one study area
(Chicago), Os ambient monitor design values were below that of the existing standard during the
2008-2010, therefore APEX simulations could not be performed for meeting the existing
standard for that 3-year period. And finally, we were not able to simulate just meeting an 8-hr
standard level of 60 ppb or below in the New York study area (see Chapter 4 for details), thus
APEX simulations for these air quality scenarios could not be performed in New York.
Figure 5-5 illustrates the exposures estimated for all school-age children in each study
area with general observations as follows. After adjusting air quality to just meet the existing and
5-26
-------
alternative standards, there are virtually no school-age children experiencing a daily maximum 8-
hr exposure > 80 ppb, with very few school-age children exposed at or above the 70 ppb
benchmark. For example, out of 87 possible study area and year combinations considering air
quality adjusted to just meet the existing standard (the least stringent standard level considered
here), only 29 resulted in > 0.1% estimated percent of all school-age children exposed at least
once at or above the 80 ppb benchmark with the maximum percent of all school-age children
exposed estimated for St. Louis (1.1%). Ninety-four percent of study area and year combinations
had fewer than 5% of all school-age children experiencing at least one daily maximum 8-hr
average exposure at or above 70 ppb considering ambient air quality adjusted to just meeting the
existing standard, again with a maximum of 8.1% occurring in St. Louis. When considering air
quality adjusted to just meet an 8-hr ambient standard level of 70 ppb, < 0.2% of all school-age
children experience at least one daily maximum 8-hr exposure at or above 80 ppb for all study
area and year combinations, while for 76 or 90 study area and year combinations, < 1% of all
school-age children exceed the 70 ppb exposure benchmark. This pattern of having very few
school-age children experiencing daily maximum 8-hr exposures > 70 and 80 ppb is as expected
given the nature of the air quality adjustment procedure that limits 8-hr ambient concentrations at
or above the selected potential alternative standard level.
In contrast, approximately 10-20% percent of all school-age children experience at least
one daily maximum 8-hr exposure > 60 ppb when considering air quality just meeting the
existing standard (Figure 5-5). And similar to that mentioned above regarding exposures
associated with the base air quality, a general year-to-year exposure pattern emerges with respect
to study area and year. For the Northeastern (Boston, New York), Mid-Atlantic (Philadelphia,
Washington DC, Cleveland) and Mid-Western (Chicago, Detroit, and St. Louis) study areas, the
maximum percent of all school-age children exposed generally occurs during year 2007. For the
Southern (Atlanta, Dallas, Houston) and Western (Denver, Los Angeles, Sacramento) study
areas, the maximum exposure occurs during year 2006. Deviations from this temporal exposure
pattern appear mostly as a result of the standard averaging period, with the 2008-2010 period
producing equal or greater maximum exposures during either 2008, 2010, or both years and most
prevalent in the Northeastern and Mid-Atlantic study areas (Baltimore, Boston, New York,
Philadelphia, Washington DC; note also a trend in Atlanta, Denver, St. Louis).
The exposure patterns observed for the 60 ppb exposure benchmark remain consistent
when considering air quality adjusted to just meet a 70 ppb ambient standard, though the percent
of all school-age children exposed is less than that observed when considering the air quality
adjusted to just meet existing standard. Further, 75 of 90 study area and year combinations are
estimated to have < 10% of all school-age children experience at least one daily maximum 8-hr
exposure > 60 ppb, though between 10-20% of all school-age children were estimated to be
5-27
-------
exposed for a few study area and year combinations (e.g., Atlanta-2006, Chicago-2007 and -
2010, and Houston-2009). When considering air quality adjusted to just meet a 65 ppb standard
level, the percent of all school-age children experiencing at least one daily maximum 8-hr
exposure > 60 ppb diminishes to 5% or less for most study areas and years (i.e., 81 of 90 study
area year combinations).
All of what has been described regarding the estimated exposures to school-age children
(i.e., the year-to-year and benchmark level patterns, and the percent of the study group exposed)
also applies to the exposures estimated for asthmatic school-age children (Figure 5-6). Different
however would be the relative number of asthmatic school-age children exposed in each study
area if compared with all school-age children, as the asthma prevalence rates vary by study area
(Table 5-2), though on average are about 10% of the population of children.
The percent of asthmatic adults (Figure 5-7) experiencing daily maximum 8-hr average
exposures above the selected benchmark levels is sharply lower than that estimated for all
school-age children. For example, only three of a possible 84 study area and year combinations
(Chicago-2007, Houston-2009, and St. Louis-2007) were estimated have > 0.1% of asthmatic
adults experience a daily maximum 8-hr average exposure > 80 ppb, and only six of a possible
84 study area and year combinations were estimated have >1% of asthmatic adults experience an
daily maximum 8-hr average exposure > 70 ppb, all occurring when considering air quality just
meeting the existing standard. No study area or year combination has more than 10% of
asthmatic adults estimated to experience at least one daily maximum 8-hr exposure > 60 ppb
when considering air quality just meeting the existing standard, with 67 of 84 study area and year
combinations estimated to have 5% or less asthmatic adults experiencing such exposures.
When considering air quality adjusted to just meeting an 8-hr standard level of 70 ppb, no
asthmatic adults experience a daily maximum 8-hr average exposure at or above 80 ppb and <
0.6% experience a daily maximum 8-hr average exposure > 70 ppb for any study area or year
combination. Less than 5% of asthmatic adults could experience a daily maximum 8-hr average
exposure at or above 60 ppb when considering air quality adjusted to just meet a standard level
of 70 ppb for 88 or 90 possible study area year combinations, with the maximum percent of adult
asthmatics exposed outside this range occurring in Denver (6.8%-2008) and St. Louis (5.5%-
2007).
Older adults are estimated to have the fewest exposures above the two highest benchmark
levels when considering the adjusted air quality. For example, only two of a possible 84 study
area and year combinations (St. Louis-2007 and Washington DC-2008) were estimated have >
0.1% of asthmatic adults experience a daily maximum 8-hr average exposure > 80 ppb, and only
six of a possible 84 study area and year combinations were estimated have > 1% of asthmatic
adults experience a daily maximum 8-hr average exposure > 70 ppb, all occurring when
5-28
-------
considering air quality just meeting the existing standard (Figure 5-8). Also, exceeding the 60
ppb exposure benchmark appears to be limited to fewer than 5% of all older adults when
considering air quality adjusted to just meet the existing standard and a standard level of 70 ppb,
and occurs in < 2% of all older adults when considering a standard level of 65 ppb.
An example of multi-day exposure results associated with adjusted air quality is provided
in Figure 5-9. The percent of all school-age children estimated to experience multi-day exposures
above benchmark levels during each study area's Os season is largely limited to two air quality
scenarios: the existing standard and air quality adjusted to just meeting an 8-hr standard of 70
ppb. This is because of the small percent of school-age children experiencing even a single
exposure above the lowest benchmark level when considering standard levels at or below 65
ppb. In addition, when experiencing multiple exposures, most school-age children appear to have
at most two days above benchmark levels per Os season, even when considering the lowest
benchmark level of 60 ppb. For example, 81 of 87 possible study area and year combinations
have < 10% of all school-age children experiencing two or more daily maximum 8-hr average
exposures > 60 ppb when considering an ambient standard level of 75 ppb, while 83 of 90
possible study area and year combinations have < 5% of all school-age children experiencing
two or more exposures > 60 ppb when considering an ambient standard of 70 ppb. Increasing the
stringency of the standard to 65 ppb yields 81 of 90 possible study area and year combinations
where < 1% of all school-age children experience two or more exposures > 60 ppb.
Multi-day exposure to the higher exposure benchmarks (either the 70 or 80 ppb) is a rare
occurrence, even when considering the air quality adjusted to the existing Os standard. For
example, there were no school-age children experiencing two or more daily maximum 8-hr
average exposures above 80 ppb in all but one study area year combination and, and when
considering that one study year having a non-zero value (St. Louis-2007), the estimated percent
of all school-age children at or above the exposure benchmark was only 0.1%. Further, 83 of 87
possible study area and year combinations have < 1% of all school-age children experiencing
two or more exposures > 70 ppb, also when considering an ambient standard level of 75 ppb.
An additional presentation of the estimated exposures is provided in Figure 5-10 and
Figure 5-11 to broadly summarize the number of people in each study group that could
experience exposures of concern. As a reminder, Figure 5-5 through Figure 5-9 were designed to
show the variability in estimated exposures across each of the urban study areas, along with the
exhibiting variability across the years of air quality and two standard averaging periods. In each
of these two additional figures, the mean number of people at or above benchmark levels in each
study area was calculated by first averaging the number of people estimated for the two 2008
standard averaging periods, then averaging the number of people exposed above benchmark
levels for the five year period (2006-2010).
5-29
-------
Figure 5-10 shows the mean number of people experiencing at least one daily maximum 8-
hr average Os exposure > 60 ppb per Os season, while at moderate or greater exertion for each of
urban the study areas and considering the adjusted air quality scenarios. On average,
approximately 2.3 million school-age children have at least one daily maximum 8-hr average
exposure > 60 ppb when considering air quality just meeting the existing standard in the 15
urban study areas (Figure 5-10, top left panel). That estimated number school-age children
experiencing exposures of concern is nearly cut in half when considering air quality adjusted to
just meet an 8-hr average ambient standard of 70 ppb, with approximately 1.2 million
experiencing, on average, at least one daily maximum 8-hr exposure > 80 ppb per Os season in
the 15 urban study areas. The mean number of school-age children experiencing at least one
daily maximum 8-hr average exposure > 60 ppb is further reduced to just under 400,000 when
adjusting air quality to just meet an 8-hr average ambient standard of 65 ppb, while
approximately 70,000, on average, are estimated to be exposed at the same level when
considering air quality adjusted to just meet a standard of 60 ppb.
As was shown above regarding the percent of each study group experiencing exposures of
concern, the reduction in exposures with the increasing stringency of the air quality standard
scenarios are similar to that of school-age children, although the number of people within each
of the study groups is less. For example, when considering air quality that just meets the existing
standard in the 15 urban study areas, nearly 250,000 asthmatic children, on average, are
estimated to experience at least one daily maximum 8-hr average Os exposure > 60 ppb per Os
season (Figure 5-10, top right panel) compared with 2.5 million school-age children exposed at
the same benchmark level. Again, fewer people exposed in one study group compared with that
of another in this case is a function of the smaller study group (asthmatic school-age children)
being a fraction of the larger exposed study group (all school-age children) although for other
study groups (e.g., older adults) fewer people exposed is largely a function of less time spent
outdoors rather than originating from a smaller overall population size.
Far fewer people are exposed to the 8-hr benchmark level of 70 ppb when compared with
the 60 ppb benchmark (Figure 5-11). For example, when considering air quality adjusted to just
meet an 8-hr average ambient standard of 70 ppb, about 94,000 school-age children on average
are estimated to experience at least one 8-hour exposure at or above the 70 ppb, while over ten
times as many would experience at least one 8-hour exposure > 60 ppb per Os season. Again,
similar patterns in number of people exposed to the 70 ppb benchmark are present across each of
the exposure study groups with increasing stringency of the ambient standard level, with on
average a few thousand to a low as a few hundred estimated to experience exposures of concern
when adjusting air quality to just meet ambient standard levels of 65 and 60 ppb, respectively.
5-30
-------
0)
§£>
0)
W (0
8. -2.
x"°
UJ <
£ =
•T O
00 *.|3
0) oj
j= X
O LU
4-> i-
(/) 0)
ro +-
c
0) 0)
<
«
*4_ re
o E
0)
c
CD
8—9 8—o 0 a—•
Figure 5-5. Percent of All School-age Children with at Least One Daily Maximum 8-hr
Average Os Exposure at or above 60, 70, and 80 ppb (red, green and blue lines) while at
Moderate or Greater Exertion, 2006-2010 Air Quality Adjusted to Just Meet 8-hr Average
Os Standards of 75, 70, 65, and 60 ppb (panels left to right) in 15 Urban Study Areas.
5-31
-------
.*
o ro
Q-OJ
111 3
*T <
00 -
c 2
O t
+j O)
re HI
g>
5 o
c o
0) OJ
2 2
.- O)
o^
O £
o
o
0)
E
"55
• m
4-1 ro
+- O)
Sm
o o
°-
<^^(^<^(^ *^(^(^(^<^ <^(^<^(^f^ <^<^r^f^<^
Figure 5-6. Percent of Asthmatic School-age Children with at Least One Daily Maximum
8-hr Average Os Exposure at or above 60, 70, and 80 ppb (red, green and blue lines) while
at Moderate or Greater Exertion, 2006-2010 Air Quality Adjusted to Just Meet 8-hr
Average Os Standards of 75, 70, 65, and 60 ppb (panels left to right),15 Urban Study Areas.
5-32
-------
> i i i t a o o o—*—
Figure 5-7. Percent of Asthmatic Adults with at Least One Daily Maximum 8-hr Average
Os Exposure at or above 60, 70, and 80 ppb (red, green and blue lines) while at Moderate
or Greater Exertion, 2006-2010 Air Quality Adjusted to Just Meet 8-hr Average Os
Standards of 75, 70, 65, and 60 ppb (panels left to right) in 15 Urban Study Areas.
5-33
-------
(y^ ry fy* fy fy fy fy fy fy fy fy fy fy fy
-------
Figure 5-9. Percent of All School-age Children with > 2, > 4, > 6 (red, green, and blue lines)
Daily Maximum 8-hr Average Os Exposures at or above 60 ppb while at Moderate or
Greater Exertion, 2006-2010 Air Quality (x-axes) Adjusted to Just Meet 8-hr Average Os
Standards of 75, 70, 65, 60 ppb (panels left to right) in 15 Urban Study Areas.
5-35
-------
2,500,000
c o
-------
25,000
Ill
p ? 1
a o 2
i||
IS"
its
3 £ S
5 re 1 10,000
x ID B
1-s-a
I a
75ppb
70 ppb 65 ppb
Air Quality Standard Level
70 ppb 65 ppb
Air Quality Standard Level
60 ppb
H Atlanta
• Baltimore
y Boston
U Chicago
• Cleveland
H Dallas
H Denver
y Detroit
y Houston
y Los Angeles
y New York
u Philadelphia
_j Sacramento
-JSt. Louis
_j Washington
• Atlanta
• Baltimore
m Boston
U Chicago
H Cleveland
H Dallas
U Denver
U Detroit
y Houston
U Los Angeles
U New York
u Philadelphia
u Sacramento
ust. Louis
u Washington
60 ppb
ra +-.
-O B
S .2
.If
S -o
.
5 v
al
1?
< O
20,000
10,000
5,000
75 ppb
70 ppb 65 ppb
Air Quality Standard Level
60 ppb
70 ppb 65 ppb
Air Quality Standard Level
60 ppb
Figure 5-11. Mean Number of People with at Least One Daily Maximum 8-hr Average Exposure at or above 70 ppb while at
Moderate or Greater Exertion, 2006-2010 Air Quality Adjusted to Just Meet 8-hr Standards of 75, 70, 65, and 60 ppb (x-axes,
from right to left), in 15 Urban Study Areas. All School-age children (top left panel), asthmatic school-age children (top right
panel), asthmatic adults (bottom left panel), older adults (bottom right panel).
5-37
-------
5.4 TARGETED EVALUATION OF EXPOSURE MODEL INPUT AND OUTPUT
DATA
This section summarizes the results of several targeted evaluations intended to provide
additional insights to APEX input data or approaches used to estimate exposures (CHAD data
attributes and activity pattern evaluations, comparison of CHAD outdoor time data with ATUS,
comparisons of asthmatic outdoor time expenditure and exertion levels to that of non-
asthmatics), exposure results for additional exposure populations of interest (outdoor workers,
school-age children during summers, impact of averting), and model performance evaluations
(personal exposure measurements and independent ventilation rate estimates compared with
APEX estimates). Detailed analysis results are provided in Appendix 5G.
5.4.1 Analysis of Time-Location-Activity Data
While CHAD is the most comprehensive and relevant source of time-location-activity
data available for use in our exposure modeling, there are a few limitations to the survey data
contained therein, many of which are founded in the individual studies from which activity
patterns were derived (Graham and McCurdy, 2004). CHAD is a collection of related survey
data, though individual study attributes can range widely (e.g., survey participant ages, region or
city of residence, time-of-year data collected). We note that many of the assumptions about use
of these activity patterns in exposure modeling are strengthened by the manner in which they are
used by APEX. This is done by focusing on selecting the most important individual attributes
that contribute to variability in human behavior (e.g., age, sex, day-of-week, ambient
temperature) and linking these attributes of simulated individuals to the population demographics
of each census tract (see section 5.2.5) and the study area temperatures (section 5.2.4). Further,
one key lifestyle attribute is also accounted for in generating longitudinal diary profiles by
simulating both the intra- and interpersonal variability in time spent outdoors (section 5.2.5; Glen
et al., 2008).
A few questions may arise as to the representativeness of the CHAD diaries to the
simulated population. For example, the year of a particular survey study may differ from our
simulated exposure population by as much as 30 years (i.e., some activity pattern data were
generated in the 1980s). In addition, there are other personal attributes (e.g., ethnicity, income
level, lifestyle factors21), health conditions (e.g., asthma, cardiovascular disease), and situational
factors (e.g., availability of parks and recreation areas) that are not used in creating the simulated
21 Examples of such factors for adults could include married/unmarried, having infants or young children/no
children. Lifestyle factors for children could include whether the child is active/non-active or whether or not there
is time spent outdoors.
5-38
-------
individuals that could be influential in estimating exposures. Considering this, a number of
evaluations were performed to answer questions regarding important personal attributes used in
generating simulated individuals and the general representativeness of the CHAD time-location-
activity data. First though, we summarize the newly acquired activity pattern data now included
in CHAD compared with data available and used in the 1st draft Os HREA.
5.4.1.1 General evaluation of CHAD study data: historical and recently acquired
data
The number of diary days having complete information and used by APEX in the Os
HREA is 41,474 (Table 5-3). This is an increase of about 8,700 diaries currently used by APEX
compared with what was used by APEX in the 1st Draft Os HREA. Further, there have been eight
new study data sets incorporated into CHAD and used in our current exposure assessment since
the previous Os NAAQS review conducted in 2007, most of which were from recently conducted
activity pattern studies (see Appendix 5B, Section 5B-4 for more information regarding these
studies). The diary data included from these new studies have more than doubled the total
activity pattern data used for 2007 Os exposure modeling and has increased the number of
children's diaries by about a factor of five. Currently, the majority of diaries (54%) from CHAD
are taken from surveys conducted in the past decade, while the pre-1990s diaries represent less
than 15% of the total diaries available by APEX.
5.4.1.2 Exposure-relevant personal attributes included in CHAD and APEX
simulated individuals
The survey participants whose diary data are within CHAD were asked a number of
questions regarding their personal attributes. The number and type of attributes present for
diaries in CHAD is driven largely by the original intent of the individual study. In our exposure
assessment, we have strict requirements to simulate individuals using several personal attributes,
namely age, sex, temperature (as a surrogate for seasonal variation in activity patterns), and day-
of-week. These attributes are considered as important drivers influencing daily activity patterns
(Graham and McCurdy, 2004) and when diaries do not have these particular attributes for a
particular day, the diary day will not be used by APEX. We compared the representation of these
and other attributes in the current CHAD used by APEX with that in the 1st draft Os HREA and
found strong similarities in the attribute distributions between both databases, suggesting little
change in the overall composition of the database regarding these influential attributes.
While there may be other personal or situational attributes that affect daily time
expenditure (e.g., socioeconomic status, occupation of an employed person), these attributes are
typically not included in our assessment to generate simulated individuals simply because the
response to the attribute is missing for most of the study participants/CHAD diary days. For
5-39
-------
example, income level is missing for about two-thirds of the CHAD diaries because either the
original study did not have an income/occupation related survey question or perhaps the
participant refused to answer the question if it were posed. If one were to select this personal
attribute in developing a simulated individual's activity pattern (among using any other attribute
having missing responses), the pool of diaries available to simulate individuals may be extremely
limited, likely leading to repetition of diaries used for individuals or groups of similar individuals
and artificially reducing both intra- and inter-personal variability in time expenditure, or perhaps
resulting in model simulation failure altogether. This is why personal attributes are carefully
selected and prioritized according to both their prevalence in CHAD and whether the attribute
has a known significant influence on activity patterns.
5.4.1.3 Evaluation of afternoon time spent outdoors for CHAD survey participants
There have been questions raised regarding the representativeness of the diaries from
studies conducted in the 1980s and whether there are any recognizable patterns in time
expenditure in the CHAD diaries across the time period when data were collected. Because time
spent outdoors is a significant factor influencing daily maximum 8-hr average Os exposures, we
evaluated the current collection of CHAD diaries used by APEX for two metrics and considering
two dimensions: outdoor participation rate (i.e., the percent of people who spent some time
outdoors during their survey day) and the mean time spent outdoors for where the people spent at
least one minute outdoors or at least 2 hours outdoors. Because time spent outdoors is an
important determinant for highly exposed individuals, we summarize the results here for the
diaries having at least 2 hours of outdoor time here, while all other results are provided in
Appendix 5G. CHAD diaries were stratified by five age groups (4-18, 19-34, 35-50, 51-64, 65+)
and three decades (1980s, 1990s, and 2000s) using the year the particular activity pattern study
was conducted. We note that CHAD is composed of primarily cross-sectional data (single diary
days per person), thus the trend evaluated over the three decades is changes (if any) in
participation rate and the time spent outdoors by the composite study population, not within
individuals.
Regardless of decade and duration of time spent outdoors, children tended to have the
highest outdoor participation rate when compared with the other age groups, while the oldest
adults (aged 65 or greater) tend to have the lowest participation rate. The CHAD diaries from the
1980's studies for children ages 4-18 have the highest outdoor participation rate (50%) compared
to other decades (35-40%) and all other age groups and decade of collection. When considering
the pool of diaries available for this age group, these 1980's studies contribute to approximately
19% of diaries having two or more hours of time spent outdoors during the afternoon. This
translates to a small effect on the overall outdoor participation rate for diary pools that would
5-40
-------
include these earlier studies (39% participation rate) compared to the participation rate excluding
these studies (36% participation rate). In general, these outdoor participation rates are similar to
the finding reported recently by Marino et al. (2012) of 37.5%, though estimated for pre-school
age children. Thus, when considering participation in outdoor activities and the
representativeness of the CHAD study data from the 1980s, it is unlikely that use of these oldest
diaries would strongly influence exposure model estimates.
There is variability in the amount of outdoor time evaluated over the three decades, with
diaries from the 2000's studies exhibiting perhaps the lowest range of mean outdoor time (190-
220 min/day) compared with the 1980's (210-240 min/day) and 1990's (212-258 min/day)
studies, a trend perhaps most notable when considering the children's diaries (a decrease in time
spent outdoors of about 30 minutes over the three decades). However, the coefficient of variation
(COV) for each of the age groups and across all decades for the cross-sectional data was
consistently about 40%, supporting a general conclusion of no appreciable differences in the
mean time spent outdoors over the three decades of data collection. Thus, when considering all
diaries having at least 2 hours of afternoon outdoors time and the representativeness of the
CHAD study data from the 1980s compared with that of the other CHAD study years, inclusion
of these earlier diaries is also unlikely to have a strong influence on exposure modeling
outcomes.
5.4.1.4 Evaluation of afternoon time spent outdoor for ATUS survey participants
We evaluated recent year (2002-2011) time expenditure data from the American Time
Use Survey (ATUS) (US BLS, 2012). As was done with the CHAD data set, the purpose was to
evaluate trends (if any) in outdoor time over the period of time data were collected. A few
strengths of the ATUS data are (1) its recent and ongoing data collection efforts, (2) large sample
size (totaling over 120,000 diary days), (3) national representativeness, and (4) that varying diary
approaches would not be an influential or confounding factor in evaluating trends over time.
ATUS does however have a few noteworthy limitations when compared with the CHAD
data: (1) there are no survey participants under 15 years of age, (2) time spent at home locations
is neither distinguished as indoors or outdoors, and (3) missing or unknown location data can
comprise a significant portion of an individual's day (on average, about 40% (George and
McCurdy, 2009)). To overcome the limitation afforded by the ambiguous home location, we
identified particular activity codes most likely to occur outdoors (e.g., participation in a sport) to
better approximate each ATUS individual's outdoor time expenditure. Missing time was
circumvented by our focused analysis: about 85% of missing time information occurs outside of
the hours of interest here (i.e., before 12:00 PM and after 8:00 PM). Data were stratified by the
5-41
-------
same five age groups as was done for the CHAD data, though here the time trends were assessed
over individual survey years.
When considering person-days having at least 2 hours of time spent outdoors, there were
no clear trends over the nine year ATUS study period regarding either the participation rate or
the mean time spent outdoors for any of the age groups. Consistent with CHAD, the participation
rate of children was greater than that of the other age groups. The range in ATUS diary outdoor
participation rate (10-20%) for all age groups is lower than that observed for the CHAD data
(generally between 20-40%), while the range in mean time spent outdoors (190-240 minutes per
day) was similar to that of the CHAD data. The lower participation rate for ATUS participants is
not surprising given the lack of distinction regarding time indoors and outdoors while at home
for ATUS participants and possibly influenced in part by not having any activity patterns for
children under 15 years old. Overall, results of the ATUS data analysis generally support the
representativeness of the CHAD data, and while participation in outdoor activities calculated
using ATUS diaries was less than CHAD diaries, ATUS survey methods obfuscate the strength
of this finding.
5.4.1.5 Evaluation of outdoor time and exertion level for asthmatics and non-
asthmatics in CHAD
Due to limited number of CHAD diaries with survey requested health information, all
CHAD diaries are assumed appropriate for any APEX simulated individual (i.e., whether
asthmatic, non-asthmatic, or no compromising health condition was indicated), provided they
concur with age, sex, temperature, and day-of-week selection criteria. In general, the assumption
of modeling asthmatics similarly to healthy individuals (i.e., using the same time-location-
activity profiles) is supported by the activity analyses reported by van Gent et al. (2007) and
Santuz et al. (1997), though other researchers, for example, Ford et al., (2003), have shown
significantly lower leisure time activity levels in asthmatics when compared with individuals
who have never had asthma. To provide additional support to the assumption that any CHAD
diary day can be used to represent the asthmatic population regardless of the study participants'
characterization of having asthma or not, we first compared participation in afternoon outdoor
activities at elevated exertion levels among asthmatic, non-asthmatic, and unknown health status
using the CHAD diaries. We then compared compatible CHAD diary days with literature
reported outdoor time participation at varying activity levels.
In the first comparison, participation in afternoon outdoor activities for non-asthmatic
children and adults in CHAD were found similar when compared with their respective asthmatic
cohorts (both about 40-50%). Outdoor participation rate for people having unknown asthma
status, a smaller fraction of the total diaries, varied ±10% from that having known asthma status
(children were higher, adults were lower). The amount of time spent outdoors by the people that
5-42
-------
did so varied little across the two study groups and three asthma categories. On average, CHAD
diaries from children indicate approximately 2Vi hours of afternoon time is spent outdoors, 80%
of which is at a moderate or greater exertion level, again regardless of their asthma status, known
or unknown. Slightly less afternoon time is spent outdoors by adults when compared with
children, and while their participation in moderate or greater exertion level activities is much less
(about 63%), there was little difference between asthmatic adults and non-asthmatic adults
considering outdoor time or percent at moderate or greater exertion.
For the second comparison, the percentage of waking hours outdoors at varying activity
levels for asthmatics reported in three independent asthma activity pattern studies (Shamoo et al.,
1994; EPRI, 1988; EPRI 1992) were compared to CHAD diary days having similar personal
attributes and stratified by asthma status. The range in the percent of waking hours outside at
moderate activity level for CHAD diaries was similar to that estimated using the three
independent literature sources (2-10%), however the range in percent of outdoor time associated
with strenuous activities using the CHAD asthmatic diaries extends beyond that of asthmatic
individuals from the three independent studies by about a factor of two higher. At this time, the
definitive reason for this difference is unknown though there are a number of potential
contributing factors. These would include the diary/survey collection methods, how activities
and their associated activity levels were classified, variable number of study subjects, and survey
sample selection methods. Overall, given the above mentioned similarities in outdoor time,
participation, and activity levels, disregarding the survey record asthma status of any CHAD
diary to simulate asthmatic and non-asthmatic individuals in this exposure assessment is
reasonably justified based on the available data analyzed.
5.4.1.6 Evaluation of time spent outdoors by CHAD subjects in four U.S Census
regions
Finally, we performed an additional analysis of outdoor time that evaluates the potential
differences that might exist in CHAD participant activity patterns across geographic regions of
the U.S. It is possible that broad cultural, climatological and other attributes that vary across the
U.S. could influence activity patterns, including time spent outdoors. If such differences in time
spent outdoors were consistent and substantial, assuming the use of activity patterns without
regard to the CHAD participant's original geographic region could lead to bias in urban study
area exposure estimates. The study group of primary interest is school-age children and considers
important influential variables used in selection of diaries to perform exposure simulations such
as temperature and day-of-week. Again, being mindful of potential sample size issues associated
with additional stratification of the CHAD data set, we characterized the U.S into four regions
5-43
-------
(Northeast, South, Midwest, and West) based on the U.S. Census.22 When considering these four
regions alone and the full CHAD data set used (i.e., 41,474 diaries), the number of available
CHAD diaries is large Table 5-6), though note that nearly 3,000 diaries do not have an identified
geographic location and would not be used in an exposure simulation that considered geographic
location.
Table 5-6. Number and Percent of All CHAD Diaries Used by APEX and Stratified by
Four U.S. Regions.
Region
Midwest (MW)
Northeast (NE)
South (SO)
West (WE)
Missing (XX)
Number Of CHAD
Diaries
5257
4323
14951
13998
2945
Percent of
Total Diaries
12.7
10.4
36
33.8
7.1
When considering the important influential attributes used in developing our diary pools
(i.e., age group, sex, temperature, and day of week) in addition to region, the number of diaries
available for simulating individuals is reasonable for most pools, though for some diary pools
there only a few available diaries (Table 5-7; e.g., school-age children in the Midwest and
Northeast, either sex, weekend days, with daily maximum temperature > 84 °F). Again, that
criteria alone (i.e., the number of diaries available) is the justification for not using region as a
diary pool attribute in developing individual activity pattern profiles for the simulated study
groups because, for some of our study areas, APEX would simply not simulate exposures due to
lack of sufficient diaries (e.g., Boston, New York, Detroit). However, it may be of some value to
evaluate whether there are differences in time spent outdoors across these four U.S. regions using
the available data to better inform discussions regarding how this uncertainty might affect the
estimated exposures in this HREA.
As was done in the above activity pattern analyses, we evaluated the distribution of
afternoon time spent outdoors for school-age children spending at least one minute outdoors and
for those spending at least two hours of afternoon time outdoors, though also considering region
as a potential influential factor. Because high Os concentration days would tend to occur more
frequently on days when daily maximum temperatures are also high, we chose to focus our
analysis on the pool comprised of CHAD diaries where daily maximum temperatures were > 84
°F. We evaluated both the participation in the event (i.e., the percent of diaries in diary pool with
afternoon time spent outdoors) and the amount of time per day for those that participated in an
outdoor event.
22https://www.census.gov/geo/maps-data/maps/pdfs/reference/us_regdiv.pdf
5-44
-------
Table 5-7. Number of CHAD Diaries for All School-age Children, Stratified by Four U.S.
Regions and Diary Pool Attributes.
Region
MW
NE
SO
WE
Sex
Female
Male
Female
Male
Female
Male
Female
Male
Daily Maximum
Temperature (°F)
>84
<55
55-83
>84
<55
55-83
>84
<55
55-83
>84
<55
55-83
>84
<55
55-83
>84
<55
55-83
>84
<55
55-83
>84
<55
55-83
Day of
Week
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
WD
WE
Number Of CHAD
Diaries
102
29
475
431
286
267
157
29
443
408
297
251
245
14
269
266
201
132
283
8
254
270
230
157
487
118
305
288
824
783
657
127
347
322
890
822
331
145
231
179
976
726
427
131
354
252
1018
699
5-45
-------
Regarding participation in outdoor activities during afternoon hours, there are general
consistencies across the four regions (Table 5-8). For example, during weekdays with daily
maximum temperatures > 84 °F, diaries collected from females indicate that between 64-69%
spent at least one minute of afternoon time outdoors and between 30-41% spent at least 2 hours.
Under these same conditions, the results for males were also consistent although slightly higher,
having from 67-70% and 37-41% of days participating in outdoor events of at least one minute
and two hours, respectively. Diaries for where region was not determined were not consistent
with these results and indicated greater participation outdoor events, possibly a function of the
smaller sample size, the general ages of study participants (mean age 14 versus about a mean of
8-10 for those having region), or the survey itself (all exclusively from the Cincinnati and
Washington Studies). When considering weekend days, there is greater variability observed in
outdoor event participation, though much of it appears to be a function of the limited number of
diaries for analysis, in particular diaries from school-age children residing in the Northeastern
U.S. (both sexes) and those for Southern school age females.
Table 5-8. Number and Percent of CHAD Diaries of School-age Children Reporting Time
Spent Outdoors on Days When Daily Maximum Temperature > 84 °F, Stratified by Four
U.S. Regions, Sex, and Day Type.
Day of
Week
Weekday
Weekend
Sex
Female
Male
Female
Male
Region
MW
NE
SO
WE
XX
MW
NE
SO
WE
XX
MW
NE
SO
WE
XX
MW
NE
SO
WE
XX
CHAD Diaries with Afternoon Time Spent Outdoors
At Least One Minute
Number
of Diaries
66
167
314
228
26
109
199
439
304
35
17
9
61
105
28
21
6
87
103
34
Participation
(% of Diaries)
65
68
64
69
93
69
70
67
71
100
59
64
52
72
97
72
75
69
79
97
At Least Two Hours
Number of
Diaries
31
101
170
104
11
61
117
263
157
22
9
8
25
55
15
13
3
53
61
23
Participation
(% of Diaries)
30
41
35
31
39
39
41
40
37
63
31
57
21
38
52
45
38
42
47
66
5-46
-------
The amount of time spent outdoors also indicates general consistency across the four U.S.
regions, along with notable deviations possibly a function of the small number of diaries
available for analysis. Based on this analysis, we conclude that there is not a substantial
difference in time spent outdoors across the four regions (and an appropriate number of diaries)
that would warrant performing additional simulations using a region specific diary pool to
estimate exposures in the urban study areas.
Distribution of aftn_-
5
1 300-
J :n>-
p 110-
J 1KI'
60-
0-
X
o
o
0
0
1 .
">• 1
o o
o
0 -i-
— T— O
0
0
V V
L 1 1 -L
' JH ;» M
MW NT SO WE .XX
U.S. Region
])istiihuti,,mif aftn j.iittimr by region
4SO-
420-
1 3m-
5
1 .MO-
i
^ 180-
1 120-
0-
N
•
T <
I
0
1 .
r
.
• T
> T I
1 I-"-!
0 * ^-
, T T T
f.v M!( :s
MW ME SO WE XX
U.S. Rcgim
480-
420-
| 300-
C 240-
| m_
60-
0-
X
Ili-tiTliiniim ..I :. flu minim.' In n-ynri
o o
a °
-
o o <
J O
>
0
! 1
> 0
T T 1 T
: : :' I
1
N « NE SO WK XX
UJS, Region
480-
420-
f
| sw -
1 240-
1 IJO-
60-
0 -
X
INitributiiHi of afhijiuiHmc by rcKkin
T i
T
o
I
O 0
. I
MW M SO ttT.
i:.S. Region
J
o
1
\\
Figure 5-12. Distribution of Afternoon Time Spent Outdoors for School-age Children on
Days When Daily Maximum Temperature was > 84 °F, Stratified by Four U.S. Regions.
Females (left panels), males (right panels), weekday days (top panels), weekend days
(bottom panels) for where at least one minute of afternoon time was spent outdoors.
5.4.2 Characterization of Factors Influencing High Ozone Exposures
We investigated the factors that influence people experiencing the highest daily
maximum 8-hr average exposures. These exposure results in six selected study areas, Atlanta,
Boston, Denver, Houston, Philadelphia, and Sacramento, considering base air quality and air
quality just meeting the existing standard were combined with each simulated individual's
5-47
-------
microenvironmental time expenditure during the afternoon hours (12:00 PM through 8:00 PM),
times of day commonly when daily peak high Os concentrations occur. We first evaluated the
relative contribution seven variables23 had on the total explained variability in daily maximum 8-
hr average exposures. We then evaluated the distribution of identified influential variables for
simulated individuals with the highest exposures. And finally, we identified the
microenvironmental locations highly exposed people occupied and the activities performed
within them, given that within an 8-hr time frame most people would likely visit multiple
locations and perform different activities.
When considering only person days having the highest daily maximum 8-hr average Os
exposures at any of the six study areas and either air quality scenario and age groupings,
collectively the main effects of ambient concentrations and outdoor time combined with their
interaction similarly contribute to approximately 80% of the total explained variance results,
suggesting that for highly exposed people, the most important influential factors are time spent
outdoors corresponding with high daily maximum 8-hr average ambient Os concentrations.
The distributions of afternoon outdoor time and ambient concentration for highly exposed
individuals were evaluated considering base air quality and air quality adjusted to just meeting
the existing standard. As an example, exposure results in Boston indicated that for about half of
the days, simulated school-age children experiencing high exposures spend about 240 minutes
outdoors during the afternoon hours along with experiencing daily maximum 8-hr average
ambient Os concentrations > 75 ppb. In contrast when adjusting ambient concentrations to just
meeting the existing standard, for about half of the days, simulated school-age children
experiencing similar high exposures need to spend about 280 minutes outdoors during the
afternoon hours along with experiencing daily maximum 8-hr average ambient Os concentrations
> 60 ppb. Simply put, under conditions of lower ambient concentrations, people need to spend a
significantly greater amount of time outdoors to experience similar exposures observed at higher
ambient concentration conditions.
When considering these highly exposed children, on average about half of children's total
afternoon time is spent outdoors on high exposure days, 40% is spent indoors, while only 10% of
time is spent near-roads or inside motor vehicles. In general, greater than half of the time highly
exposed children spent outdoors specifically involves performing a moderate or greater exertion
level activity, such as a sporting activity. While apportionment of afternoon microenvironmental
time was similar for highly exposed adults in other age groups considered (e.g., 19-35),
23 The seven variables include the main effects of (1) daily maximum 8-hr average ambient Cb concentration, (2-4)
afternoon time spent outdoors, near-roads, and inside vehicles, and (5) physical activity index (PAI), while also
including interaction effects from (6) afternoon time outdoors by daily maximum 8-hr average ambient Os
concentration and (7) PAI by afternoon time outdoors.
5-48
-------
important high exertion activities performed outdoors also included those associated with paid
work and performing chores.
5.4.3 Exposure Results for Additional At-Risk Populations and Lifestages, Exposure
Scenarios, and Alternative Air Quality Input Data
5.4.3.1 Exposures estimated for all school-age children during summer months,
neither attending school nor performing paid work
As mentioned earlier in describing the longitudinal approach used in the main body of the
exposure assessment, the sequence of activity diaries for all simulated individuals is determined
by a user-selected profile variable of interest. In this assessment our longitudinal diary approach
uses time spent outdoors to link together CHAD diary days, an attempt to appropriately balance
intra- and inter-personal variability in that variable. For the primary exposure results, all
available diaries were used in developing any one sample pool without restriction outside of the
particular characteristics on interest in developing the pool (i.e., age, sex, day-of-week,
temperature, time spent outdoors). In this targeted simulation in Detroit during three summer
months of 2007 (June, July, and August), we restricted the diary pool of all school-age children
to include only those diary days that did not have any time spent inside a school nor had time
spent performing paid work during any day of the week. The results of this targeted simulation
were compared to an identical simulation, only differing in that all CHAD diary days were used
i.e., including any diary day for individuals having school time or paid work, and as was done for
the main body of this exposure assessment.
Figure 5-13 indicates that when restricting the CHAD diary pool to include only those
diaries having no time spent at school or performing paid work activities, there is about 1/3 or
33% increase in the number (or percent) of all school-age children at or above the 60 ppb
benchmark, a relationship also consistent across the alternative standards and when considering
multiple exposures. A similar relationship was found for the other benchmarks (not shown, see
Appendix 5G). Clearly, based on the analysis results reported in section 5.4.2 regarding factors
influencing those highly exposed, using only activity pattern data that do not include school or
work-related events (which would likely occur more so indoors than outdoors) and sampling
from a pool of diaries consistent with summer temperatures would increase the likelihood
simulated individuals spend time outdoors and be exposed to concentrations at or above the
selected benchmarks. This suggests that, for urban study areas having a traditional school
calendar (i.e., school not in session during the months of June, July and August), exposures at or
above selected benchmark levels could be underestimated by about 33%.
5-49
-------
>= 1 Exposure-All CHAD Diaries
>= 1 Expo sure -No School/Work Diaries
>= 2 Exposures-All CHAD Diaries
>= 2 Expo sure s -No School/Work Diaries
>= 3 Exposures-All CHAD Diaries
>= 3 Expo sure s -No School/Work Diaries
0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20%
Percent of Children with 8-hr Daily Max Exposure > 60 ppb
standard level (ppb)
! 60
I 65
170
175
Figure 5-13. Comparison of the Percent of All School-age Children in Detroit Having Daily
Maximum 8-hr Average Os Exposures at or above 60 ppb while at Moderate or Greater
Exertion, June-August 2007. Using all available CHAD diaries ("All CHAD Diaries") or
using CHAD diaries having no time spent in school or performing paid work ("No
School/Work Diaries").
5.4.3.2 Exposures estimated for outdoor workers during summer months
A targeted APEX simulation was performed for the Atlanta study area to simulate
summertime exposures for two hypothetical outdoor worker study groups, people between the
age 19-35 and 36-55, using 2006 air quality just meeting the existing standard. To do this, both
the daily and longitudinal activity patterns used by APEX were adjusted to best reflect patterns
expected for outdoor workers (e.g., a standardized work schedule during weekdays) while also
maintaining variability in those patterns across various occupation types. Briefly, the distribution
of all employed individual's occupations was estimated using data provided by the U.S. Bureau
of Labor and Statistics (US BLS, 2012b)24 and linked with 144 occupation titles from the
Occupational Information Network (O*NET)25 identified as having one or more days per week
where paid work was performed outdoors. These data were then aggregated to twelve broadly
defined BLS occupation groups, generating a data set containing the number of days per week
work time would be performed outdoors by that occupation group and properly weighted to
reflect the population distribution of people employed in each outdoor work group. Then,
24 U.S. employment data by SOC codes were obtained from: http://www.bls.gov/emp/ffiables: Table 1.2
Employment by occupation, 2010 and projected 2020.
25 Additional information is available at http://www.onetonline.org.
5-50
-------
existing CHAD diary days reflecting outdoor paid work were identified, isolated and replicated
to reflect this BLS/O*NET outdoor participation rate and occupation group frequencies. A
10,000 person simulation was performed by APEX using this adjusted CHAD activity pattern
database designed to simulate outdoor workers and compared with exposure results generated
from an identical APEX simulation of all employed people, though differing by using the
standard CHAD database and population-based modeling approach used in the main body
HREA. Details regarding the development of CHAD activity patterns used as input to simulate
outdoor workers, as well as other settings and conditions for APEX is described in Appendix 5G.
Estimated exposures are presented in Figure 5-14 for one of two age study groups
investigated (results for both age groups were similar) and considering either a longitudinal
approach designed specifically to reflect an outdoor worker weekday schedule (top panel) or
when using our general population-based modeling approach (bottom panel). The results indicate
that when accounting for a structured schedule that includes repeated occurrences of time spent
outdoors for a specified study group, all while simulated individuals are likely to be more
consistently performing work tasks that may be at or above moderate or greater exertion levels,
there are a greater percent of the study group experiences exposures at or above the selected
health effect benchmark levels than that estimated using our general population-based modeling
approach. Outdoor workers are expected to experience more exposures at or above benchmark
levels, though represent a fraction of the total employed population. It is possible that, in using
the general population-based approach along with the longitudinal algorithm that accounts for
within and between variability in outdoor time, a number of outdoor workers are incidentally
simulated and represent a significant portion of those who experienced exposures at or above
benchmark levels.26 However, the differences between exposures estimated for the two
longitudinal approaches become much greater when considering the percent of people
experiencing multiple exposure days at or above benchmark levels, primarily when considering
the 60 ppb benchmark. For example, 2% or less of the general population-based exposure group
was estimated to have two or more daily maximum 8-hr exposures > 60 ppb, while >17% of
specifically simulated outdoor workers were estimated to experience exposures at or above that
exposure benchmark.
26 Approximately 30% of our outdoor worker study group were estimated to experience at least one daily maximum
exposure > 60 ppb while at moderate or greater exertion. Assuming outdoor workers comprise 12% of the
workforce (Appendix 5G, Table 5G-8), outdoor workers experiencing at least one exposure at or above the 60 ppb
benchmark could contribute a value of 3.6% to the total exposed population (i.e., one comprised of outdoor and
non-outdoor workers). For the same air quality scenario and using the general population-based approach, we
estimated 5-8% of a total employed study group (incidentally comprised of outdoor and non-outdoor workers)
would experience exposures at or above the same benchmark, suggesting 48-75% of people experiencing
exposures at or above the 60 ppb benchmark have similar activity pattern characteristics as outdoor workers.
5-51
-------
Outdoor Worker Scenario-based Approach (ages 19-35)
Atlanta, 2006, Just Meet Existing 75 ppb Standard
II
I/I 4->
X —
UJ £
M >
01
30% f
25% -<
20% -'
X -* >-
s- ™ s
c Ot
O u m
i. 10%
O
Q. "O
£
-------
5.4.3.3 Exposures estimated for all school-age children when accounting for
averting behavior
A growing area of air pollution research involves evaluating the actions people might
perform in response to high Os concentration days (ISA, section 4.1.1). Most commonly termed
averting behaviors, they can be broadly characterized as personal activities that either reduce
pollutant emissions or limit personal exposure levels. The latter topic is of particular interest in
this HREA due to the potential negative impact it could have on Os concentration-response (C-
R) functions used to estimate health risk and on time expenditure and activity exertion levels
recorded in the CHAD diaries used by APEX to estimate Os exposures. To this end, we have
performed an additional review of the available literature here beyond that summarized in the
ISA to include several recent technical reports that collected and/or evaluated averting behavior
data (Graham, 2012). The purpose was to generate a few reasonable quantitative approximations
that allow us to better understand how averting behavior might affect time-location-activity
patterns, and then simulate how such personal adjustments might affect our population exposure
estimates.
Based on the elements evaluated in our literature review (i.e., air pollution awareness,
prevalence and duration of averting response), we conclude that most people are aware of alert
notification systems (in particular those people having compromised health and reside in an
urban area). We approximate that 30% of all asthmatics (or 15% of the general population) may
reduce their outdoor activity level on alert days (e.g., KS DOH, 2006; McDermott et al., 2006;
Wen et al., 2009; Zivin and Neidell, 2009) and that outdoor time/exertion during afternoon hours
may be reduced by about 20-40 minutes in response to an air quality alert notification
(Bresnahan et al., 1997; Mansfield et.al, 2006, Neidell, 2010; Sexton, 2011). We used these
literature derived estimates to generate an adjusted activity diary pool27 used by APEX to
simulate a 2-day exposure period (August 1-August 2, 2007) in Detroit to approximate the effect
this degree of averting may have on exceedances of exposure benchmarks.
When considering base air quality and our designed target to represent averting
performed by the general population - 15.3 % of all simulated school-age children spent on
average 44 minutes less time outdoors - resulting in approximately one percentage point or
fewer children experienced exposures at or above any of the selected benchmark levels (Figure
5-15, top panel). When considering base air quality and our designed target to represent an
averting response by the population of asthmatics - 30.3% of simulated asthmatic school-age
27 Because most activity diaries are limited to a single day and the survey participants were not directly asked if they
altered their daily activities in response to a high air pollution event, we do not know if any diary day represents
the activities of an individual who averted. Thus it is entirely possible that the "no averting" simulation includes,
to an unknown extent, individuals who spent less time outdoors than would have occurred if absolutely no
individuals averted.
5-53
-------
children spent on average 44 minutes less time outdoors - resulting in approximately two
percentage points or fewer experienced exposures at or above any of the selected benchmark
levels (Figure 5-15, bottom panel).
14%
ttj ra
g S! 12%
I No Averting
115.3% avert a mean of 44 minutes
70
80
8-hr Exposure Benchmark (ppb)
14%
No Averting
30.4% avert a mean of 44 minutes
70 80
8-hr Exposure Benchmark (ppb)
Figure 5-15. Percent of All School-age Children (left panel) and Asthmatic School-age
Children (right panel) Having at Least One Daily Maximum 8-hr Average Os Exposure at
or above Benchmark Levels while at Moderate or Greater Exertion During a 2-day
Simulation in Detroit, Base Air Quality, August 1-2, 2007. Red bars indicate exposure
results when considering effect of averting.
5-54
-------
5.4.3.4 Comparison of APEX estimated exposures using three different base air
quality data sets: AQS, VNA, and eVNA
For this exposure assessment, we elected to use a modeling approach to estimate the
ambient input concentration field and better account for spatial gradients that may exist (Chapter
4). To support the selection of Voronoi Neighbor Averaging (VNA), we compared exposure
results separately generated using ambient monitor (AQS), enhanced Voronoi Neighbor
Averaging (eVNA; a combined VNA and air quality modeling approach), and VNA as input to
APEX for three study areas: Atlanta, Detroit, and Philadelphia. All APEX settings were
generally consistent with the simulations discussed previously, though the air quality data
differed in that the year selected was 2005 (based on the available CMAQ data) and that a 4 Km
grid was used to define the spatial area for this evaluation rather than census tracts. Daily
maximum 8-hr average exposures were estimated for asthmatic school-age children residing in
the same census tracts comprising each air quality domain and summarized in Figure 5-16.
Atlanta
Philadelphia
o £
2 3
« Of
45%
40%
2.2 '
I s
.c O
u «M
£ c
00 O
•
E £
x
M oj
O *•
Q. n»
S!
oo
E
70 80 60 70 80 60
8-hr Exposure Benchmark (ppb)
70
80
Figure 5-16. Comparison of APEX Exposure Results Generated for Three Urban Study
Areas (Atlanta, Detroit, and Philadelphia) using Three Different 2005 Air Quality Input
Data Sets: AQS, VNA, and eVNA.
5-55
-------
Exposure results for all three air quality input data sets were very comparable, with a few
notable differences. Using AQS monitor concentration data tended to result in a 1-3% greater
percent of asthmatic school-age children at or above each of the selected benchmark levels when
compared with exposures estimated using VNA concentrations. While the VNA concentrations
are based on the AQS monitor data, the approach generates a concentration gradient with
distance from areas of known concentration that are typically less than the observed values, thus
yielding fewer individuals exposed to the highest concentrations. Using the eVNA approach to
generate ambient concentrations tended to result in 2-5% greater percent of asthmatic school-age
children at or above each of the selected benchmark levels when compared with exposures
estimated using either the AQS or VNA approaches. This is because at times, the eVNA
approach estimated high concentrations in areas where no observations were present, based on
modeling which captures gradients in Os that may result from nearby sources (see Chapter 4).
5.4.3.5 Comparison of APEX estimated exposures using two different adjusted air
quality data sets: quadratic rollback and HDDM
We elected to use an air quality modeling based approach rather than the previously used
statistical approach to adjust air quality to just meet the existing and alternative standard levels
(Chapter 4). To support the selection of the HDDM approach, we compared exposure results for
the scenario of just meeting the existing standard, separately generated using air quality inputs
obtained using the quadratic rollback and HDDM method to adjust air quality for the Atlanta
study area. All APEX settings were generally consistent with the simulations discussed
previously, though both the air quality data sets used in this comparison differed from that done
in the main exposure results above in that only the ambient monitor locations were used to define
the air districts and assumed a 30 km radius of influence, as was done for the 1st draft Os HREA.
Daily maximum 8-hr average exposures were estimated for asthmatic school-age children in
census tracts within 30 km of each air district and summarized in Figure 5-17.
The quadratic adjusted air quality resulted in slightly fewer percent of asthmatic school-
age children exposed at or above the highest exposure benchmark (80 ppb) when compared with
exposures estimated using the HDDM model simulation approach, though a significantly greater
percent of asthmatic school-age children were exposed to the lowest exposure benchmark (60
ppb) using the quadratic approach. This is because the quadratic approach generally targets the
highest concentrations for adjustment, while the HDDM approach accounts for changes across
the full concentration distribution to a varying degree to meet the adjusted concentration level of
interest.
5-56
-------
Quadratic Rollback
1 CO/
ge Children with at
Benchmark While a
Air Quality Adjusted
ng Standard
I-* h
3 IvJ I
? a? 5
< c tn
o o o •=
O -° '•? u
4S « fe
u !_ QJ OJ
fl ° i2 •£
W j_i co/
Percent of Asthmati
;ast One Exposure a
Moderate or Greater
to Just Meet
3 U» €
> Sp X
s d^ 0
CM
8
CM
(M
60
70
HDDM-Model Simulation
_
80 60 70 80
8-hr Exposure Benchmark (ppb)
Figure 5-17. Comparison of Exposure Results Generated by APEX using Two Different
Air Quality Adjustment Approaches to Just Meet the Existing Standard in Atlanta:
Quadratic Rollback (left panel) and HDDM-model Simulation (right panel).
5.4.4 Limited Performance Evaluations
5.4.4.1 Personal exposure comparisons
A new evaluation of APEX was performed using a subset of personal Os exposure
measurements obtained from the Detroit Exposure and Aerosol Research Study (DEARS) (Meng
et. al, 2012). For five consecutive days, personal Os outdoor concentrations along with daily
time-location activity diaries were collected from 36 adult study participants in Wayne County
Michigan during July and August 2006.28 An APEX simulation was performed considering these
same geographic and temporal features, followed with the sub-setting of APEX output data
according to important personal attributes of the DEARS study participants (5-day collection
study periods, age/sex distributions, outdoor time, ambient concentrations, and air exchange
rate). A comparison sample was generated randomly from the complete simulation, selecting for
50 APEX simulated individuals.
28 The DEARS was primarily designed to estimate human exposure to paniculate matter (Rea et al., 2003).
5-57
-------
For both data sets and considering the two output variables separately (outdoor time and
daily exposure), the median daily values for each study participant were ranked, then plotted
along with each individual's corresponding minimum and maximum value using each
individual's 5 person-days of data (Figure 5-18). In spite of the distinct matching of influential
personal attributes, over 50% of APEX simulated individuals had median daily Os exposure
concentrations above 10 ppb, while only 3% of DEARS participants' median values exceeded 10
ppb. These differences between the APEX-modeled daily exposures and DEARS-measured
personal exposures could possibly be related to differences in precision and assumptions
regarding both methods, and are discussed with the following.
DEARS ESTIMATED
Dally Mean O3 Exposure (ppb)
10 15 20
25
APEX ESTIMATED
Dally Mean O, Exposure (ppb)
10 15 20
120 180 240 300 360 420 480 540 600
Dally Total Outdoor Time (minutes)
120 180 240 300 360 420 480
Dally Total Outdoor Time (minutes)
Figure 5-18. Distribution of Estimated Daily Average Os Exposures (top panels) and Daily
Total Outdoor Time (bottom panels) for DEARS Study Adult Participants (left panels) and
APEX Simulated Adults (right panels) in Wayne County, MI, July-August 2006.
Regarding the DEARS measurement data, personal exposures were estimated using a
passive sampling device and time-averaged over 24-hours. The estimated detection limits for this
study using this approach were 3 ppb, though commonly reported detection limits for most
studies using these devices are generally within 5-10 ppb (Os ISA, section 4.3.1). As shown in
Figure 5-18, nearly 80% of the sampled individuals had a median daily personal exposure at or
5-58
-------
below the DEARS reported 24-hr detection limit of 3 ppb, approximately 85% of sampled
individuals had a median daily personal exposure at or below a minimum detection value of 5
ppb, and 97% of sampled individuals had a median daily personal exposure at or below a
minimum detection value of 10 ppb. Given the significant removal and decay of ambient Os
within indoor microenvironments (indoor is commonly -60-90% less than ambient
concentrations; Os ISA, section 4.3.2), the large amount of daily time spent indoors by most
people, particularly adults (Os ISA, section 4.4.1), and that daily average ambient concentrations
are usually about 30 ppb (Os ISA, Table 3-7), the 24-hr passive sampler limits of detection are an
important consideration in understanding the reported DEARS results and other personal Os
exposure measurement studies where, commonly, a substantial portion of daily samples are
reported as being below the limit of detection (e.g., Sarnat et al., 2006).
In general, comparisons of the concentrations measured using passive sampling devices
with those derived from active sampling methods shows reasonable agreement between the two
approaches, giving some degree of confidence in concentrations estimated using these passive
sampling devices (e.g., Liu et al., 1993). However, it is not uncommon for these evaluation
studies to be conducted over longer time periods (days to weeks) due to a lessened confidence in
their ability to reasonably measure short-term (< 24-hr time averaged samples) low-level Os
concentrations, particularly at levels observed when estimating personal Os exposures and Os
concentrations in indoor microenvironments (e.g., Winter et al., 2003; Bytnerowicz et al., 2004).
Further, a greater measurement bias exists between passive and active sampling at higher Os
concentrations (Winter et al., 2003) and potentially the result of a reduced sampling rate due to
saturation of the passive device receptors. In addition, relative errors in passive measurement of
up to 90% have been reported for Os concentrations between 0-10 ppb (Liu et al., 1993). Thus, it
is possible that the DEARS passively measured personal Os exposures are biased low due to the
inability of the sampler to capture short-term (1-min to 80 ppb).
To explore this issue further here, Liu et al. (1993) reported that 12-hr outdoor passive
measurements were strongly correlated (r=0.95, p<0.01) with 12-hr measurements made at a co-
located continuous monitoring site, a site having mean 24-hr ambient Os concentrations of 37.8
ppb (min 8.3 ppb, max 64.3 ppb). Twelve-hour personal daytime (8AM-8PM) passive Os
exposure samples measured in 23 children ranged from 0.5 to 78.8 ppb (mean 23.9 ppb),
concentrations that were also well correlated with 12-hr outdoor (r=0.41, p<0.01) and 12-hr
indoor (r=0.55, p<0.01) passive measurements. Ratios of the daytime 12-hr personal Os
5-59
-------
concentrations to 12-hr ambient Os concentrations were on average 0.44 (± 0.29 standard
deviation). Further, time activity diaries collected in the Liu et al. (1993) study subjects indicated
that on average, the children spent 30% (± 22%) of their time outdoors, and based on evaluations
in this REA, are an important influential factor driving the personal concentration levels
(particularly those having the highest exposures) and likely contributing to the overall strong
correlations between personal and ambient concentrations.
For comparison, the DEARS 24-hr ambient concentrations measured using a passive
sampling device ranged from 17-52 ppb (mean 30.2 ppb), lower but generally comparable to the
Liu et al. 24-hr values. However, the DEARS 24-hr personal exposures ranged from 1-16 ppb
(mean 3.4 ppb), much lower than a mean of 17.2 ppb approximated for a 24-hr personal
exposure in the Liu et al. (1993) study, even when assuming the remaining 12-hrs of day was
spent entirely indoors (i.e., having a mean indoor nighttime concentration of 10.5 ppb reported
by Liu et al., 1993). Ratios of the DEARS 24-hr personal Os concentrations to 24-hr ambient Os
concentrations were on average 0.12 (± 0.09 standard deviation) also lower than the values
reported by Liu et al. (1993) and below that reported by most exposure measurement studies
discussed in the Os ISA (generally ranging from 0.3-0.8, section 4.3.3). In addition, the DEARS
personal Os exposure measurements were weakly correlated with ambient concentrations
(r=0.19, p=0.014) and within the lower range of values reported in the Os ISA for comparisons
of personal Os exposures to indoor residential microenvironmental Os concentrations.
Substantially less time spent outdoors on average for DEARS study subjects (mean 6.8% of daily
time) is likely a major factor contributing to the overall weak relationship in the 24-hr personal
Os exposure in relation to the 24-hr ambient Os concentrations.
Contrary to these uncertainties and influential factors in the DEARS 24-hr personal Os
exposure measurements, modeling Os exposure is as precise as is allowed by the APEX input
data. The time resolution and level of the ambient Os concentrations used in this HREA (1-hr
and < 1 ppb) combined with that of the time location activity diaries used to represent simulated
individuals (minutes) result in Os exposure concentrations that can be calculated for events
having a minimum duration of 1-minute and at concentrations less than 1 ppb. Simulated
individuals having variability in exposure concentration of any magnitude will be precisely and
immediately modeled, including when the individual transitions from short-term high
concentration exposure events to low concentration exposure events, for any duration of time.
Thus, simply when accounting for standard methodological differences in calculating exposures,
particularly when including those that are at or below the measurement device limits of detection
(such as indoor Os concentrations), APEX modeled daily exposures would likely in most
instances be greater (as is observed in Figure 5-18) than that estimated using a passive personal
measurement device.
5-60
-------
It is also possible that the indoor microenvironmental concentrations estimated by APEX
are biased high, potentially resulting an upward bias in the 24-hour time averaged
concentrations. Two variables that influence indoor Cb concentrations are the air exchange rate
and the indoor decay rate. As shown in Appendix 5G Figure 5G-11, the overall distribution of
AER values estimated for APEX simulated individuals was actually lower than that estimated for
DEARS study participants, a factor that would lead to generally lower indoor exposures for the
APEX simulated individuals compared to exposure estimated for the DEARS subjects, if having
a similar indoor decay rate. APEX estimated exposures account for the Os decay expected to
occur within all indoor microenvironments based on findings of Lee et al. (1999). Unfortunately,
neither of these variables (indoor Os concentrations and Os decay rate) were measured in the
DEARS study so a direct comparison of these two variables cannot be made between the two
approaches.
APEX estimated daily average indoor residential microenvironmental concentrations for
the simulated individuals were divided by daily average ambient concentrations to generate a
metric comparable to commonly reported Indoor/Outdoor (I/O) ratios (Os ISA, section 4.3.2).
Median I/O values of APEX simulated individuals range from 0.1-0.6, having a central tendency
of about 0.26, while minimum and maximum values occurring on a single day ranged from about
0.03-0.74 (Figure 5-19). These central tendency values of I/O estimated using APEX are
consistent with daily mean I/O ratios reported in the Os ISA for most studies (0.1-0.4), with
summertime I/O ratios reported as high as 0.58. Higher I/O values are likely related to instances
where infiltration of outdoor ambient concentrations is increased, such as when windows are
opened. It is possible that some of the highest I/O ratios estimated here are a result of the highest
AER used in estimating indoor residential Os concentrations (Appendix 5G, Figure 5G-11) but
again, the rate of Os decay could also play a role if underestimated for the simulated study group.
In addition, there are limited instances of high AER rates (e.g., AER > 2 hr"1) used for all APEX
simulated individuals, while high AER for tended to occur for only about half of the DEARS
study subjects, indicating that there may be instances where APEX indoor simulated
concentrations are greater than would be expected where an actual residential AER would be less
variable than that modeled. Thus, it is possible that indoor Os concentrations could be biased
high on occasion due to a wider range of variability in AER equally applied across simulated
individuals' than should be as well as the potential for underestimation of indoor Os decay.
5-61
-------
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Daily Average Indoor/Outdoor Concentration Ratio
0.9
Figure 5-19. Distribution of Indoor/Outdoor Concentration Ratios for APEX Simulated
Adults in Wayne County, MI, July-August 2006.
This observed difference between the DEARS personal exposure measurements and the
APEX modeled exposures is important, though given the above discussion and understanding the
most important influential factors leading to exposures at or above the benchmarks (i.e.,
substantial time spent outdoors and outdoor ambient concentrations) this difference may be of
limited relevance to the key analysis results in this REA. In this particular evaluation, none of the
APEX simulated individuals experienced a daily maximum 8-hr exposure at or above any
exposure benchmark while at moderate or greater exertion. There were however a number of
person-days where simulated adults had an 8-hr exposure at or above 60 ppb, just not occurring
at elevated activity levels. In using the existing output data and variables available from the
simulation, we evaluated these occurrences of elevated daily maximum 8-hr exposures along
with the total time spent outdoors (Figure 5-20). While the daily maximum 8-hr exposure is
calculated as a running average and is comprised of all microenvironmental exposures that occur
for the simulated individual (including those occurring indoors), most critical to the occurrence
of benchmark exceedances here is when individuals spend a significant amount of their time
spent outdoors, generally ranging from between 420 to 800 minutes of total outdoor time per
day.
5-62
-------
I £.
3"
Q.70
|"
068
Q.
X
LU
«
o
. cc
«
ERA
'x
1
>l
•— fi9
CO ^
Q
Rn
•
•
,
•
**
- ;
*
• ..•*
"!• * *(
• r • ^
* ™'
* * •
i • ^ * £• 1
*
«
•
*
*
• • . •
i i _ jf
*
1 '** I
i • • t*
•I * ** *•
;l ••;*•
«!*',* I*
^1% •* *
•i'.j:'i:i
•
*
•
* .
v ••
• •. • ,
•:'• • :*
Jv
•
•
* *
• •
*
•
• •
*
*
I
*
*
*
:
*
•
.
.
•
*
1
180
360 540 720 900 1080
Total Time Spent Outdoors (minutes)
1260
1440
Figure 5-20. Daily Maximum 8-hr Exposures at or above 60 ppb Estimated by APEX for
Simulated Adults in Wayne County, MI, July-August 2006.
A second model-to-measured comparison is provided here that is also relevant to the
current REA. APEX modeled exposures have previously been compared with personal exposure
measurements for Os for the 2007 NAAQS review (US EPA, 2007b). Briefly, APEX Os
simulation results were compared with 6-day personal Os concentration measurements for
children ages 7-12 (Xue et al., 2004; Geyh et al., 2000). Two separate areas of San Bernardino
County were surveyed: urban Upland CA, and the combined small mountain towns of Lake
Arrowhead, Crestline, and Running Springs, CA. Available ambient monitoring data for these
locations during the same study years (1995-1996) were used as the air quality input to APEX.
APEX predicted personal exposures, averaged similarly across a 6-day period, matched
reasonably well for much of the concentration distribution considering both locations, but tended
to underestimate exposures at the upper percentiles of the distribution. The average difference
between the 6-day means was less than 1 ppb, with a range of-11 ppb to +8 ppb, though
predicted upper bounds for a few averaged exposures having higher exposure concentrations
were under-predicted by up to 24 ppb (e.g., Figure 5-21). In addition, modeled exposure
concentration variability was less than that observed in the personal exposure measurements. At
5-63
-------
the time of analysis, these differences were proposed to be largely driven by under-estimation of
the spatial variability of the outdoor concentrations used by APEX (US EPA, 2007b).
b
CD
O
rirTfllirll:
measured
Figure 5-21. Means (and range) of 6-day Average Personal Os Exposures, Measured and
Modeled (APEX), Upland Ca. Obtained from Figure 8-22 of US EPA (2007b).
5.4.4.2 Ventilation rate comparisons
The algorithm used by APEX to estimate minute-by-minute ventilation rate serves as the
basis for recent updates to the ventilation rate distributions provided in EPAs Exposure Factors
Handbook (U.S. EPA, 2009b; US EPA, 2011). During the development of the ventilation
distributions for EPA at that time, two peer-reviewed studies were identified as providing
somewhat relevant measurement data to evaluate the APEX energy expenditure and ventilation
algorithm (see Graham, 2009 for additional comparison details). The results of this evaluation
are summarized below.
Briefly, Brochu et al. (2006a,b) presents data for ventilation rates derived from tracking
doubly-labeled water (DLW) consumption/elimination to estimate energy expenditure in healthy
normal-weight males and females, ages from 1 month to 96 years (n=l,252). Estimates of energy
expended were combined with a fixed oxygen uptake factor (H=0.21) and using a fixed
ventilatory equivalent (VQ)29 of 27. The DLW measurement period ranged from 7-21 days,
resulting in time-averaged metrics that may in some instances provide reasonable estimates for a
mean daily ventilation rate, but not useful for estimating variability in an individual's ventilation
rate over shorter time periods (as is needed by APEX). Further, while DLW is considered by
some as a 'gold standard' for measuring energy expenditure, this characterization would not
necessarily be directly transferable to approximations that use this measured value (i.e.,
ventilation rate in Brochu et al. (2006a,b) is a calculated value, not measured). Reported
' The ventilatory equivalent (VQ) is the ventilation rate (VE) divided by the oxygen consumption rate
5-64
-------
ventilation rates are daily averages for several age groupings (e.g. ages 1 to < 2, 2 to < 5, 5 to <
7, etc.) along with derived percentiles, each assuming the existence of normally distributed data.
A 14-day APEX simulation was performed (i.e., the median of 7-21 days for the DLW
measurement study) to estimate daily ventilation rates for comparison with the time-averaged
Brochu et al. (2006a) data. Twenty-five thousand people were simulated by APEX to generate
reasonable numbers of people within each year of age and other potential categorical variables
(e.g., 100-200, although a few older age groups resulted in having fewer individuals). It is
important when comparing the two types of data for them to be similar as possible, particularly
since age and body mass are important influential variables in both estimation methods. A total
of 9,613 normal-weight individuals were simulated by APEX and used for the following
analysis. Multi-day ventilation rates were averaged across the 14-day simulation period, yielding
a mean daily ventilation rate for each person to best represent the DLW time averaging done by
Brochu et al. (2006a).
Figure 5-22 compares the APEX simulated individuals body mass normalized mean daily
ventilation rates with those reported by Brochu et al. (2006a; Table 2, page 684) for several age
groupings of normal-weight individuals. The two largest differences appear for children of both
sexes less than age 10 (i.e., Brochu et. al (2006a) estimates are systematically lower than APEX
estimates) and for ages 16-33 (i.e., APEX estimates are lower than Brochu et al (2006a). Body
mass normalized ventilation rates also appear to be slightly higher using APEX when
considering people above age 64 and for both sexes.
One principal issue identified by us as potentially responsible for some of the above
differences in ventilation estimates is in the VQ used by Brochu et al. (2006a). A single value of
27 was used in estimating ventilation rates for both children and adults, however it is widely
recognized that while a VQ of 27 may be a reasonable approximation for estimating mean
ventilation rates of adults, it is not appropriate for use in estimating mean ventilation rates in
children. With this in mind, the Brochu et al. (2006a) ventilation estimates were modified here
using the VQ estimates offered by Arcus-Arth and Blaisdell (2007). Figure 5-23 illustrates the
comparison of APEX body mass normalized mean daily ventilation rates with that of Brochu et
al. (2006a) corrected ventilation estimates. The body mass normalized ventilation estimates for
school-age children are more similar to those generated by APEX when correcting the Brochu et
al. (2006a) VQ parameter. Thus, mean ventilation rates generated by APEX are reasonably
correlated with independent measures from the Brochu et al. (2006a, b) estimates, particularly
when correcting the Brochu et al. (2006a) ventilation estimates for children using a more
appropriate estimate of VQ for children.
In a second study identified for comparison with APEX estimates, Arcus-Arth and
Blaisdell (2007) provide ventilation estimates for children <19 years of age using energy intake
5-65
-------
(El, or calories consumed) and body mass data provided by the USDA's Continuing Survey of
Food Intake for Individuals (CSFII; USD A, 2000). Two-day daily average Els were combined
with a values of H (i.e., 0.22 for infants, 0.21 for non-infant children) and VQ (i.e., 33.5 for
children 0-8, 30.6 for boys 9-18, 31.5 for girls 9-18 years old). Again, time-averaging of the data
may provide reasonable estimates of a daily mean, but offer no variability in ventilation
estimates for shorter durations. Furthermore, data for both sexes are combined and reported by
age, with stratified results by sex reported only for aggregated age groups (males and females, 9-
18 years old).
m
Q
c
n
g
o -r
^ >»
^ ro
^ ~o
N <*>
"re ^
!i
z^S
181
re £
s£
>»
T3
O
00
0.1 --
0.0
1
A
.1
A
A
"*/
AAA!
-
-
1 1
IA
A^ffAAAAAA
o:Bi4|3ca
A*
1 1
xo
/WWW\.
*****
t^W
•APEXFemales
o Brochu etal. (2006) Females
A APEX Males
* Brochu etal. (2006) Males
I nrV
'L'"
0 10 20 30 40 50 60
Age (years)
70 80
90 100
Figure 5-22. Comparison of Body Mass Normalized Mean Daily Ventilation Rates
Estimated by APEX (closed symbols) and by Brochu et al., 2006 (open symbols).
5-66
-------
1.1
-^ 1.0
re
Q_ 0.9
re
o>
0.8 i
;>»
re
p 0.7
=reS0.6
I | °'5
w| °-4
5 o 0.3
" 0.2
0.1
0.0
•a
o
GO
-
i
C
-
1
1
* \
5 1
!
,
3 i
t
<
5 i
! !
» ,
1 !
1 <
!
•APEX Females
.
o Brochuetal. (2006) Females- Corrected VQ
A APEX Males
* Brochu et al. (2006) Males - Corrected VQ
i i i i i i i i
L
' i
\ j
! 1
i
I i
«
!
i
l I
> i
^ i
"> i
i /
?_ t
^ * 1
1
! 1
t
?. t
1 <
,
?. !
> -
!
*
5.
k
I
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Age (years)
Figure 5-23. Comparison of Body Mass Normalized Mean Daily Ventilation Rates in Male
and Female School-age Children (5-18) when Correcting Brochu et al. (2006a) Results with
Child Appropriate VQ Estimates.
A 2-day model simulation was performed by APEX to generate ventilation estimates for
children to compare with results of Arcus-Arth and Blaisdell (2007).30 APEX ventilation
estimates were time-averaged to generate mean daily values, and since the data reported in
Arcus-Arth and Blaisdell (2007) were not separated by sex (outside of broad age categories), the
APEX estimates were also combined by sex to provide a comparable mean estimate for each
year of age (5-18). Body mass was also not used as a categorical variable in Arcus-Arth and
Blaisdell (2007), therefore all APEX simulated individuals were used, regardless of whether they
could be classified as overweight or of normal weight. In addition, daily ventilation rates for a
few age groups of children were obtained from Tables 3 and 4 of Brochu et al. (2006a), though
considering both estimates for normal and overweight individuals (there were no combined data
available). The Brochu et al. (2006a) results have been corrected for VQ as noted above using
VQ estimates of Arcus-Arth and Blaisdell (2007) and added for comparison. Figure 5-24
illustrates ventilation rate estimates from the APEX simulation, along with associated data for
1 Table III, page 103 of Arcus-Arth and Blaisdell (2007) provided body mass normalized ventilation rates.
5-67
-------
school-age children (ages 5-18) obtained from the two publications. Daily mean ventilation
estimates are quite similar at each year of age, with slightly higher estimates by Arcus-Arth and
Blaisdell (2007) at ages 9 and above, particularly when compared with APEX ventilation
estimates. Ventilation estimates are remarkably similar for school-age children for all three
sources of data, particularly when considering the differences in the type of input data used and
the varied approaches of APEX, Brochu et al. (2006a), and Arcus-Arth and Blaisdell (2007).
This overall agreement suggests reasonable confidence can be conferred to the algorithm used by
APEX to estimate at a minimum, daily mean ventilation rates.
1.1
O->-
c
.0
JS 0.
£ 0.
1»°-
N •*
l£°-
E p
3
o.<
Z^0'
w u-v
tn
m
^
>»
•o
o
GO
o.:
0.'
0.(
-
-
-
-C
t
-
>.7r^
j,
,
\ j.
—
—
-•-APEX Mean
o Brochu etal. (2006) Normals Mean- Corrected VQ
-x- Arcus-Arth and Blaiedell (2007) Mean
A Brochuetal. (2006) OW/OB Mean -Corrected VQ
3
• ©"""--•«.>
•i
—
'---s
C:.~::i
!— 1
'ItT±-=n>vJ
^ i
--.-^
y c
tn—ni
3 (
* — i
3 (
^
3 C
1 i
C )
t- 1
3 C
i
*—• s
1
3
^
K
•
7 8 9 10 11 12 13 14 15 16 17 18 19
Age (years)
Figure 5-24. Comparison of Body Mass Normalized Daily Mean Ventilation Rates in
School-age Children (5-18) Estimated using APEX and Literature Reported Values.
5.4.4.3 Evaluation of longitudinal profile methodology
We evaluated the APEX approach used for linking together cross-sectional activity
pattern diaries to generate longitudinal profiles for our simulated individuals (Appendix 5G,
Section 5G-3). Of particular interest were how well variability in outdoor participation rate and
the amount of time expended were represented in our population-based exposure simulations.
Our goal in developing the most reasonable longitudinal profiles is to capture expected,
important features of population activity patterns, i.e., there is correlation within an individual's
day-to-day activity patterns (though neither exactly repeated nor entirely random for individuals)
5-68
-------
and variability across the modeled study group in day-to-day activity patterns (i.e., not every
simulated individual in the study group does the same activity on the same day).
The simulated longitudinal profiles indicate the method for linking together cross-
sectional diaries generates a diverse mixture of people having variable, though expected, activity
patterns: A small fraction of the simulated population spend a limited amount of afternoon time
outdoors and occurring at a low frequency across an Os season, a small fraction consistently
spends a greater amount (> 2 hours) of time outdoors and occurring at greater frequency (e.g.,
4/5 days per week), while the remaining simulated individuals fall somewhere in between
regarding participation and total time. While we are not aware of a population database available
to compare with these simulated results, we are comfortable with the method performance in
representing the intended variability in longitudinal activity patterns (see section 5G-3 for
details).
5.4.4.4 Comparison of exposures estimated using the 2000 and 2010 U.S. Census
Data
The population demographics used in to generate the main body exposure results are
based on the 2000 Census. These include census tract-level counts of urban study area
population by age and sex, tract-level employment probabilities by age and sex, and tract-to-tract
commuting patterns. In this HREA, we are modeling the years 2006 to 2010, thus there is some
uncertainty associated with use of population demographics for a different year than being
modeled. To assess the extent of this uncertainty, APEX simulations were conducted for seven
study areas31 using 2010 air quality adjusted to just meet the existing standard and employing the
2010 Census information and compared estimated exposures with those generated from identical
APEX runs, though differing by use of the 2000 Census data for input. The exposure study group
of interest was all school-age children and were evaluated for occurrences of daily maximum 8-
hr exposures above the three exposure benchmarks concomitant with moderate or greater
exertion. Because the number, identification, and location of tracts can change from one census
to the next, these new simulations used the county to define the exposure modeling domain. As
such, results presented here are not necessarily directly comparable to the main body results in
this chapter.
Table 5-9 provides the estimated exposure results for each of the seven study areas,
indicating mostly small differences (<1%) in the percent of school-age children experiencing at
least one daily maximum 8-hr exposure at or above any of the exposure benchmarks when using
31 The seven study areas modeled here were selected for overall regional representation and to capture areas having
either negative and positive growth based on a comparison of their respective 2000 and 2010 CSA populations:
Atlanta (+23.5%), Boston (+3.6%), Cleveland (-2.2%), Detroit (-2.6%), Houston (+25.7%), Los Angeles (9.2%),
Sacramento (19.0%). http://www.iweblists.com/us/population/CombinedStatisticalAreaPop.html.
5-69
-------
the 2010 Census data compared to exposures estimated using the 2000 Census data. However,
for 6 of 7 study areas, the percent of school-age children exposed using the 2000 Census data
was higher. Sacramento was the only study area where use of the 2010 Census data resulted in
fewer school-age children experiencing exposures of concern, however, at most the percentage
point difference was 0.2%.
Table 5-9. Percent of All School-age Children with Daily Maximum 8-hr Exposures at or
above Exposure Benchmarks while at Moderate of Greater Exertion, 2010 Air Quality
Adjusted to Just Meet the Existing Standard, When Using Either 2000 or 2010 Census
Population Data.
Exposure
Benchmark
60 ppb
70 ppb
80 ppb
Study Area
Atlanta
Boston
Cleveland
Detroit
Houston
Los Angeles
Sacramento
Atlanta
Boston
Cleveland
Detroit
Houston
Los Angeles
Sacramento
Atlanta
Boston
Cleveland
Detroit
Houston
Los Angeles
Sacramento
Percent of All School-age Children
Experiencing at Least One
Exposure at or above Benchmark
2000 Census
13.2
11.1
8.3
15.7
14.4
6.5
4.3
1.7
1.6
0.4
2.4
2.2
0.2
0.6
0.1
0
0
0.1
0.1
0
0
2010 Census
12.1
10.8
7.3
14.9
14.3
6.4
4.5
1.4
1.6
0.4
2.2
2.1
0.3
0.6
0.1
0
0
0.1
0.1
0
0
Exposure
Difference1
Absolute
1.1
0.3
1.0
0.8
0.1
0.1
-0.2
0.3
0
0
0.2
0.1
-0.1
0
0
0
0
0
0
0
0
Percent
8.3
2.7
12.0
5.1
0.7
1.5
-4.7
17.6
0
0
8.3
4.5
-50.0
0
0
0
0
0
0
0
0
1 Absolute is the percentage point difference between 2000 and 2010 census generated exposures. Percent is the
absolute difference relative to the 2000 results, scaled by 100.
5.5 VARIABILITY AND UNCERTAINTY
An important issue associated with any population exposure or risk assessment is the
characterization of variability and uncertainty. Variability refers to the inherent heterogeneity in
a population or variable of interest (e.g., residential air exchange rates). The degree of variability
cannot be reduced through further research, only better characterized with additional
5-70
-------
measurement. Uncertainty refers to the lack of knowledge regarding the values of model input
variables (i.e., parameter uncertainty), the physical systems or relationships used (i.e., use of
input variables to estimate exposure or risk or model uncertainty), and in specifying the scenario
that is consistent with purpose of the assessment (i.e., scenario uncertainty). Uncertainty is,
ideally, reduced to the maximum extent possible through improved measurement of key
parameters and iterative model refinement. The approaches used to assess variability and to
characterize uncertainty in this HREA are discussed in the following two sections. Each section
also contains a concise summary of the identified components contributing to uncertainly and
how each source may affect the estimated exposures.
5.5.1 Treatment of Variability
The purpose for addressing variability in this HREA is to ensure that the estimates of
exposure and risk reflect the variability of ambient Os concentrations, population and lifestage
characteristics, associated Os exposure and dose, and potential health risk across the study area
and for the simulated at-risk study groups. In this HREA, there are several algorithms that
account for variability of input data when generating the number of estimated benchmark
exceedances or health risk outputs. For example, variability may arise from differences in the
population residing within census tracts (e.g., age distribution) and the activities that may affect
population and lifestage exposure to Os (e.g., time spent inside vehicles, time performing
moderate or greater exertion level activities outdoors). A complete range of potential exposure
levels and associated risk estimates can be generated when appropriately addressing variability in
exposure and risk assessments; note however that the range of values obtained would be within
the constraints of the input parameters, algorithms, or modeling system used, not necessarily the
complete range of the true exposure or risk values.
Where possible, we identified and incorporated the observed variability in input data sets
to estimate model parameters within the exposure assessment rather than employing standard
default assumptions and/or using point estimates to describe model inputs. The details regarding
variability distributions used in data inputs are described in Appendix 5B. To the extent possible
given the data available for the assessment, we accounted for variability within the exposure
modeling. APEX has been designed to account for variability in some of the input data,
including the physiological variables that are important inputs to determining ventilation rates.
As a result, APEX addresses much of the variability in factors that affect human exposure.
Important sources of the variability accounted for in this analysis are summarized in Appendix
5D.
5-71
-------
5.5.2 Characterization of Uncertainty
While it may be possible to capture a range of exposure or risk values by accounting for
variability inherent to influential factors, the true exposure or risk for any given individual within
a study area is unknown, though can be estimated. To characterize health risks, exposure and risk
assessors commonly use an iterative process of gathering data, developing models, and
estimating exposures and risks, given the goals of the assessment, scale of the assessment
performed, and limitations of the input data available. However, significant uncertainty often
remains and emphasis is then placed on characterizing the nature of that uncertainty and its
impact on exposure and risk estimates.
The HREA's for the previous Os, NCh, SCh, and CO NAAQS reviews each presented a
characterization of uncertainty of exposure modeling (Langstaff, 2007; US EPA 2008, 2009a,
2010). The qualitative approach used in this and other HREAs is described by WHO (2008).
Briefly, we identified the key aspects of the assessment approach that may contribute to
uncertainty in the exposure and risk estimates and provided the rationale for their inclusion.
Then, we characterized the magnitude and direction of the influence on the assessment results
for each of these identified sources of uncertainty. Consistent with the WHO (2008) guidance,
staff scaled the overall impact of the uncertainty by considering the degree of uncertainty as
implied by the relationship between the source of uncertainty and the exposure concentrations. A
qualitative characterization of low, moderate, and high was assigned to the magnitude of
influence and knowledge base uncertainty descriptors, using quantitative observations relating to
understanding the uncertainty, where possible. A summary of the key findings of those prior
characterizations that are most relevant to the current Os exposure assessment are provided in
Table 5-10.
5-72
-------
Table 5-10. Characterization of Key Uncertainties in Historical and Current Exposure Assessments using APEX.
Sources of Uncertainty
Category
Ambient Monitoring
Concentrations
Element
Database Quality
Instrument
Measurement Error
Missing Data
Substitution Method
Temporal
Representation
Spatial Representation:
Large Scale
Spatial Representation:
Neighborhood Scale (1)
Spatial Representation:
Neighborhood Scale (2)
Historical Uncertainty Characterization
Influence of
Uncertainty on
Exposure/Intake Dose
Estimates
Direction
Both
Over
Both
Both
Both
Both
Over
Magnitude
Low
Low
Low
Low
Low
Low
Low
Knowledge-
base
Uncertainty
Low
Low
Low
Low
Low
Low
Low
Comments
All ambient pollutant measurements available from
AQS are both comprehensive and subject to quality
control.
Mean bias estimated as 1.2% (CV of 4.4%). See
Table 2 and Figure 6 of Langstaff (2007).
Overall completeness of data yield negligible mean
bias (~0) along with an estimated standard deviation
of 4 ppb when replacing missing values. See Table
3 of Langstaff (2007).
Appropriately uses 1-hr time-series of Os
concentrations for 5 years. No missing data for any
hour input to APEX.
Tens of monitors used in each study area.
Spatial interpolation using jackknife method
(removal of a single monitor) yielded generally
unbiased observed/predicted ratios (mean 1.06),
having an estimated standard deviation of 0.2.
Langstaff (2007).
When reducing the APEX radius setting from an
unlimited value (actual value used) to 10 km (i.e.,
the tendency would be to more accurately represent
exposure), a smaller fraction (1-3 percentage points)
of population exceeds benchmark levels. See
Figures 7 - 9 of Langstaff (2007).
Is rating
appropriate for
current APEX Os
exposure
assessment?
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
Yes. For the
uncertainties
characterized, the
historical rating is
appropriate if and
when using ambient
monitor data alone
to represent air
quality surface.
However in this
HREA, local-scale
air quality was
estimated using
VNA (see below).
5-73
-------
Sources of Uncertainty
Category
Adjustment of Air
Quality to Simulate
Just Meeting the
Existing Standard
Element
Spatial Representation:
Local Scale VNA
estimates
Spatial Representation:
Vertical Profile
Quadratic Approach
HDDM Simulation
Approach
Historical Uncertainty Characterization
Influence of
Uncertainty on
Exposure/Intake Dose
Estimates
Direction
Both
Both
Both
Both
Magnitude
Low
Moderate
Low-
Moderate
Low-
Moderate
Knowledge-
base
Uncertainty
Low-
Moderate
Moderate
Moderate
Low-
Moderate
Comments
Scenario-based evaluation in three study areas
indicated small differences in exposure results when
comparing ambient monitor data or statistically
interpolated concentrations to 4 Km grid as an input
to APEX (Figure 5-16). General dependencies of the
approaches used could lead to observed lack of
distinction in exposure results.
Differences between ground-level (0-3 meters) and
building rooftop sited (25 meters) monitor
concentrations can be significant. Most importantly,
use of higher elevation monitors would tend to
overestimate ground-level exposures (i.e., for when
time is spent outdoors).
Variable differences (e.g., none to a factor of two or
three) in the estimated number of people exposed
across study areas when using differing 3-year roll-
back periods for a single year of air quality
(Langstaff, 2007).
Expected patterns in both air quality and exposure
result from HDDM/emissions reduction approach
(full distribution affected rather than only upper
percentiles, Figure 5-17). Variable differences
remain (e.g., none to a factor of two or three) in the
estimated percent of people exposed across study
areas when using differing 3-year roll-back periods
for 2008 air quality (Figures 5-5 to 5-9). New York
study area could not be simulated to just meet 60
and 55 ppb alternative standards.
Is rating
appropriate for
current APEX OT,
exposure
assessment?
Yes. Newly
evaluated.
Yes. Given judged
impact to exposure,
additional
characterization is
possibly warranted.
Yes. Uncertainty in
the approach has
resulted in use of
HDDM approach.
Yes. Newly
evaluated.
5-74
-------
Sources of Uncertainty
Category
APEX: General Input
Databases
Element
Population
Demographics and
Commuting (US
Census)
Activity Patterns
(CHAD)
Meteorological (NWS)
Historical Uncertainty Characterization
Influence of
Uncertainty on
Exposure/Intake Dose
Estimates
Direction
Both
Both
Both
Magnitude
Low
Low-
Moderate
Low
Knowledge-
hi 9 GO
UctOC
Uncertainty
Low
Low-
Moderate
Low
Comments
Comprehensive and subject to quality control.
Differences in 2000 data versus modeled years
(2006-2010) are likely small when estimating
percent of population exposed. Direct comparison of
exposures estimated using 2000 and 2010 Census
data in seven study areas indicate small differences,
though 6 of 7 indicated that exposures estimated
using the 2000 Census data were greater than that
estimated using the 2010 Census data.
Comprehensive and subject to quality control.
Significantly increased number of diaries used to
estimate exposure from prior review and 1st draft
HREA for this review (Table 5-3). Thoroughly
evaluated trends and patterns in historical data - no
major issues noted with use of historical data to
represent current patterns (Figures 5G-1 and 5G-2).
Compared outdoor participation and time with ATUS
data base - CHAD participation is higher than
ATUS, likely due to ATUS survey methods. Activity
data for asthmatics generally similar to non-
asthmatics (Tables 5G2-to 5G-5). There is little
indication of regional differences in time spent
outdoors though sample size may be a limiting factor
in drawing significant conclusions. Remaining
uncertainty exists with other influential factors that
cannot be accounted for (e.g., SES, region/local
participation in outdoor events and associated
amount of time).
Comprehensive and subject to quality control, few
missing values. Limited application in selecting
CHAD diaries and AERs. However, while using five
years of varying meteorological conditions, the
2006-2010 MET data set may not reflect the full
suite of conditions that could exist in future air
quality scenarios.
Is rating
appropriate for
current APEX OT,
exposure
assessment .
Yes. No further
characterization
needed.
Yes. Newly
evaluated.
Yes. No further
characterization
needed.
5-75
-------
Sources of Uncertainty
Category
APEX:
Microenvironmental
Concentrations
Element
Poverty Status (US
Census) Weighted
Asthma Prevalence
(CDC)
Outdoor Near-Road
and Vehicular:
Proximity Factors
Indoor: Near-Road
Indoor: Air Exchange
Rates
Historical Uncertainty Characterization
Influence of
Uncertainty on
Exposure/Intake Dose
Estimates
Direction
Both
Both
Over
Both
Magnitude
Low
Low
Low
Low
Knowledge-
hi 9 GO
UctOC
Uncertainty
Low-
Moderate
1 n\A/-
\—\J Vv
Modsrsts
Low
Moderate
Comments
Data used are from a peer-reviewed quality
controlled source. Application accounts for variability
in most important influential variables (age, sex,
region, poverty) though possible that variability in
microscale prevalence not entirely represented.
Uncertainty in mean value used approximated as 15
percentage points. See Figure 10 and Table 7 of
Langstaff (2007). May be of greater importance in
certain study areas or under varying conditions,
though even with this mean difference, in-vehicle
penetration/decay decreases exposures and hence
importance of in-vehicle microenvironments.
Expected reduction in Os for people residing near
roads not modeled here, but when included, there is
a small reduction (-3%) in the number of people
experiencing exposure above benchmark levels
(Langstaff, 2007).
Uncertainty due to random sampling variation via
bootstrap distribution analysis indicated the AER GM
and GSD uncertainty for a given study area tends
range to at most from fitted ±1.0 GM and ± 0.5 GSD
hr1. Non-representativeness remains an important
issue as city-to-city variability can be wide ranging
(GM/GSD pairs can vary by factors of 2-3) and data
available for city-specific evaluation are limited
(Langstaff, 2007). Also, indoor exposures are
estimated as not important to daily maximum 8-hr
average Os exposure.
Is rating
appropriate for
current APEX Os
exposure
assessment .
Yes. Newly
identified. Could
possibly use further
characterization,
though typically
available local
prevalence rates
are not well
stratified by
influential variables.
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
5-76
-------
Sources of Uncertainty
Category
Element
Indoor: A/C Prevalence
(AHS)
Indoor: Removal Rate
Vehicular: Penetration
Factors
Historical Uncertainty Characterization
Influence of
Uncertainty on
Exposure/Intake Dose
Estimates
Direction
Both
Both
Both
Magnitude
Low
Low
Low
Knowledge-
base
Uncertainty
Low
Low
Moderate
Comments
Comprehensive and subject to quality control,
estimated 95th percentile confidence bounds range
from a few to just over ten percentage points, though
some urban areas use older year data (Table 9 of
Langstaff, 2007). Note, variable indicates
presence/absence not actual use. Also, indoor
exposures are estimated here as limited in
importance to daily maximum 8-hr average
exposures and sensitivity analyses in NO2 HREA
(in-vehicle was most influential exposure ME)
concluded indoor prevalence variable was of limited
importance.
Greatest uncertainty in the input distribution
regarded representativeness, though estimated as
unbiased but correct to within 10% (Langstaff,
2007).
Input distribution is from an older measurement
study though consistent with recent, albeit limited
data.
Is rating
appropriate for
current APEX OT,
exposure
assessment?
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
5-77
-------
Sources of Uncertainty
Category
APEX: Simulated
Activity Profiles
Element
Longitudinal Profiles
Commuting
At-Risk Population and
Lifestages
Historical Uncertainty Characterization
Influence of
Uncertainty on
Exposure/Intake Dose
Estimates
Direction
Under
Both
Both
Magnitude
Low-
Moderate
Low
Low
Knowledge-
hi 9 GO
UctOC
Uncertainty
Moderate
Moderate
Low-
Moderate
Comments
Depending on the longitudinal profile method
selected, the number of people experiencing
multiple exposure events at or above a selected
level could differ by about 15 to 50% (see Appendix
B, Attachment 4 of NO2 REA). Long-term diary
profiles (i.e., monthly, annual) do not exist for a
population, limiting the evaluation.
The general population-based modeling approach
used for main body HREA results does not assign
rigid schedules, for example explicitly representing a
5-day work week for employed people. However,
when considering such scheduling (e.g., outdoor
workers or all children spending entire summer
season not in-school), estimated exposures are
greater than when not considering rigid
weekly/seasonal schedules. For our hypothetical
outdoor worker scenario, the number of multiday
exposures at or above benchmark levels was
primarily affected (though mainly the 60 ppb level,
Figure 5-14), while both percent of children
experiencing single and multiday exposures were
increased by about 30% when simulating a rigid
schedule (Figure 5-15).
New method used in this assessment is designed to
link Census commute distances with CHAD vehicle
drive times. Considered an improvement over the
former approach that did not match distance and
time. While vehicle time accounted for through diary
selection, not rigidly scheduled. However, In-vehicle
exposures are not important drivers for people
exceeding benchmark levels (section 5.4.2).
An updated evaluation shows activity patterns of
asthmatics are similar to that of non-asthmatics
(section 5.4.1, Tables 5G-2 to 5G-5).
Is rating
appropriate for
current APEX OT,
exposure
assessment .
Yes. Newly
evaluated.
Yes. Newly
evaluated.
Yes. Newly
evaluated.
5-78
-------
Sources of Uncertainty
Category
APEX: Physiological
Processes
Exposure Benchmark
Level
Element
Body Mass (NHANES)
NVO2max
RMR
METS distributions
Ventilation rates
EVR characterization of
moderate or greater
exertion
Historical Uncertainty Characterization
Influence of
Uncertainty on
Exposure/Intake Dose
Estimates
Direction
Unknown
Unknown
Unknown
Over
Over
Over
Magnitude
Low
Low
Low
Low-
Moderate
Low-
Moderate
Moderate
Knowledge-
base
Uncertainty
Low
Low
Low
Low-
Moderate
Low-
Moderate
Moderate
Comments
Comprehensive and subject to quality control,
though older (1999-2004) than current simulated
population, possible small regional variation is not
represented by national data.
Upper bound control for unrealistic activity levels
rarely used by model, thus likely not very influential.
Approach from older literature (Schofield, 1985),
used in ventilation equation. Note ventilation rate
estimates are reasonable.
APEX estimated daily mean METs range from about
0.1 to 0.2 units (between about 5-10%) higher than
independent literature reported values (Table 15 of
Langstaff, 2007). However, shorter-term values are
of greater importance in this assessment.
APEX estimated daily ventilation rates can be
greater (2-3 m3/day) than literature reported
measurement values (Table 25 of Langstaff, 2007),
though if accounting for measurement bias this
minimizes the discrepancy (Graham and McCurdy,
2005; see Figure 5-23 and Figure 5-24). Also, a
shorter-term comparison (hours rather than daily),
while more informative, cannot be performed due to
lack of data.
Given that the EVR serves as a cut point for
selecting individuals performing moderate or greater
exertion activities and is a lower bound value (~5th
percentile), the simulated number of people
achieving this level of exercise could be
overestimated.
Is rating
appropriate for
current APEX OT,
exposure
assessment?
Yes. No further
characterization
Yes. No further
characterization
needed.
Yes. Newly
identified. Could be
further
characterized.
Yes. Newly
identified. Could be
further
characterized if
short-term METS
available.
Yes. Could be
further
characterized if
minute or hourly
ventilation rate data
were available.
Yes. Newly
identified. Could be
further
characterized.
5-79
-------
5.6 KEY OBSERVATIONS
Four additional tables are provided to additionally summarize the exposure results across
all study areas and years of air quality data. Table 5-11 contains the percent of all school-age
children experiencing at least one exposure at or above the three exposure benchmark levels,
while Table 5-12 contains the percent of all school-age children experiencing at least two
exposures at or above the three exposure benchmark levels, with both tables considering results
associated with each of the adjusted air quality scenarios. Two descriptive statistics are provided
in each of these two tables using the exposure results for each urban study area: the mean percent
of persons exposed in each study area per Os season, averaged across the 5 years simulated;32
and the maximum percent of persons exposed in each study area for an Os season, capturing the
year having the highest ambient concentration where adjusted air quality met a particular
standard level within the 3-year averaging period.33 Two summary tables are provided to capture
the mean and maximum number of people experiencing at least one daily maximum 8-hr
exposure at or above 60 ppb (Table 5-13) and 70 ppb (Table 5-14) per Os season considering the
2006-2010 adjusted air quality. The number (mean and maximum) of person-days34 for each of
the two benchmark levels are also included in these two summary tables.
And finally, Figure 5-25 illustrates the estimated mean and maximum percent of all
school-age children exposed for each study area when considering the 60 ppb exposure
benchmark and adjusted air quality scenarios, and using the data provided in Table 5-11 and
Table 5-12.
Presented below are key observations resulting from the Os exposure analysis:
• General: The estimated percent of any study group exposed at least one time at or above
the selected exposure benchmark levels were highest considering the base air quality
though the percent of people exposed varied by study area, year, and benchmark level
(Appendix 5F). Very few people within any exposure study group, any study area, and
any year experienced any benchmark exceedances when considering an alternative 8-hr
standard level of 55 ppb (all are estimated to be < 0.3%, data not shown).
32 For most urban study areas, the sample size is five years based on years simulated and first averaging the two
2008 values generated for each of the two standard averaging periods. The number of years used in calculating the
mean for each study area could vary based on the available air quality (e.g., Chicago does not have 3 years
simulated for just meeting the existing standard during 2008-2010 period because air quality was below the
existing standard, thus the total sample size for this study area is 3).
33 In this analysis, the maximum year may not necessarily be the same for each urban study area, though all years
appropriately fall within the scenario of just meeting the particular standard level in that urban study area.
34 The person-day metric reflects the total number of days per O3 season members of the study group experienced an
exposure at or above a particular benchmark and when compared with the number of exposed study group
members can broadly indicate an the overall likelihood of multiple exposure days.
5-80
-------
Study Group: The percent of all school-age children exposed at or above the selected
benchmark levels across all study areas, years, and air quality scenarios were similar to
exposures for asthmatic school-age children (e.g., Figure 5-5 and Figure 5-6,
respectively) with both of these study groups having consistently higher percent of
persons exposed than that estimated for asthmatic adults and all older adults (Figure 5-7
and Figure 5-8, respectively), generally by about a factor of three or more. The percent of
all older adults experiencing exposures at or above any benchmark level tended to be
only a few percentage points or less when compared with corresponding benchmark
exceedances for asthmatic adults.
80 ppb Exposure Benchmark: In general, less than 1% of any study group, including all
school-age children and any urban study area, was exposed at least once at or above the
highest exposure benchmark, 80 ppb, when considering the existing standard air quality
scenario (Table 5-11). When considering an 8-hr standard level of 70 ppb, < 0.2% of any
study group and any study area was exposed at least once at or above that same
benchmark. When considering an 8-hr standard level of 65 ppb, no simulated individuals
experienced a daily maximum 8-hr exposure at or above 80 ppb.
70 ppb Exposure Benchmark: Less than 10% of any study group, including all school-
age children and any study area, was exposed at least once at or above an exposure
benchmark of 70 ppb, when considering the existing standard air quality scenario (Table
5-11). On average, approximately 360,000 school-age children across all 15 urban study
areas combined would experience at least one 8-hr exposure at or above 70 ppb (Table
5-14) for that same scenario. In considering a worst-case year of air quality in each urban
study area however, the number of school-age children exposed could be as high as
420,000 totaled across the 15 study areas. When considering an 8-hr standard level of 70
ppb, < 3.5% of any study group and in any study area was exposed at least once at or
above the 70 ppb benchmark. For all school-age children, this equates to, on average, just
less than 100,000 exposed in all 15 urban study areas at least once at or above the 70 ppb
benchmark, even considering the worst-case air quality year simulated in each of the 15
study areas (Table 5-14). An 8-hr standard level of 65 ppb is estimated to reduce the
percent of persons at or above an exposure benchmark of 70 ppb to < 0.5% of any study
group and in any study area, an air quality scenario leading to at most, approximately
15,000 school-age children exposed in all 15 urban study areas, even considering the
worst-case air quality year simulated in each of the 15 study areas (Table 5-14).
60 ppb Exposure Benchmark: In general, no more than 26% of any study group in any
study area was exposed at least once at or above the lowest exposure benchmark, 60 ppb,
when considering the existing standard air quality scenario (Table 5-11, Figure 5-25). On
5-81
-------
average, approximately 2.3 million school-age children are exposed at least once per Os
season to the 60 ppb benchmark when considering all 15 urban study areas combined and
air quality adjusted to just meet the existing standard (Table 5-13). In considering a
worst-case year of air quality in each urban study area however, the number of school-
age children experiencing such exposures could be as high as 4.0 million totaled across
the 15 study areas. When considering a standard level of 70 ppb, < 20% of any study
group in any study area was exposed at least once at or above that same benchmark. On
average about 1.2 million school-age children experience at least one 8-hr exposure to 60
ppb per Os season considering air quality adjusted to meet a 70 ppb standard level,
though during a high ambient concentration year within the standard averaging period,
the number of school-age children exposed per Os season to this same benchmark could
be as high as 1.7 million totaled across the 15 urban study areas (Table 5-13). An 8-hr
standard level of 65 ppbr is estimated to reduce the percent of persons at or above an
exposure benchmark of 60 ppb to < 10% of any study group and study area.
Multi-day Benchmark Exceedances: When considering air quality adjusted to just meet
the existing standard, multi-day exposure benchmark exceedances are largely limited to
two or more exceedances of the 60 ppb exposure benchmark, all occurring for < 15% of
any study group in any study area (e.g., Table 5-12, Figure 5-9). There were no people
estimated to experience any multi-day exposures at or above 80 ppb for any study group
in any study area, while < 2.2% of persons were estimated to experience two or more
daily maximum 8-hr exposures at or above 70 ppb, each considering any adjusted air
quality scenario. When the number of people exposed and the number of person-days at
or above the 70 ppb benchmark are very similar in value, this also reflects that finding of
limited occurrences of multi-day benchmark level exceedances (Table 5-14). For
example, we estimated 94,000 school-age children could experience at least one daily
maximum 8-hr exposure at or above 70 ppb when considering an 8-hr ambient standard
of 70 ppb, while approximately 99,000 school-age children person-days had an
exceedance of the 70 ppb exposure benchmark (Table 5-14), indicating that on average
most simulated individuals experienced only a single day at or above the exposure of
concern (i.e., the person to person-day ratio is 1.05).
Targeted Input Data Evaluations: Frequent occurrence and large amounts of afternoon
time spent outdoors, along with high ambient Os concentrations are the most influential
factors when considering the highest exposed individuals. There is no apparent temporal
trend in the amount of outdoor time or participation in outdoor events when comparing
historical CHAD diaries (1980s studies) to recently collected diary data (2000s studies);
regardless, the majority of CHAD data, in particular those from school-age children, are
5-82
-------
from studies conducted since 2000. Use of activity pattern data from non-asthmatic
individuals to represent asthmatics appears reasonably justified based on an evaluation
indicating their having similar outdoor time expenditure and attaining similar outdoor
exertion levels. Afternoon time spent outdoors does not differ when considering the
CHAD data set stratified by four U.S. Census regions, though the number of diaries
available for the analysis could be a limiting factor in this lack of an observed difference.
APEX estimated daily exposures in are somewhat comparable to personal exposure
measurements; however, both over- and under- estimations occurred to varying degrees
(Figure 5-18; Figure 5-21). APEX estimated ventilation rates were comparable to
literature provided estimates, particularly those of school-age children (Figure 5-24).
Targeted Exposure Scenarios: When considering a modeling approach that more
rigidly schedules longitudinal time location activity patterns compared with the standard
longitudinal approach used by APEX, a greater percent of persons experience at least one
or more exposures at or above benchmark levels. For example, an APEX model
simulation using only summer time (no school) CHAD diary days for non-working
school-age children generated approximately 30% more persons at or above exposure
benchmark levels compared with exposures estimated using our population-based
modeling approach (Figure 5-13). When accounting for a fraction of the population to
avert in response to a bad air quality day, approximately 1-2 percentage point fewer
persons experienced exposures at or above benchmark levels compared with exposures
estimated using our population based modeling approach (Figure 5-15).
5-83
-------
Table 5-11. Mean and Maximum Percent of all School-age Children Estimated to
Experience at Least One Daily Maximum 8-hr Average Os Exposure at or above Selected
Health Benchmark Levels while at Moderate or Greater Exertion.
Study Area
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
8-hr
Standard
Level
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
Percent of All School-Age Children Experiencing At Least
One Exposure at or above Selected 8-hr Benchmark Level1
60 ppb
mean
14.8
7.5
2.9
0.6
12.2
7.1
3.0
0.6
13.8
9.0
3.4
0.7
13.7
9.2
4.2
1.2
10.2
4.2
1.1
0.1
12.9
7.5
3.0
0.8
17.0
10.2
3.8
0.2
14.1
7.3
2.9
0.3
11.4
6.6
2.7
0.3
9.5
4.4
1.1
0
10.9
3.3
0
na
13.8
7.1
2.4
0.6
max
19.3
10.8
4.8
1.2
19.0
11.8
5.4
1.2
21.9
15.7
6.7
1.7
24.7
16.0
8.1
2.2
18.0
9.3
3.0
0.4
22.9
16.0
7.6
1.9
25.6
18.9
9.5
0.5
19.1
10.3
4.6
0.7
17.8
11.9
5.7
0.7
10.2
5.0
1.5
0.2
19.0
6.6
0.1
na
20.5
11.8
4.6
1.7
70
mean
2.8
0.7
0.2
0
2.0
0.7
0.2
0
2.8
1.2
0.2
0
3.2
1.0
0.2
0
1.4
0.3
0.1
0
1.9
0.6
0.1
0
1.7
0.5
0.1
0
2.4
0.5
0.1
0
2.3
0.8
0.1
0
0.6
0.1
0
0
1.6
0.2
0
na
2.1
0.6
0.1
0
ppb
max
4.4
1.4
0.5
0.1
4.0
1.2
0.3
0
6.6
3.2
0.5
0
7.5
2.7
0.4
0.1
3.7
0.9
0.2
0
4.5
1.5
0.3
0.1
4.1
1.7
0.4
0
4.2
0.9
0.2
0
5.5
2.1
0.4
0
1.0
0.2
0
0
3.7
0.5
0
na
4.2
1.5
0.3
0.1
80 ppb
mean
0.3
0.1
0
0
0.2
0.1
0
0
0.3
0.1
0
0
0.2
0
0
0
0.1
0
0
0
0.1
0
0
0
0.1
0
0
0
0.1
0
0
0
0.3
0
0
0
0
0
0
0
0.1
0
0
na
0.2
0
0
0
max
0.7
0.2
0
0
0.4
0.1
0
0
1.0
0.2
0
0
0.7
0.1
0
0
0.2
0
0
0
0.3
0.1
0
0
0.5
0.1
0
0
0.2
0
0
0
0.7
0.1
0
0
0.1
0
0
0
0.3
0
0
na
0.4
0.1
0
0
5-84
-------
Study Area
Sacramento
St. Louis
Washington
8-hr
Standard
Level
75
70
65
60
75
70
65
60
75
70
65
60
Percent of All School-Age Children Experiencing At Least
One Exposure at or above Selected 8-hr Benchmark Level1
60 ppb
mean
10.3
5.8
2.7
0.4
16.3
10.2
3.9
0.6
13.2
6.6
2.3
0.3
max
16.5
10.0
4.7
0.8
25.8
16.9
7.3
1.5
23.4
12.5
5.0
0.6
70
mean
1.6
0.4
0.1
0
3.3
1.0
0.1
0
2.4
0.6
0.1
0
ppb
max
2.7
0.9
0.2
0
8.1
2.7
0.4
0
6.0
1.4
0.2
0
80 ppb
mean
0.1
0
0
0
0.3
0.1
0
0
0.2
0
0
0
max
0.2
0
0
0
1.1
0.2
0
0
0.8
0.1
0
0
1 The mean is the arithmetic average of the estimated percent of all school-age children exposed across 2006-2010
year air quality; max is the highest estimated percent of all school-age children exposed in a year.
na - we were unable to simulate just meeting a standard level of 60 ppb in the New York study area.
Table 5-12. Mean and Maximum Percent of All School-age Children Estimated to
Experience at Least Two Daily Maximum 8-hr Average Os Exposures at or above Selected
Health Benchmark Levels while at Moderate or Greater Exertion.
Study Area
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
8-hr
Standard
Level
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
Percent of All School-Age Children Experiencing At Least
Two Exposures at or above Selected 8-hr Benchmark Level1
60 ppb
mean
6.0
2.1
0.4
0
4.6
1.8
0.4
0
4.5
2.2
0.4
0
5.3
2.5
0.8
0.2
3.1
0.9
0.1
0
4.8
2.2
0.5
0
max
8.9
3.3
0.8
0.1
8.4
3.7
0.9
0.1
9.7
5.5
1.1
0.1
11.6
5.7
1.8
0.3
7.5
2.6
0.5
0
12.2
7.1
2.0
0.2
70
mean
0.4
0
0
0
0.2
0
0
0
0.3
0.1
0
0
0.5
0.1
0
0
0.1
0
0
0
0.2
0
0
0
ppb
max
0.7
0.1
0
0
0.5
0.1
0
0
1.1
0.4
0
0
1.3
0.2
0
0
0.5
0
0
0
0.8
0.1
0
0
80 ppb
mean
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
max
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
5-85
-------
Study Area
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
8-hr
Standard
Level
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
Percent of All School-Age Children Experiencing At Least
Two Exposures at or above Selected 8-hr Benchmark Level1
60 ppb
mean
7.6
3.5
0.7
0
5.0
1.9
0.4
0
3.8
1.5
0.3
0
4.1
1.6
0.3
0
3.4
0.5
0
na
5.0
1.7
0.3
0
3.7
1.5
0.4
0
7.0
3.2
0.7
0
5.5
2.0
0.4
0
max
14.4
9.2
2.8
0
8.6
3.6
1.1
0
6.3
2.9
0.7
0
4.5
1.8
0.3
0
8.0
1.4
0
na
8.7
3.3
0.6
0.1
7.4
3.4
0.9
0.1
13.8
7.0
2.0
0.1
12.5
5.0
1.2
0
70
mean
0.2
0
0
0
0.3
0
0
0
0.2
0
0
0
0.1
0
0
0
0.1
0
0
na
0.2
0
0
0
0.2
0
0
0
0.6
0.1
0
0
0.4
0
0
0
ppb
max
0.4
0.1
0
0
0.8
0.1
0
0
0.6
0.1
0
0
0.1
0
0
0
0.4
0
0
na
0.5
0.1
0
0
0.5
0.1
0
0
2.2
0.3
0
0
1.4
0.1
0
0
80 ppb
mean
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
na
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
max
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
na
0
0
0
0
0
0
0
0
0.1
0
0
0
0
0
0
0
1 The mean is the arithmetic average of the estimated percent of all school-age children exposed across 2006-2010
year air quality; max is the highest estimated percent of all school-age children exposed in a year.
na - we were unable to simulate just meeting a standard level of 60 ppb in the New York study area.
5-86
-------
Table 5-13. Mean and Maximum Number of People and Person-days with at Least One
Daily Maximum 8-hr Average Os Exposure at or above 60 ppb while at Moderate or
Greater Exertion for All Exposure Study Groups, All 15 Urban Study Areas Combined.
Study Group
(Total Simulated
Population)
All School-age
Children
(19,049,557)
Asthmatic
School-age
Children
(1,992,762)
Asthmatic Adults
(4,993,077)
Older Adults
(35,174,540)
8-hr
Standard
Level
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
Number of People or Person-days with at Least One Daily
Maximum 8-hr Average Exposure at or above 60 ppb
Mean per Os Season
People
2,316,000
1,176,000
392,000
70,000
246,000
126,000
42,000
7,700
180,000
83,000
25,000
4,100
282,000
129,000
38,000
6,800
Person-days
4,043,000
1,707,000
483,000
76,000
432,000
184,000
52,000
8,600
258,000
107,000
29,000
4,400
414,000
163,000
43,000
7,100
Maximum per Os Season
People
3,588,000
1,988,000
762,000
156,000
385,000
214,000
81,000
17,000
295,000
145,000
47,000
9,000
497,000
246,000
83,000
16,000
Person-days
7,187,000
3,178,000
986,000
171,000
779,000
345,000
108,000
20,000
446,000
192,000
55,000
10,000
806,000
334,000
97,000
17,000
Table 5-14. Mean and Maximum Number of People and Person-days with at Least One
Daily Maximum 8-hr Average Os Exposure at or above 70 ppb while at Moderate or
Greater Exertion for All Exposure Study Groups, All 15 Urban Study Areas Combined.
Study Group
(Total Simulated
Population)
All School-age
Children
(19,049,557)
Asthmatic
School-age
Children
(1,992,762)
Asthmatic Adults
(4,993,077)
Older Adults
(35,174,540)
8-hr
Standard
Level
75
70
65
60
75
70
65
60
75
70
65
60
75
70
65
60
Number of People or Person-days with at Least One Daily
Maximum 8-hr Average Exposure at or above 70 ppb
Mean per Os Season
People
362,000
94,000
14,000
1,400
40,000
10,000
1,700
200
22,000
5,100
700
100
39,000
9,700
1,400
100
Person-days
418,000
99,000
15,000
1,400
46,000
11,000
1,700
200
24,000
5,300
700
100
43,000
10,000
1,400
100
Maximum per Os Season
People
792,000
236,000
41,000
4,500
88,000
27,000
4,500
500
46,000
12,000
1,900
200
96,000
26,000
3,800
400
Person-days
950,000
256,000
43,000
4,500
108,000
29,000
4,600
500
51,000
13,000
1,900
200
107,000
27,000
3,900
400
5-87
-------
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
0°
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
I
i i
!
1 1 1
1
1
1
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
4 2% 4% 6% 8% 10% 1 2% 14% 1 6% 1 8% QC
Percent of All School-Age Children with at Least One 8-hr Daily Max Exposure >=60ppb
standard level (ppb1) r I 60 i i 65 I 1 70 i i 75
1
| |
1
1
1 1
0% 1 % 2% 3% 4% 5% 6% 7% 8°
Percent of All School-Age Children with at Least Two 8-hr Daily Max Exposure >= 60 ppb
standard level (ppb) I I 60 L I 65 I 1 70 i 1 75
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
•o 0°
1, 1
1
1
I
I | ] |
1
1
1
1 1 1
•i 2% 4% 6% 8% 10% 12% 14% 16% 18% 20% 22% 24% 26°
Percent of All School-Age Children with at Least One 8-hr Daily Max Exposure >= 60 ppb
standard level (ppb) 1 1 60 1 1 65 1 1 70 1 ;'"
1
1 1
1 1
1 1
1
1
1
1 1
| ]
4 1% 2% 3% 4% 5% 6% 7% 8% 9% 10% 11% 12% 13% 14% 15°,
Percent of All School-Age Children with at Least Two 8-hr Daily Max Exposure >= 60 ppb
standard level ('ppb) I i 60 I i 65 r 1 70 I I 75
Figure 5-25. Incremental Decreases in Percent of All School-age Children Experiencing Daily Maximum 8-hr Os Exposures at
or above 60 ppb, with Increasing Stringency in Adjusted 2006-2010 Air Quality Standard Levels. Average percent (left panels),
maximum percent (right panels), at least one exposure (top panels), at least two exposures (bottom panels) per Os season.
5-88
-------
5.7 REFERENCES
Ainsworth, B.E.; W.L. Haskell; A.S. Leon; D.R. Jacobs Jr.; H.J. Montoye; J.F. Sallis; R.S.
Paffenbarger Jr. 1993. Compendium of physical activities: classification of energy costs of
human physical activities. Medicine & Science in Sports Exercise. 25:71-80.
Arcus-Arth, A. and J. Blaisdell. 2007. Statistical distributions of daily breathing rates for narrow
age groups of infants and children. Risk Analysis. 27(1):97-110.
Bresnahan, B.W.; M. Dickie and S. Gerking. 1997. Averting behavior and air pollution. Land
Economics. 73:340-357.
Brochu, P.; J.F. Ducre-Robitaille and J. Brodeur. 2006a. Physiological daily inhalation rates for
free-living individuals aged 1 month to 96 years, using data from doubly labeled water
measurements: a proposal for air quality criteria, standard calculations and health risk
assessment. Human and Ecological Risk Assessment. 12:675-701.
Brochu, P.; J.F. Ducre-Robitaille and J. Brodeur. 2006b. Supplemental material for physiological
daily inhalation rates for free-living individuals aged 1 month to 96 years, using data from
doubly labeled water measurements: a proposal for air quality criteria, standard
calculations and health risk assessment. Human and Ecological Risk Assessment. 12:1-12.
Bytnerowicz A.; M.J. Arbaugh; M. Fenn; R. Musselman; P. Padgett; T. Procter; and R. Cisneros.
2004. Improving Model Estimates of Smoke Contributions to Regional Haze Using Low -
Cost Sampler Systems.
.
Burmaster, D.E. 1998. LogNormal distributions for skin area as a function of body weight. Risk
Analysis. 18(l):27-32.
EPRI. 1988. A Study of Activity Patterns among a Group of Los Angeles Asthmatics. Research
Project 940-5. Electric Power Research Institute. Prepared by Roth Associates, November
1988.
EPRI. 1992. A Survey of Daily Asthmatic Activity Patterns in Cincinnati. TR-101396. Research
Project 940-05. Electric Power Research Institute. Prepared by Roth Associates. November
1992.
Ford, E.S.; G.W. Heath; D.M. Mannino and S.C. Redd. 2003. Leisure-time physical activity
patterns among U.S. adults with asthma. Chest. 124:432-437.
George, B.J. and T. McCurdy. 2009. Investigating the American time use survey from an
exposure modeling perspective. Journal of Exposure Science and Environmental
Epidemiology. 21:92-105.
Geyh, A.S.; J. Xue, H. Ozkaynak and J.D. Spengler. 2000. The Harvard Southern California
chronic ozone exposure study: assessing ozone exposure of grade-school-age children in
two Southern California communities. Environmental Health Perspectives. 108:265-270.
Glen, G.; L. Smith; K. Isaacs; T. McCurdy and J. Langstaff. 2008. A new method of longitudinal
diary assembly for human exposure modeling. Journal of Exposure Science and
Environmental Epidemiology. 18:299-311.
Graham, S.E. and T. McCurdy. 2004. Developing meaningful cohorts for human exposure
model s. Journal of Exposure Analysis and Environmental Epidemiology. 14:23-43.
5-89
-------
Graham, S.E. and T. McCurdy. 2005. Revised Ventilation Rate (VE) Equations for Use in
Inhalation-oriented Exposure Models. Washington, DC: EPA Office of Air and Radiation.
(EPA document number EPA/600/X-05/008, Appendix A).
.
Graham, S. 2009. Response to Peer-review Comments on Appendix A, prepared by S. Graham
(U.S. EPA). Response is Appendix D of US EPA (2009b). Research Triangle Park, NC:
EPA Office of Air and Radiation.
.
Graham, S. E. 2012. Comprehensive Review of Published Averting Behavior Studies and
Available Technical Documents. Memo to Bryan Hubbell, Group Leader, Risk and
Benefits Group, Office of Air Quality Planning and Standards. (EPA Docket # EPA-HQ-
OAR-2008-0699-0085).
Isaacs, K. and L. Smith. 2005. New Values for Physiological Parameters for the Exposure Model
Input File Physiology, text memorandum submitted to the U.S. Environmental Protection
Agency under EPA Contract EP-D-05-065. NERL WA 10. Alion Science and Technology.
Found in US EPA. (2009). Risk and Exposure Assessment to Support the Review of the SO2
Primary National Ambient Air Quality Standard. (EPA document number EPA-452/R-09-
007, August 2009).
.
Isaacs, K.; G. Glen; T. McCurdy and L. Smith. 2008. Modeling energy expenditure and oxygen
consumption in human exposure models: accounting for fatigue and EPOC. Journal of
Exposure Science and Environmental Epidemiology. 18:289-298.
Kansas Department of Health and Environment. 2006. Environmental Factors, Outdoor Air
Quality, and Activity Level: Results from 2005 Kansas Behavioral Risk Factor Surveillance
System. Office of Health Promotion, KDOH, available at:
.
Langstaff, I.E. 2007. Analysis of Uncertainty in Ozone Population Exposure Modeling., OAQPS
Staff Memorandum to Ozone NAAQS Review, January 31. Washington, DC: Office of Air
Radiation. (EPA docket number OAR-2005-0172). Available at:
.
Lee, K.; J. Vallarino; T. Dumyahn; H. Ozkaynak; and J.D. Spengler. 1999. Ozone decay rates in
residences. Journal of the Air and Waste Management Association. 49:1238-1244.
Lioy, PJ. 1990. The analysis of total human exposure for exposure assessment: a multi-
discipline science for examining human contact with contaminants. Environmental Science
and Technology. 24:938-945.
Liu, S.-J.; P. Koutrakis; H.H. Suh; J.D. Mulik; andR.M. Burton. 1993. Use of personal
measurements for ozone exposure assessment: a pilot study. Environmental Health
Perspectives. 101:318-324.
Mansfield, C.; F.R. Johnson and G. Van Houtven. 2006. The missing piece: averting behavior
for children's ozone exposures. Resource Energy Economics. 28:215-228.
Marino, A.J.; E.N. Fletcher; R.C. Whitaker and S.E. Anderson. 2012. Amount and
environmental predictors of outdoor playtime at home and school: a cross-sectional
analysis of a national sample of preschool-aged children attending head start. Health &
Place. 18:1224-1230.
5-90
-------
McCurdy, T. 2000. Conceptual basis for multi-route intake dose modeling using an energy
expenditure approach. Journal of Exposure Analysis and Environmental Epidemiology.
10:1-12.
McCurdy, T.; G. Glen; L. Smith and Y. Lakkadi. 2000. The National Exposure Research
Laboratory's consolidated human activity database. Journal of Exposure Analysis and
Environmental Epidemiology. 10:566-578.
McDermott, M.; R. Srivastava and S. Croskell. 2006. Awareness of and compliance with air
pollution advisories: a comparison of parents of asthmatics with other parents. Journal of
Asthma. 43:235-239.
McDonnell, W.F.; H.R. Kehrl; S. Abdul-Salaam; PJ. Ives; LJ. Folinsbee; R.B. Devlin; JJ.
O'Neil and D.H. Horstman. 1991. Respiratory response of humans exposed to low levels of
ozone for 6.6 hours. Archives of Environmental Health. 46(3): 145-150.
Meng, Q.; R. Williams and J.P. Pinto. (2012). Determinants of the associations between ambient
concentrations and personal exposures to PM2.5, NCh, and Os during DEARS. Atmospheric
Environment. 63:109-116.
Montoye, H.J.; H.C. G. Kemper; W.H.N. Saris andR.A. Washburn. \996.MeasuringPhysical
Activity and Energy Expenditure. Human Kinetics: Champaign, IL.
National Research Council. 1991. Human Exposure Assessment for Airborne Pollutants:
Advances and Opportunities. Washington, DC: National Academy of Sciences.
Neidell, M. 2010. Air quality warnings and outdoor activities: evidence from Southern California
using a regression discontinuity approach design. Journal of Epidemiology Community
Health. 64:921-926.
Rea, A; A. Vette; R. Williams; J. Suggs; L. Sheldon; R. Fortmann; and L. Wallace. 2003.
Exposure to Airborne Particles and Gases and Apportionment to Major Sources in the
Detroit Area (The Detroit Exposure Aerosol Research Study). Revised Study Plan after
Peer Review, September 9, 2003. .
Santuz, P.; E. Baraldi; M. Filippone and F. Zacchello. 1997. Exercise performance in children
with asthma: is it different from that of healthy controls? European Respiratory Journal.
10:1254-1260.
Sarnat, S.E.; B.A. Coull; J. Schwartz; D.R. Gold; and H.H. Suh. 2006. Factors affecting the
association between ambient concentrations and personal exposures to particles and gases.
Environmental Health Perspectives. 114(5):649-654.
Schofield, W.N. 1985. Predicting basal metabolic rate, new standards, and review of previous
work. Human Nutrition - Clinical Nutrition. 39C(S1):5-41.
Sexton, A. L. (2011). Responses to Air Quality Alerts: Do Americans Spend Less Time
Outdoors?.
.
Shamoo, D.A.; W.S. Linn; R.C. Peng; J.C. Solomon; T.L. Webb; J.D. Hackney and H. Hong.
1994. Time-activity patterns and diurnal variation of respiratory status in a panel of
asthmatics: implications for short-term air pollution effects." Journal of Exposure Analysis
and Environmental Epidemiology. 4(2): 133-148.
5-91
-------
U.S. Bureau of Labor Statistics. 2012. American Time Use Survey User's Guide. "Understanding
ATUS 2003 to 2011," August 2012. .
U.S. Census Bureau. 2007. 2000 Census of Population and Housing. Summary File 3 (SF3)
Technical Documentation, . Individual
SF3 files '30' (for income/poverty variables pctSO) for each state were downloaded from:
.
U.S. Department of Agriculture. 2000. Continuing Survey of Food Intake by Individuals (CSFII)
1994-96, 1998. Agricultural Research Service, CD-ROM, available at:
.
U.S. EPA. 1986. Air Quality Criteria for Ozone and Other Photochemical Oxidants. Research
Triangle Park, NC: Office of Health and Environmental Assessment, Environmental
Criteria and Assessment Office. (EPA document number EPA-600/8-84-020aF-eF).
Available from the National Technical Information Service, Springfield, VA. (NTIS
publication number PB87-142949).
U.S. EPA. 1996a. Review of National Ambient Air Quality Standards for Ozone: Assessment of
Scientific and Technical Information - OA QPS Staff Paper. Research Triangle Park, NC:
EPA Office of Air Quality Planning and Standards. (EPA document number EPA/452/R-
96-007). .
U.S. EPA. 1996b. Air Quality Criteria for Ozone and Related Photochemical Oxidants. Research
Triangle Park, NC: EPA Office of Research and Development, National Center for
Environmental Assessment. (EPA document number EPA/600/P-93/004aF-cF).
.
U.S. EPA. 2002. Consolidated Human Activities Database (CHAD) Users Guide.
.
U.S. EPA. 2005. Guidance on Selecting Age Groups for Monitoring and Assessing Childhood
Exposures to Environmental Contaminants. (EPA document number EPA/630/P-03/003F).
.
U.S. EPA. 2006. Air Quality Criteria for Ozone and Related Photochemical Oxidants (Final).
Research Triangle Park, NC: EPA National Center for Environmental Assessment. (EPA
document number EPA/600/R-05/004aF-cF).
.
U.S. EPA. 2007a. Review of National Ambient Air Quality Standards for Ozone: Policy
Assessment of Scientific and Technical Information - OA QPS Staff Paper. Research
Triangle Park, NC: EPA Office of Air Quality Planning and Standards. (EPA document
number EPA-452/R-07-007).
.
U.S. EPA. 2007b. Ozone Population Exposure Analysis for Selected Urban Areas. Research
Triangle Park, NC: EPA Office of Air Quality Planning and Standards.
.
U. S. EPA. 2008. Risk and Exposure Assessment to Support the Review of the NO2 Primary
National Ambient Air Quality Standard. Washington, DC: EPA Office of Air and
Radiation. (EPA document number EPA-452/R-08-008aO, November).
.
5-92
-------
U.S. EPA. 2009a. Risk and Exposure Assessment to Support the Review of the SO 2 Primary
National Ambient Air Quality Standard. (EPA document number EPA-452/R-09-007,
August).
.
U.S. EPA. 2009b. Metabolically Derived Human Ventilation Rates: A Revised Approach Based
Upon Oxygen Consumption Rates. Washington, DC: EPA National Center for
Environmental Assessment. (EPA document number EPA/600/R-06/129F).
.
U.S. EPA. 2010. Quantitative Risk and Exposure Assessment for Carbon Monoxide -Amended.
Research Triangle Park, NC: EPA Office of Air Quality Planning and Standards. (EPA
document number EPA-452/R-10-009, July).
.
U.S. EPA. 2011. Exposure Factors Handbook, 2011 edition. Washington, DC: EPA National
Center for Environmental Assessment. (EPA document number EPA/600/R-09/052F).
.
U.S. EPA. 2012a. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
Documentation (TRIM. Expo / APEX, Version 4.4) Volume I: User's Guide. Research
Triangle Park, NC: EPA Office of Air Quality Planning and Standards. (EPA document
number EPA-452/B-12-001a). .
U.S. EPA. 2012b. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
Documentation (TRIM. Expo / APEX, Version 4.4) Volume II: Technical Support
Document. Research Triangle Park, NC: EPA Office of Air Quality Planning and
Standards. (EPA document number EPA-452/B-12-001b).
.
U.S. EPA. 2013. Integrated Science Assessment of Ozone and Related Photochemical Oxidants.
Research Triangle Park, NC: EPA National Center for Environmental Assessment. (EPA
document number EPA 600/R-10/076F).
.
Winter, S.E.; R.A. Bilonick; J.A. Withum; V.B. Conrad; R.M. Statnick; S.T. Ebelt; H.H. Suh;
and P. Koutrakis. 2003. A Field Evaluation to Compare the Performance of Personal
Exposure Multi-Pollutant Samplers vs. Federal Method Monitors at a Central Ambient Air
Monitoring Station in Steubenville, Ohio. Presented at Air Quality IV Conference,
Arlington, VA September 23, 2003.
.
van Gent, R.; K. van der Ent; L.E.M. van Essen-Zandvliet; M.M. Rovers; J.L.L. Kimpen; G. de
Meer and P.H.C. Klijn. 2007. No difference in physical activity in (Un)diagnosed asthma
and healthy controls. PediatricPulmonology. 42:1018-1023.
Wen, X.J.; L. Balluz and A. Mokdad. 2009. Association between media alerts of air quality
index and change of outdoor activity among adult asthma in six states, BRFSS, 2005.
Journal of Community Health. 34:40-46.
Whitfield, R.; W. Biller; M. Jusko and J. Keisler. 1996. A Probabilistic Assessment of Health
Risks Associated with Short- and Long-Term Exposure to Tropospheric Ozone. Argonne,
IL: Argonne National Laboratory.
5-93
-------
World Health Organization. 2008. Uncertainty and Data Quality in Exposure Assessment. "Part
1: Guidance Document on Characterizing and Communicating Uncertainty in Exposure
Assessment."
-------
6 CHARACTERIZATION OF HEALTH RISKS BASED ON
CONTROLLED HUMAN EXPOSURE STUDIES
6.1 INTRODUCTION
This chapter presents information regarding the methods and results for a controlled
human exposure-based ozone (Os) health risk assessment that builds upon the methodology used
in the assessment conducted as part of the Os NAAQS review completed in 2008 and also
introduces a new method for estimating risk. In the previous review, EPA conducted a health risk
assessment that produced risk estimates for the number and percent of school-aged children,
asthmatic school-aged children, and the general population experiencing lung function
decrements associated with Os exposures for 12 urban study areas, where lung function is
measured as forced expiratory volume in one second (FEVi). That portion of the risk assessment
was based on exposure-response (E-R) relationships developed from analysis of data from
several controlled human exposure studies which were combined with population-level exposure
distributions developed for children and adults. Risk estimates for lung function decrements were
developed for recent air quality levels and for just meeting the existing daily maximum 8-hour
(8-hr) average standard and several alternative 8-hr standards. The methodological approach
followed in the last risk assessment and risk estimates resulting from that assessment are
described in the 2007 Staff Paper (U.S. EPA, 2007a).
The goals of the current Os risk assessment are to provide estimates of the number and
percentage of select study groups that would experience adverse respiratory effects associated
with recent Os levels and with meeting the existing and potential alternative Os standards in
specific urban study areas; and to develop a better understanding of the influence of various
inputs and assumptions on the risk estimates. The current assessment includes estimates of risks
of lung function decrements in school-aged children (ages 5 to 18), asthmatic school-aged
children, and the adult population (19 and above). We recognize that there are many sources of
uncertainty in the inputs and approach used in this portion of the health risk assessment which
make the specific estimates uncertain, however, we have sufficient confidence in the magnitude
and direction of the estimates provided by the assessment for it to serve as a useful input to
decisions on the adequacy of the Os standard and risk reductions associated with alternative
standards.
We are estimating lung function risk using two methodologies in this review. The
primary results are based on a new model that estimates FEVi responses for individuals
associated with short-term exposures to Os (McDonnell, Stewart, and Smith, 2007, 2010;
McDonnell et al., 2012). We refer to this model as the McDonnell-Stewart-Smith (MSS) model.
We also provide estimates following the methodology used in previous reviews which provides
6-1
-------
population level estimates of the percent and number of people at risk. We refer to this model as
the population E-R model used in previous reviews. Both of these models are implemented in the
air pollution exposure model APEX (U.S. EPA, 2012a,b). Following this introductory section,
this chapter discusses the scope of the controlled human exposure study based risk assessment,
describes the risk models, and provides key results from the assessment. The results of sensitivity
analyses are reported and key uncertainties are identified and summarized. More detailed
descriptions of several parts of the analyses are included in appendices that accompany this
Health Risk and Exposure Assessment (HREA).
6.1.1 Development of Approach for Current Risk Assessment
The lung function risk assessment described in this chapter builds upon the methodology
and lessons learned from the risk assessment work conducted for previous reviews (U.S. EPA,
1996, 2007a). The current risk assessment also is based on the information evaluated in the ISA
(U.S. EPA, 2013). The general approach used in the current risk assessment was described in the
Scope and Methods Plan for Health Risk and Exposure (U.S. EPA, 2011), that was released to
the CAS AC and general public in April 2011 for review and comment and which was the subject
of a consultation with the CASAC Cb Panel in May 2011. The first and second drafts of this
HREA were reviewed by CASAC in September 2012 and March 2014. The approach used in the
current risk assessment reflects consideration of the comments offered by CASAC members and
the public on the Scope and Methods Plan and the two draft HREAs.
Controlled human exposure studies involve volunteers, primarily healthy adults, who are
exposed while engaged in different exercise regimens to specified levels of Os under controlled
conditions for specified amounts of time. For the current health risk assessment, we are using
probabilistic E-R relationships based on analysis of individual data that describe the relationship
between measures of personal exposure to Cb and measures of lung function recorded in the
studies. Therefore, a risk assessment based on exposure-response relationships derived from
controlled human exposure study data requires estimates of personal exposure to ambient Os.
Because data on personal hourly exposures to Os of ambient origin are not available, estimates of
personal exposures to varying ambient concentrations are derived through exposure modeling, as
described in Chapter 5.
While the quantitative risk assessment based on controlled human exposure studies
addresses only lung function responses, it is important to note that other respiratory responses
have been found to be related to Os exposures in these types of studies, including increased lung
inflammation, increased respiratory symptoms, increased airway responsiveness, and impaired
host defenses. Sufficient information is not available to quantitatively model these other
endpoints. Section 6.2 of the ISA provides a discussion of these additional health endpoints
6-2
-------
which are an important part of the overall characterization of risks associated with ambient Os
exposures.
6.1.2 Comparison of Controlled Human Exposure- and Epidemiologic-based Risk
Assessments
In contrast to the exposure-response (E-R) relationships derived from controlled human
exposure studies, epidemiological studies provide estimated concentration-response (C-R)
relationships based on data collected in real world community settings. The assessment of health
risk based on epidemiological studies is the subject of Chapter 7. The characteristics that are
relevant to carrying out a risk assessment based on controlled human exposure studies versus one
based on epidemiology studies can be summarized as follows:
• The relevant controlled human exposure studies in the ISA provide data that can be used
to estimate E-R functions, and therefore a risk assessment based on these studies requires
as input (modeled) personal exposures to ambient Os. The relevant epidemiological
studies in the ISA provide C-R functions, and, therefore, a risk assessment based on these
studies requires as input (actual monitored or adjusted based on monitored) ambient Os
concentrations, and personal exposures are not required as inputs to the assessment.
• Epidemiological studies are carried out in specific real world locations (e.g., specific
urban areas). To minimize extrapolation uncertainty, a risk assessment based on
epidemiological studies is best performed in locations where the studies took place.
Controlled human exposure studies, carried out in laboratory settings, are generally not
specific to any particular real world location. A risk assessment based on controlled
human exposure studies can therefore appropriately be carried out for any location for
which there are adequate air quality and other data on which to base the modeling of
personal exposures.
• To derive estimates of risk from C-R relationships estimated in epidemiological studies,
it is usually necessary to have estimates of the baseline incidences of the health effects
involved. Such baseline incidence estimates are not needed in a controlled human
exposure studies-based risk assessment.
• Widely accepted ethical standards preclude the use of infants and young children in
controlled human exposure studies, leaving a number of the most susceptible lifestages
effectively unstudied under laboratory conditions. Similarly, other susceptible groups,
such as people with lung disease, are not often participants in controlled human exposure
studies. In contrast, epidemiology studies are able to include all lifestages and groups in a
population or focus on a particular at-risk group.
6-3
-------
6.2 SCOPE OF LUNG FUNCTION HEALTH RISK ASSESSMENT
The current controlled human exposure-based Os health risk assessment is one approach
used to estimate risks associated with exposure to ambient Os in a number of urban study areas
selected to illustrate the public health impacts of this pollutant. The short-term exposure related
health endpoints selected for this portion of the Os health risk assessment include those for which
the ISA concludes that the evidence as a whole supports the general conclusion that Cb, acting
alone and/or in combination with other components in the ambient air pollution mix is causal or
likely to be causally related to the endpoint.
In the 2007 Os NAAQS review, the controlled human exposure-based health risk
assessment involved developing risk estimates for lung function decrements (> 10, > 15, and
> 20% changes in FEVi) in school-aged children (ages 5 to 18 years old). The strong emphasis
on children reflects the finding of previous Os NAAQS reviews that children are an important at-
risk group. Due to the increased amount of time spent outdoors engaged in relatively high levels
of physical activity (which increases intake), school-aged children as a group are particularly at
risk for experiencing Os-related health effects.
Outdoor workers and other adults who engage in moderate exertion for prolonged
periods or heavy exertion for shorter periods during the day also are clearly at risk for
experiencing similar lung function responses when exposed to elevated ambient Os
concentrations. In this HREA, we focus the quantitative risk assessment for lung function
decrements on all and asthmatic school-aged children (ages 5-18), and the adult population (ages
19 and above).
For this assessment, lung function risks are estimated for 15 urban study areas, Atlanta,
Baltimore, Boston, Chicago, Cleveland, Dallas, Denver, Detroit, Houston, Los Angeles, New
York, Philadelphia, Sacramento, St. Louis, and Washington, DC.
6.2.1 Selection of Health Endpoints
The ISA identifies several responses to short-term Os exposure that have been evaluated
in controlled human exposure studies (U.S. EPA, 2013, sections 6.2.1.1, 6.2.2.1, 6.2.3.1, and
6.3.1). These include decreased inspiratory capacity; decreased forced vital capacity (FVC) and
forced expiratory volume in one second (FEVi); mild bronchoconstriction; rapid, shallow
breathing patterns during exercise; symptoms of cough and pain on deep inspiration (PDI);
increased airway responsiveness; and pulmonary inflammation. Such studies provide direct
evidence of relationships between short-term Os exposure and an array of respiratory-related
effects, however, there are only sufficient E-R data at different concentrations to develop
quantitative risk estimates for Os-related decrements in FEVi. Other responses to Os which may
be equally or more important than FEVi decrements (e.g., inflammation) do not necessarily
6-4
-------
correlate with FEVi responses (ISA, section 6.2.3.1) and this risk assessment is not able to
address these other responses.
As stated in the 2006 Criteria Document (Table 8-3, p.8-68) for adults with lung disease,
even moderate functional responses (e.g., FEVi decrements > 10% but < 20%) would likely
interfere with normal activities for many individuals, and would likely result in more frequent
medication use. In a letter to the Administrator, the CASAC Os Panel stated that '"Clinically
relevant' effects are decrements > 10%, a decrease in lung function considered clinically relevant
by the American Thoracic Society" (Samet, 2011, p.2). The CASAC Os Panel also stated that:
a 10% decrement in FEVi can lead to respiratory symptoms, especially in
individuals with pre-existing pulmonary or cardiac disease. For example,
people with chronic obstructive pulmonary disease have decreased ventilatory
reserve (i.e., decreased baseline FEVi) such that a > 10% decrement could
lead to moderate to severe respiratory symptoms (Samet, 2011, p. 7).
This is consistent with the most recent official statement of the American Thoracic Society on
what constitutes an adverse lung function health effect of air pollution:
The committee recommends that a small, transient loss of lung function, by
itself, should not automatically be designated as adverse. In drawing the
distinction between adverse and nonadverse reversible effects, this committee
recommended that reversible loss of lung function in combination with the
presence of symptoms should be considered adverse (ATS, 2000, p.672).
For this lung function risk assessment, a focus on the mid- to upper-end of the range of
moderate levels of functional responses and higher (FEVi decrements > 15%) is appropriate for
estimating potentially adverse lung function decrements in active healthy adults, while for people
with asthma or lung disease, a focus on moderate functional responses (FEVi decrements down
to 10%) may be appropriate.
6.2.2 Approach for Estimating Health Risk Based on Controlled Human Exposure
Studies
The major components of the health risk assessment based on data from controlled
human exposure studies are illustrated in Figure 3-3 in Chapter 3. As shown in this figure, under
this portion of the risk assessment, exposure estimates for a number of different air quality
scenarios (i.e., recent year of air quality, just meeting the existing 8-hr and alternative standards)
are combined with probabilistic E-R relationships derived from the controlled human exposure
studies to develop risk estimates associated with recent air quality and after simulating just
meeting the existing and alternative standards. The health effect included in this portion of the
risk assessment is lung function decrement, as measured by changes in FEVi. The population
6-5
-------
risk estimates for a given lung function decrement (e.g., > 15% reduction in FEVi) are estimates
of the expected number of people who will experience that lung function decrement, the number
of times that people experience repeated occurrences of given lung function decrements, and the
number of occurrences (person-days) of the given lung function decrement. The air quality and
exposure analysis components that are integral to this portion of the risk assessment are
discussed in Chapters 4 and 5.
We used two approaches to estimate health risk. As done for the risk assessment
conducted during the previous Os NAAQS review, a Bayesian Markov Chain Monte Carlo
approach was used to develop probabilistic E-R functions. These functions were then applied to
the APEX estimated population distribution of 8-hr maximum exposures for people at or above
moderate exertion (> 13 L/min-m2 body surface area) to estimate the number of people expected
to experience lung function decrements. The primary approach, based on the McDonnell-
Stewart-Smith (MSS) FEVi model, uses the time-series of Os exposure and corresponding
ventilation rates for each APEX simulated individual to estimate their personal time-series of
FEVi reductions, selecting the daily maximum reduction for each person. A key difference
between these approaches is that the previous method estimates a population distribution of
FEVi reductions, where the MSS model estimates FEVi reductions at the individual level. Each
of these approaches is discussed in detail below.
6.2.3 Controlled Human Exposure Studies
Modeling of risks of lung function decrements as a function of exposures to Os is based
on application of results from controlled human exposure studies. As discussed in Chapter 6 of
the ISA (U.S. EPA, 2013), there is a significant body of controlled human exposure studies
reporting lung function decrements and respiratory symptoms in adults associated with 1- to 8-hr
exposures to Os. In the ISA sections on controlled human exposure (Sections 6.2.1.1, 6.2.2.1,
6.2.3.1, and 6.3.1) over 140 references to human clinical studies are reported.
6.2.3.1 Lifestages
Consistent with the approach used in the previous Os NAAQS review and lacking a
significant body of controlled human exposure studies on children, we judge that it is reasonable
to estimate E-R relationships for lung function decrements associated with Os exposures in
children 5-18 years old based on data from young adult subjects (18-35 years old). As discussed
in the ISA (U.S. EPA, 2013), findings from clinical studies for children and summer camp field
studies of children 7-17 years old in at least six different locations in the U.S. and Canada found
lung function decrements in healthy children similar to those observed in healthy young adults
exposed to Os under controlled chamber conditions. There are fewer studies of young children
6-6
-------
than adolescents to draw upon, which may add to uncertainties in the modeling. Additional
uncertainties are likely introduced since the lungs and airways of children are developing, while
development is complete in adults (Dietert et al., 2000). The primary period of alveolar
development is from birth to around eight years of age, but there is evidence for continued
development through adolescence. The adult number of alveoli is reached by 2-3 years of age
and the size and surface area of the alveoli increase until after adolescence (Hislop, 2002;
Narayanan et al., 2012). Since it is unlikely that there will ever be a significant body of
controlled human exposure studies on children (for ethical reasons), it may be that the best way
to collect data on children's lung function response to Os is through summer camp studies, where
children wear personal continuous 5-minute Os monitors and have periodic lung function tests,
with some method for estimating continuous ventilation rates.
Lung function responses to Os exposure for adults older than 18 decrease with age until
around age 55, when responses are minimal. "Children, adolescents, and young adults appear, on
average, to have nearly equivalent spirometric responses to Os, but have greater responses than
middle-aged and older adults when similarly exposed to Os" (ISA p. 6-21). "In healthy
individuals, the fastest rate of decline in Os responsiveness appears between the ages of 18 and
35 years (Passannante et al., 1998; Seal et al., 1996), more so for females than males (Hazucha et
al., 2003). During the middle age period (35-55 years), Os sensitivity continues to decline, but at
a much lower rate. Beyond this age (>55 years), acute Os exposure elicits minimal spirometric
changes" (ISA p. 6-23).
6.2.3.2 Asthma
There have been several controlled human exposure studies of the effects of Os on
asthmatic subjects, going back to 1978 (Linn et al., 1978). In reference to these studies, the ISA
states that "[b]ased on studies reviewed in the 1996 and 2006 Os AQCDs, asthmatic subjects
appear to be at least as sensitive to acute effects of Os as healthy nonasthmatic subjects" (ISA p.
6-20). Studies published since the 2006 Os AQCD do not alter this conclusion (ISA, p. 6-20 to 6-
21). In the 2010 Os NAAQS proposal (75 FR 2969-2972), EPA describes the evidence that
people with asthma are as sensitive as, if not more sensitive than, normal subjects in manifesting
Os-induced pulmonary function decrements.
In reference to epidemiologic studies, the ISA states that "[t]he evidence supporting
associations between short-term increases in ambient Os concentration and increases in
respiratory symptoms in children with asthma is derived mostly from examination of 1-h max,
8-h max, or 8-h avg Os concentrations and a large body of single-region or single-city studies.
The few available U.S. multicity studies produced less consistent associations." (ISA, p. 6-101 to
6-102). "Although recent studies contributed mixed evidence, the collective body of evidence
6-7
-------
supports associations between increases in ambient Os concentration and increased asthma
medication use in children" (ISA, p. 6-109).
6.2.3.3 Ethnicity
There are two controlled human exposure studies that have assessed differences in lung
function responses comparing ethnic groups (ISA, p. 6-23 to 6-24). Both of these studies show
greater FEVi decrements in blacks than whites, however, epidemiologic studies were less
supportive of this difference in response. The data available are insufficient to quantify any
differences that might exist, due to the limited number of studies and a lack of consistency
between disciplines.
6.2.3.4 Body mass index
Some studies have found greater FEVi decrements to be associated with increasing BMI.
BMI was included in some of the models of McDonnell et al. (2012); however, the BMI terms
were found to be statistically insignificant, indicating that the effect of BMI on FEVi in the
presence of Os is likely to be small, within the range of BMIs of the subjects studied.
6.2.3.5 Outdoor workers
Although there are no controlled human exposure studies that have had specifically
outdoor workers as subjects, the studies are applicable to outdoor workers: the 6.6-hour
experimental protocol was intended to simulate the performance of heavy physical labor for a
full workday (ISA, p. 6-9).
6.2.3.6 Variability of responses
Responses to Os exposure are variable within the population, even within cohorts of
similar people (e.g., healthy young adult white males) (ISA, p. 6-16 to 6-20). Factors which
contribute to interindividual variability include health status, body mass index, age, sex,
race/ethnicity, and the intrinsic responsiveness of individuals. Other factors which contribute to
the variability of responses include the duration and concentration of Os exposure, the level of
exercise and breathing rate, attenuation due to repeated exposures, and co-exposures with other
pollutants. For specific individuals, lung function responses tend to be reproducible over a period
of several months.
6.2.3.7 Effects of altitude
The controlled human exposure studies drawn upon for this lung function risk assessment
were conducted at low altitudes (< 600 feet). It is not known whether it is more appropriate to
use a mass concentration (ug/m3) or a mixing ratio concentration (ppm) at higher altitudes where
6-8
-------
the barometric pressure is lower. The E-R relationships used in the HREA are based on mixing
ratio concentrations. If in fact mass concentrations are the appropriate exposure metric, then the
lung function responses in a high-altitude city such as Denver (5300 feet asl1) could be
overestimated by about 15%.
A known effect of altitude is that people must breathe more rapidly to maintain their
oxygen intake with increasing altitude (Bittar, 2002). This is not accounted for in the model of
ventilation rates in APEX.
6.2.4 The McDonnell-Stewart-Smith (MSS) Model
In this review, EPA is investigating the use of a new model that estimates FEVi
responses for individuals associated with short-term exposures to Os (McDonnell, Stewart, and
Smith, 2007; McDonnell, Stewart, and Smith, 2010; McDonnell et al., 2012). This is a
fundamentally different approach than the previous approach, for which the E-R function is at a
population level, not an individual level. This model was developed using controlled human
exposure data from studies using different exposure durations and different exertion levels and
breathing rates. These data were from 15 controlled human Os exposure studies that included
exposure of 541 volunteers (ages 18 to 352) on a total of 864 occasions. These data are described
in McDonnell et al. (1997). Schelegle et al. (2009) found that there appears to be a delay in
response when modeling FEVi decrements as a function of accumulated dose and estimated a
threshold associated with the delay. McDonnell et al. (2012) refit their 2010 model using data
from eight additional studies with 201 subjects and incorporating a threshold parameter into the
model. Their threshold parameter allows for modeling a delay in response until accumulated
dose (taking into account decreases over time according to first order reaction kinetics) reaches a
threshold value and is found by McDonnell et al. (2012) to slightly improve model fit. That latest
model is the model described here and is the model used in this risk assessment. The threshold is
not a concentration threshold and does not preclude responses at low concentration exposures.
Schelegle et al. (2012) have also developed a 2-compartment model for predicting FEVi
decrements (ISA, p. 6-15,16). Their model is similar to the MSS model in that it accounts for the
effects of accumulated dose coupled with an exponential decay and also has a threshold, below
which response is delayed. The primary difference between this model and the MSS model is
that in the Schelegle et al. model the net accumulated dose is multiplied by an individual's
responsiveness coefficient to obtain a predicted FEVi decrement, whereas in the MSS model the
FEVi decrement increases as a sigmoid-shaped function of the net accumulated dose. Also, the
1 The average height of the O3 monitors in the city of Denver.
2 The ages in these studies range from 18 years 1 month to 35 years 1 month.
6-9
-------
Schelegle et al. model's threshold is based on accumulated intake dose (the integral of
concentration x volume inhaled), where the MSS model's threshold is based on net accumulated
dose (taking into account the first order decay). A direct comparison of the results of these two
models has not been performed.
The MSS model is conceptually a two-compartment model. The accumulated amount of
exposure to Cb (exposure concentration times ventilation rate, loosely speaking a measure of
dose) is modeled in the first compartment and modified by an exponential decay factor to yield
an intermediate quantity X. The response (lung function decrement) of the individual to X is
modeled in the second compartment (Figure 6-1). The threshold parameter imposes the
constraint that there is no response while the value of X is below the threshold value.
compartment 1
compartment 2
C(t)V(t)
(dose)
-------
-o) + i^7(t )£6n _ e-/?5(ti-«m Equation (6-2)
0s
and V(h) denote the (constant) values of C(t) and Ff£) during the event from time to
to time h.
This model calculates the FEVi decrement due to Os exposure (compartment 2) as:
+ £2 (40eifc - A}] \ *„.., - -±4 + em Equation (6-3)
where Jijk = max{0, Xijk - $9}. $9 is a threshold parameter which allows X to increase up to the
threshold before the median response is allowed to exceed zero.
The variables in the above equations are defined as:
The indices i,j,k refer to the 7th subject at the/h time for the A*h experiment for that subject,
C(t) is the Os exposure concentration at time t (ppm) during the event,
V(t) = VE(t)IBSA is the ventilation rate normalized by body surface area at time t
(L/min-m2),
VE(t) is the expired minute volume at time t (L min"1),
BSA is the body surface area (m2),
t is the time (minutes),
to is the time at the start of the event, ti is the time at the end of the event,
Agdk is age in years of the 7th subject in the A*h study,
A is an age parameter (taken to be the approximate mean age of the clinical study subjects in
the McDonnell, Stewart, and Smith 2007 (A=25), 2010 (A=25), and 2012 (A=23.8) papers),
Ui is a subject-level random effect (between-individual variability not otherwise captured by
the model), and
Sijkis a variability term, which includes measurement error and within-individual variability
not otherwise captured by the model.
The values of the PS and the variances of the {Ui} and {£yk} were estimated from fits of
the model to the data (see McDonnell, et al. (2007, 2010, and 2012) for details). In APEX, values
of Ui and Sijkare drawn from Gaussian distributions with mean zero and variances var(U) and
var(e), constrained to be within ±2 standard deviations from the means (when samples are
outside of this range, they are discarded and resampled). The values of Ui are chosen once for
each individual and remain constant for individuals throughout the simulation. The By-hare
sampled daily for each individual. The best fit values (based on maximum likelihood) for these
parameters are listed in Table 6-1. The values in parentheses are standard errors of the estimates
(given here to two significant digits; the values in the papers are given to up to five significant
digits). Although some of the parameters are quite different in the three models in Table 6-1, the
6-11
-------
predictions of these three models are similar. The relative influences of the parameters are
discussed in Section 6.5.1.
Table 6-1. Estimated Parameters in the MSS Models.
Model
20071,
201 02
201 23
2012T4
P1
9.9047
(0.61)
9.8057
(0.74)
10.916
(0.84)
P2
-0.4106
(0.11)
-0.1907
(0.28)
-0.2104
(0.31)
P3
0.0164
(0.0030)
0.01839
(0.0051)
0.01506
(0.0033)
P4
46.9397
(7.3)
65.826
(12)
13.497
(4.7)
P5
0.003748
(0.00027)
0.003191
(0.00021)
0.003221
(0.00021)
P6
0.9123
(0.054)
0.8753
(0.086)
0.8839
(0.065)
P9
0
59.284
(10)
var(U)
0.835
(0.080)
0.9449
(0.083)
0.9373
(0.082)
var(E)
13.8279
(0.36)
17.120
(1.2)
17.0816
(1.2)
The values in parentheses are standard errors of the estimates.
1 McDonnell, Stewart, and Smith (2007). A = 25.
2 McDonnell, Stewart, and Smith (2010). A = 25.
3 McDonnell, et al. (2012). A = 23.8. No-threshold.
4 McDonnell, et al. (2012). A = 23.8. Threshold.
We are using this model to estimate lung function decrements for people ages 5 and
older. However this model was developed using only data from individuals aged 18 to 35 and the
age adjustment term [Pi + $2 (Agdjk - A)] in the numerator of equation 6-3 is not appropriate for
all ages. In addition to this age term, the effects of age are also taken into account through the
dependence of ventilation rate and body surface area on age. The APEX estimates of lung
function risk for different age groups are also influenced by the time spent outdoors and the
activities engaged in by those groups, which vary by age (see Appendix 6E).
Clinical studies data for children which could be used to fit the model for children are not
available at this time. In the absence of data, we are extending the model to ages 5 to 18 by
holding the age term constant at the age 18 level. Since the response increases as age decreases
in the range 18 to 35, this trend may extend into ages of children, in which case the responses of
children could be underestimated. However, the slope of the age term in the MSS model is
estimated based on data for ages 18 to 35 and does not capture differences in age trend within
this range; in particular, we don't know at what age the response peaks, which could be above or
below 18. The evidence from clinical studies indicates that the responsiveness of children to Os
is about the same as for young adults (ISA, 2012, p. 6-21). This suggests that the age term for
children should not be higher than the age term for young adults. See Sections 6.4.2 and 6.5.3,
and Appendices 6D and 6E for more details.
Because the responses to Os decline from age 18 until around age 55 and for ages older
than 55 the response are minimal, we let the age term for ages 35 to 55 linearly decrease to zero
and set it to zero for ages > 55.
6-12
-------
"/« healthy individuals, the fastest rate of decline in Os responsiveness
appears between the ages of 18 and 35 years .... During the middle age period
(35-55 years), Os sensitivity continues to decline, but at a much lower rate.
Beyond this age (>55 years), acute Os exposure elicits minimal spirometric
changes." (ISA, 2012, p. 6-23)
In order to extend the age term to ages outside the range of ages the MSS model is based
on (ages 18-35), we reparameterize the age term in the numerator of equation 6-3 by
[Pi + P2(ai Age + 012)], for different ranges of ages (ai and 012 depend on age), requiring that these
terms match at each boundary to form a piecewise linear continuous function of age. The
foregoing assumptions result in the following values of ai and 012 for four age ranges (Table 6-2).
Table 6-2. Age Term Parameters for Application of the 2012 MSS
Threshold Model to All Ages.
Age Range
5-17
18-35
36-55
>55
Pi
10.916
10.916
10.916
0
P2
-0.2104
-0.2104
-0.2104
0
CM
0
1
2.0341
0
0(2
-5.8
-23.8
-59.994
0
The lung function decrements estimated by the MSS (2010) model for a particular case
are illustrated in Figure 6-2 and Figure 6-3. Figure 6-2 shows the predictions of the MSS model
for 20-year old individuals with a (typical) body surface area (BSA) of 2 m2 and a target
ventilation rate of 40 L/min (moderate exertion) and an Os exposure level of 100 ppb, under the
conditions of a typical 6.6-hour clinical study. Subjects alternated 50 minutes of moderate
exercise with 10 minutes of rest for the first three hours, with the exercise occurring first. For the
next 35 minutes (lunch), subjects continued exposure at rest. For the remaining three hours of the
exposure period, subjects again alternated 50 minutes of exercise with 10 minutes of rest. The
inter-individual variability predicted by this model is depicted by the boxplots in this figure. The
predictions for the median individual over time are given by the line. Minute-by-minute
predictions for the median individual for an exposure level of 100 ppb are shown in Figure 6-3.
The stairstep response results from the pattern of exercise and rest during the experiment.
6-13
-------
45 i
0
7
hour
Figure 6-2. Distribution of Responses (Lung Function Decrements in FEVi) Predicted by
the MSS Model for 20-Year Old Individuals. Exposure to 100 ppb Os at Moderate Exercise
(40 L/min, BSA=2 m2) Under the Conditions of a Typical 6.6-hour Clinical Study.
The bottom and top edges of the boxes are at the 25th and 75th percentiles. The center horizontal line is
drawn at the 50th percentile (median). The whiskers are at the 1st and 99th percentiles.
6-14
-------
7:
£ 5'
I 4
o
(D
li
T^
0
hour
Figure 6-3. Median Response (Lung Function Decrements in FEVi) Predicted by the MSS
Model for 20-Year Old Individuals. Exposure to 100 ppb Os at Moderate Exercise (40
L/min, BSA=2 m2) Under the Conditions of a Typical 6.6-hour Clinical Study.
Figure 6-4 and Figure 6-5 illustrate the threshold effect based on McDonnell et al. (2012).
Figure 6-4 is a graph of the median response for a population of 20-year old individuals over a
6.6-hour time period. The exposure concentration is a constant 100 ppb over this time period,
while the individuals are exercising (at 30 L/min-m2 BSA) from hour 1 to hour 3 and at rest (5
L/min-m2 BSA) otherwise. There is a 1.3-hour delay in response due to the threshold; without
the threshold, the response starts increasing after exposure starts (hour 0). Figure 6-5 shows the
corresponding probability of a response (FEVi decrement) > 10% over the time period for the
two models. There is very little difference in response between the threshold and non-threshold
models.
6-15
-------
LJJ
LL
1234567
hour
Model no-threshold ~ threshold
Figure 6-4. Median Response (FEVi Decrements) Predicted by the MSS Threshold and
Non-Threshold Models for 20-Year Old Individuals, Constant 100 ppb Os Exposure For
6.6 Hours, 2 Hours Heavy Exercise From Hour 1 to 3 (30 L/min-m2 BSA).
6-16
-------
CD
on
O
o
>>
26%
24%
22%
20%
18%
16%
14%
12%
10%
8%
6%
4%
2%
o
2345
hour
~ no-threshold ~~ threshold
Model
Figure 6-5. Probability of Response > 10% Predicted by the MSS Threshold and Non-
Threshold Models for 20-Year Old Individuals, Constant 100 ppb O3 Exposure For 6.6
Hours, 2 Hours Heavy Exercise From Hour 1 to 3 (30 L/min-m2 BSA).
6.2.5 The Exposure-Response Function Approach Used in Prior Reviews
As described in section 3.1.2 of the 2007 Risk Assessment Technical Support Document
(U.S. EPA, 2007b), a Bayesian Markov Chain Monte Carlo approach (Lunn et al., 2012) was
used to estimate probabilistic E-R relationships for lung function decrements associated with 8-
hr Os exposures occurring at moderate exertion. In the previous review, summary data from the
Folinsbee et al. (1988), Horstman et al. (1990), McDonnell et al. (1991), and Adams (2002,
2003, 2006) studies were combined to estimate E-R relationships for 8-hr exposures at moderate
exertion for each of the three measures of lung function decrement (> 10, > 15, > 20%
decrements in FEVi). In this HREA we have updated this E-R function with the results from two
additional studies (Kim et al., 2011; Schelegle et al., 2009). The controlled human exposure
study data were corrected on an individual basis for study effects in clean filtered air to remove
any systemic bias that might be present in the data attributable to the effects of the study itself
(e.g. exercise, diurnal effect, etc.) (ISA, Section 6.2.1.1). This is done by subtracting the FEVi
decrement in filtered air from the FEVi decrement (at the same time point) during exposure to
Os. An example of this calculation is given in Appendix 6D.
Table 6-3 presents a summary of the study-specific results based on correcting all
individual responses for the effect on lung function decrements of exercise in clean air.
6-17
-------
Table 6-3. Study-specific Os E-R Data for Lung Function Decrements Based on Correcting
Individual Responses for the Effect on Lung Function of Exercise in Clean Air, Ages 18-35.
Study, Grouped by
Average Os
Exposure
0.04 ppm Os
Adams et al. (2002)
Adams et al. (2006)
0.06 ppm Os
Adams et al. (2006)
Kimetal. (2011)
Schelegle et al. (2009)
0.07 ppm Os
Schelegle et al. (2009)
0.08 ppm Os
Adams et al. (2002)
Adams et al. (2003)
Adams et al. (2006)
F-H-M3
Kimetal. (2011)
Schelegle et al. (2009)
0.087 ppm Os
Schelegle et al. (2009)
0.1 ppm Os
F-H-M3
0.12 ppm Os
Adams et al. (2002)
F-H-M3
Protocol
Square-wave, face mask
Triangular
Square-wave
Triangular
Square-wave
Variable levels (0.06 ppm
avg)
Variable levels (0.07 ppm
avg)
Square-wave, face mask
Square-wave, chamber
Square-wave, face mask
Variable levels (0.08 ppm
avg), chamber
Variable levels (0.08 ppm
avg), face mask
Square-wave
Triangular
Square-wave
Square-wave
Variable levels (0.08 ppm
avg)
Variable levels (0.087 ppm
avg)
Square-wave
Square-wave, chamber
Square-wave, face mask
Square-wave
Number
Exposed
30
30
30
30
59
31
31
30
30
30
30
\J\J
30
\J\J
30
30
60
30
31
31
32
30
30
30
Number of Responses
AFEVi > AFEVi > AFEVi >
10% 15% 20%
2
0
2
2
3
4
6
6
6
5
6
\J
5
\J
7
9
17
4
10
14
13
17
21
18
0
0
0
2
1
2
3
5
2
2
1
1
1
1
2
3
11
1
5
10
11
12
13
15
0
0
0
0
0
1
2
2
1
2
1
1
1
1
1
1
8
0
4
7
6
10
7
10
1F-H-M combines data from Folinsbee et al. (1988), Horstman et al. (1990), and McDonnell et al. (1991).
6-18
-------
For the risk assessment conducted during the 2007 Os NAAQS review (U.S. EPA,
2007b), EPA considered both linear and logistic functional forms in estimating the E-R
relationship and chose a 90 percent logistic/10 percent piecewise-linear split using a Bayesian
Markov Chain Monte Carlo approach. This Bayesian estimation approach incorporates both
model uncertainty and uncertainty due to sampling variability.
For each of the three measures of lung function decrement, EPA assumed a 90 percent
probability that the E-R function has the following 3-parameter logistic form:3
( •
y(X,
, ,
Equation (6-4)
where x denotes the Os concentration (in ppm) to which the individual is exposed, y
denotes the corresponding response (decrement in FEVi > 10%, > 15% or > 20%), and a, /?, and
7 are the three parameters whose values are estimated.
We assumed a 10 percent probability that the E-R function has the following linear with
threshold (hockey stick) form:
\a + fie, for a + j3x> 0
y(x;a,j3) = -\ Equation (6-5)
[0, fora + (3x
-------
For each of the two functional forms (logistic and linear), we derived a Bayesian
posterior distribution using this binomial likelihood function in combination with prior
distributions for each of the unknown parameters (Box and Tiao, 1973). We assumed lognormal
priors with maximum likelihood estimates of the means and variances for the parameters of the
logistic function, and normal priors, similarly with maximum likelihood estimates for the means
and variances, for the parameters of the linear function. For each of the two functional forms
considered, we used 1,000 iterations as the "burn-in" period4 followed by 9,000 iterations for the
estimation. Each iteration corresponds to a set of values for the parameters of the (logistic or
linear) E-R function. We combined the 9,000 sets of values from the logistic model runs with the
last 1,000 sets of values from the linear model runs to get a single combined distribution of
10,000 sets of values reflecting the 90 percent/10 percent assumption stated above. WinBUGS
version 1.4.3 was used for these analyses (WinBUGS; Lunn et al., 2012).
For any Os concentration, x, we can derive the nth percentile response value, for any n, by
evaluating the E-R function at x using each of the 10,000 sets of parameter values (9,000 of
which were for a logistic model and 1,000 of which were for a linear model). The resulting 2.5th
percentile, median (50th percentile), and 97.5th percentile E-R functions for changes in FEVi >
10% are shown in Figure 6-6, along with the response data to which they were fit. The
corresponding E-R functions for changes in FEVi > 15% and > 20% are shown in Appendix 6A.
The values of the functions are also provided in Appendix 6A.
4 Markov chain Monte Carlo (MCMC) simulations require an initial adaptive "burn-in" set of iterations, which are
not used. This allows the MCMC sampling to stabilize.
6-20
-------
90%
80%
0%
0.02
0.04
0.06
0.08
0.1
0.12
Ozone Exposure (ppm)
0.14
0.16
— -97.5th Pctl (FEV1>10%)
Median (FEV1>10%)
2.5th Pctl (FEV1>10%)
• Data for FEV1>10%
Functional form mix:
90% logistic &
10% linear
Figure 6-6. Probabilistic E-R Relationships for FEVi Decrements > 10% for 8-hr
Exposures At Moderate Exertion, Ages 18-35. Values associated with data points are the
number of subject-exposures at each exposure concentration.
The population risk is estimated by multiplying the expected risk by the number of
people in the relevant population, as shown in Equation (6-6) below. The risk (i.e., expected
fractional response rate) for the kth fractile, Rk is estimated as:
Equation (6-6)
where:
ej = (the midpoint of) the/ category of personal exposure to Os;
Pj = the fraction of the population having personal exposures to Os concentration
of e}ppm;
RRk | ej = k-fractile response rate at Os exposure concentration ej;
N= number of intervals (categories) of Os personal exposure concentration.
6-21
-------
Exposure estimates used in this portion of the risk assessment were obtained from APEX
for each of the 15 urban study areas and the five air quality scenarios. Chapter 5 provides
additional details about the inputs and methodology used to estimate population exposure in the
urban study areas. Exposure estimates for all and asthmatic school-aged children (ages 5 to 18)
were combined with probabilistic E-R relationships for lung function decrements associated with
8-hr exposure while engaged in moderate exertion. Individuals engaged in activities that resulted
in an average equivalent (BSA-normalized) ventilation rate (EVR) for the 8-hr period at or above
13 L/min-m2 BSA were included in the exposure estimates for 8-hr moderate or greater exertion.
This range was selected based on the EVRs for the group of subjects in the controlled human
exposure studies that were the basis for the E-R relationships used in this portion of the risk
assessment.
6.3 OZONE RISK ESTIMATES
This section provides lung function risk estimates associated with several air quality
scenarios: five recent years of air quality as represented by 2006 to 2010 monitoring data, and air
quality in those years after simulating just meeting the existing Os standard and alternative
standard levels of 0.070, 0.065, and 0.060 ppm. The risk measures presented here are the
percents of the population estimated to experience lung function responses greater than 10, 15,
and 20%, one or more times or six or more times during an Os season, for three age groups:
school-aged children (ages 5-18), young adults (ages 19-35) and adults ages 36-55. Results for
adults older than 55 are not presented since the responses for this age group are estimated to be
minimal. People with multiple events with large lung function decreases are more at risk than
those with only one such event during the Os season. Although six events is less than once per
month, we see dramatic decreases in population risk in going from one or more to six or more
events during a season, which is why we report on six or more events rather than a higher
number.
In the figures and tables that follow, "base" indicates the base case scenario of recent air
quality for the indicated year. "75," "70," "65," and "60" respectively indicate the existing Os
standard and alternative standard levels of 0.070, 0.065, and 0.060 ppm. "75 6-8" indicates the
0.075 ppm existing 8-hr standard based on rollback for the 2006-2008 period, while "75 8-10"
indicates the existing standard scenario based on rollback for the 2008-2010 period. There are
two estimates of results for the 2008 existing and alternative standard scenarios (because 2008
overlaps the two rollback periods) and one for each of the other four years. These two estimates
for 2008 can be quite different because of the relationship between the design value over the
three-year period and the amount of adjustment to the air quality distribution in 2008 that can
result.
6-22
-------
6.3.1 Lung Function Risk Estimates Based on the McDonnell-Stewart-Smith Model
Results based on the McDonnell-Stewart-Smith (2012) threshold model are summarized
in this section; detailed results can be found in Appendix 6B.
Figure 6-7 shows the results for school-aged children in the same format used in exposure
results, explained in Section 5.3.1. Figure 6-7 depicts results for all urban study areas, year, and
scenarios for ages 5 to 18 with > 1 occurrences of FEVi decrements > 10, 15, 20% and illustrates
the variation of results across study area, year, and scenarios.
Figure 6-8 shows the variation across study areas (horizontally) and years (vertically) for
the percent of school-aged children with > 1 occurrences of FEVi decrements > 10% with air
quality just meeting the potential alternative standard of 0.07 ppm. The points above each study
area on this graph represent the risk for the six years for the study area (2008 has two points,
corresponding to the different 2006-2008 and 2008-2010 design values used to adjust the air
quality to meet 0.07 ppm). There is substantial variability both across years and across study
areas. Denver has the highest overall risks, while Cleveland and New York have the lowest. Los
Angeles has the smallest variation across years, with a range of 2.3% (from 14.3% to 16.6%).
The other study areas have a range of around 4% to 7.5% across years.
Table 6-4 and Table 6-5 present summary results (ranges over study areas and years) of
FEVi decrements > 10 and 15% estimated by the MSS model for the different age groups. The
results for asthmatic school-aged children are very similar to the results for all school-aged
children and are not presented here.
6-23
-------
xp
(^
O
Al
"S
s
O
(L)
PH
(L)
O
o
O
o
Al
(L)
O
T3
CD
o
O
o
fin
Fev
FEV10
FEV15
FEV20
/ / / // / / / // / / /
s> 5>
Figure 6-7. Risk Results For All School-Aged Children With > 1 Occurrences of FEVi
Decrements > 10,15, 20% For All Study Areas, Year, and Scenarios (y-axis is percent of children
affected).
6-24
-------
21%:
20%:
19%:
18%:
17%:
S :
S 16%:
I o :
d, 15/0.
° 14%:
8 :
| 13%:
12%:
io%-
9%:
3! *
*
*
*
* * *
* * *
*
ABB
t a o
1 1 s
a t t
n i o
t m n
a o
r
e
*
*
*
*
*
i
c
h
i
c
a
g
o
*
*
*
*
C
1
e
V
e
1
a
n
d
*
*
*
*
*
*
*
*
*
*
*
i i i
ODD
a e e
1 n t
1 v r
a e o
s r i
t
*
*
i
H
o
u
s
t
o
n
*
*
*
L
o
s
A
n
g
e
1
e
s
*
*
*
*
N
e
w
Y
o
r
k
*
*
*
*
*
i
P
h
i
1
a
d
e
1
P
h
i
a
*
*
*
*
*
*
S
a
c
r
a
m
e
n
t
0
*
*
*
*
*
*
i
S
t
L
0
u
i
s
*
*
*
W
a
s
h
i
n
g
t
0
n
Urban area
Figure 6-8. Risk Results For All School-Aged Children With > 1 Occurrences of FEVi
Decrements > 10% Under The 0.07 ppm Alternative Standard Showing Variability Across
Study Areas (Horizontally) and Years (Vertically).
6-25
-------
Table 6-4. Ranges of Percents of Population Experiencing One or More and Six or More
Days during the Os Season with Lung Function Decrement (AFEVi) More than 10%. The
numbers in this table are the minimum and maximum percents estimated over all study
areas and years, for each age group and scenario.
Age group
5 to 18
5 to 18
5 to 18
5 to 18
5 to 18
19 to 35
19 to 35
19 to 35
19 to 35
19 to 35
36 to 55
36 to 55
36 to 55
36 to 55
36 to 55
>55
Scenario
base
75
70
65
60
base
75
70
65
60
base
75
70
65
60
All
Percent of People Experiencing >
1 day with AFEVi > 10%
Minimum
11%
11%
8%
2%
4%
3%
3%
2%
1%
1%
1%
1%
0%
0%
0%
0%
Maximum
31%
22%
20%
18%
13%
13%
9%
8%
6%
5%
4%
2%
2%
2%
1%
0%
Percent of People Experiencing >
6 days with AFEVi > 10%
Minimum
1%
1%
1%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
Maximum
9%
6%
5%
4%
3%
1%
1%
1%
1%
0%
0%
0%
0%
0%
0%
0%
6-26
-------
Table 6-5. Ranges of Percents of Population Experiencing One or More and Six or more
Days during the Os Season with Lung Function Decrement (AFEVi) More than 15%. The
numbers in this table are the minimum and maximum percents estimated over all study
areas and years, for each age group and scenario.
Age group
5 to 18
5 to 18
5 to 18
5 to 18
5 to 18
19 to 35
19 to 35
19 to 35
19 to 35
19 to 35
36 to 55
36 to 55
36 to 55
36 to 55
36 to 55
>55
Scenario
base
75
70
65
60
base
75
70
65
60
base
75
70
65
60
All
Percent of People Experiencing >
1 day with AFEVi > 15%
Minimum
2%
2%
2%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
Maximum
12%
6%
5%
4%
3%
3%
2%
1%
1%
1%
1%
0%
0%
0%
0%
0%
Percent of People Experiencing >
6 days with AFEVi > 15%
Minimum
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
Maximum
3%
1%
1%
1%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
0%
Figure 6-9 illustrates the distribution of responses (FEVi decrements > 10%) across
ranges of ambient concentrations of Os for school-aged children for one city and scenario (Los
Angeles, 2006 recent air quality). The concentrations are daily 8-hr average ambient
concentrations during the 8-hr period with maximum 8-hr average exposure for that day.
These concentrations are less than but close to daily maximum 8-hr average ambient
concentrations and are greater than daily maximum 8-hr average exposure concentrations. The
percents in this chart reflect the frequencies of person-days with FEVi decrements > 10% within
a concentration bin as percents of all person-days with FEVi decrements > 10%. Figure 6-9
shows that more than 90% of daily instances of FEVi decrements > 10% occur when 8-hr
average ambient concentrations are above 40 ppb for this modeled scenario. This distribution
will be different for different study areas, years, and air quality scenarios.
6-27
-------
I
o
w
22
<+H
O
20;
18;
16;
14
12:
6;
4;
2;
0:
' OO \
8-hour average concentration (ppb)
Figure 6-9. Distribution of Daily FEVi Decrements > 10% Across Ranges of 8-hr Average
Ambient Os Concentrations (Los Angeles, 2006 recent air quality).
Outdoor Workers
Outdoor workers spend more time outdoors than the general population and therefore are
at higher risk for health effects due to Os. We conducted simulations of outdoor workers ages 19-
35 for Atlanta (2006) for the existing and alternative standards to estimate the risk of this group
for experiencing FEVi decrements > 15%. The methodology for simulating outdoor workers
involves modifying activity diaries to represent outdoor workers and is described in Section 5.3.3
in Chapter 5. Table 6-6 shows the results of these simulations and compares them with the
results for the general population, ages 19-35. The percents of people experiencing one or more
FEVi decrements > 15% during the 2006 Os season in Atlanta are 3.6 times higher for outdoor
workers than for the general population (ages 19-35) under the existing standard, and range up to
5.3 times higher for the alternative standards. The percents of people experiencing six or more
FEVi decrements > 15% during the 2006 Os season in Atlanta are 24 times higher for outdoor
workers than for the general population under the existing standard, and range up to 150 times
higher for the alternative standards. As expected, we see that the risk of repeated occurrences of
6-28
-------
FEVi decrements > 15% is much greater for outdoor workers than for the general population.
Part of the reason for this is that APEX tends to underestimate the number of individuals who
have very repetitive activity patterns (e.g., 9 to 5 weekdays office workers) when using the
CHAD activity database and the method selected for generating longitudinal diary profiles (see
Section 5.3.1).
Table 6-6. Percents of the General Population and Outdoor Workers (ages 19-35)
Experiencing One or More and Six or More FEVi Decrements > 15% (based on Atlanta
2006 APEX simulations).
Scenario
General
population
ages 1 9-35
1 or more occurrences
Existing standard
70 ppb alt. std.
65 ppb alt. std.
60 ppb alt. std.
1 .2%
0.84%
0.55%
0.32%
6 or more occurrences
Existing standard
70 ppb alt. std.
65 ppb alt. std.
60 ppb alt. std.
0.05%
0.018%
0.005%
0.005%
Outdoor workers
ages 1 9-35
4.3%
3.2%
2.5%
1 .7%
1 .2%
0.93%
0.74%
0.55%
6.3.2 Lung Function Risk Estimates Based on the Exposure-Response Functions
Approach Used in Prior Reviews
In this section we present lung function risk estimates for all school-aged children
following the methodology used in previous reviews, based on updated E-R functions. In
Appendix 6C we compare these estimates with those from the previous review. Table 6-7
provides an overall summary of results for each air quality scenario by tabulating the minimum
and maximum estimates over all study areas and years of percents of all school-aged children
(ages 5 to 18) experiencing one or more days (during the Os season) with FEVi decrement more
than 10 and 15%. This table can be compared with Table 6-4 and Table 6-5, which have
analogous results for the MSS model. These results are much lower than the MSS model results.
The reasons for this are described in Section 6.3.3 below.
6-29
-------
Table 6-7. Ranges of Percents of all School-aged Children Experiencing One or More Days
during the Os Season with Lung Function Decrement (AFEVi) More than 10 and 15%,
based on the E-R Function Approach. The numbers in this table are the minimum and
maximum percents estimated over all study areas and years.
Scenario
base
75
70
65
60
minimum
percent
experiencing
> 1 day with
AFEVi>10%
4%
3%
3%
1%
2%
maximum
percent
experiencing
> 1 day with
AFEVi>10%
11%
6%
6%
5%
3%
minimum
percent
experiencing
> 1 day with
AFEVi>15%
1%
1%
1%
0%
0%
maximum
percent
experiencing
> 1 day with
AFEVi>15%
5%
2%
2%
1%
1%
6.3.3 Comparison of the MSS Model with the Exposure-Response Function Approach
There are two key differences between the MSS and E-R models. The E-R model
estimates the distribution of FEVi decrements across the population or study group, whereas the
MSS model estimates FEVi decrements at the individual level and then these are aggregated to
obtain the population distribution. Thus the MSS model allows for detailed analyses of
conditions that influence risk. Second, the E-R model estimates FEVi decrements only for 8-hr
average exposures when the 8-hr average exertion level is moderate or greater. The MSS model
estimates FEVi decrements for any averaging time and therefore accounts for a wider range of
activities that might result in FEVi decrements.
A comparison of the MSS model with the E-R function approach for the 2006 existing
standard scenarios is summarized in Table 6-8 which lists estimates of the percents of school-
aged children estimated to experience lung function responses greater than 10, 15, and 20%. The
MSS model estimates are significantly higher than the E-R function approach estimates. In most
cases, the MSS model gives results about a factor of three higher than the E-R function model for
school-aged children. This is expected, since, as discussed above, the MSS model includes
responses for a wider range of exposure protocols (under different levels of exertion, lengths of
exposures, and patterns of exposure concentrations) than the E-R model of previous reviews.
6-30
-------
Table 6-8. Comparison of Responses from the MSS Model with Responses from the
Population E-R Method, 2006 Existing Standard Air Quality Scenario, Ages 5 to 18.
Urban Study
Area
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
> 10% FEVi decrement
MSS model
19.2%
18.6%
13.6%
14.4%
13.5%
21 .6%
20.2%
13.6%
16.2%
18.2%
12.7%
16.4%
17.9%
18.6%
15.9%
E-R method
5.6%
5.4%
4.5%
4.7%
4.2%
6.0%
5.8%
4.4%
4.7%
4.8%
4.2%
4.8%
5.1%
5.4%
4.6%
> 15% FEVi decrement
MSS model
5.3%
5.2%
3.7%
3.9%
3.3%
6.4%
5.6%
3.5%
4.2%
4.2%
3.2%
4.2%
4.8%
5.1%
4.0%
E-R method
1 .7%
1 .6%
1 .2%
1 .3%
1.1%
1 .8%
1 .7%
1 .2%
1 .3%
1 .2%
1.1%
1 .3%
1 .4%
1 .6%
1 .3%
> 20% FEVi decrement
MSS model
2.1%
2.1%
1 .4%
1 .6%
1 .2%
2.7%
2.2%
1 .3%
1 .6%
1 .5%
1 .2%
1 .6%
2.8%
2.0%
1 .5%
E-R method
0.7%
0.7%
0.5%
0.5%
0.4%
0.8%
0.7%
0.4%
0.5%
0.5%
0.4%
0.5%
0.6%
0.6%
0.5%
Since the E-R method of the previous reviews only looks at 8-hr exposures concomitant
with EVR > 13 L/min-m2 BSA (hereafter, EVR > 13), it is of interest to compare the E-R method
results with the corresponding MSS model results (instances of AFEVi > 10, 15, 20%
concomitant with EVR > 13).
We performed this comparison for four APEX simulations: the Atlanta March 1-October
30, 2006 base case, ages 18-35; the Los Angeles May 29-July 28, 2006 base case, age 25; the
Los Angeles May 29-July 28, 2006 base case, ages 18-35; and the Los Angeles Jan 1-Dec 31,
2006 base case, ages 18-35.
For the Atlanta simulation, the E-R function approach gives 5.0, 1.8, and 0.9%
responding for AFEVi > 10, 15, 20%. The MSS model approach gives 11.54, 3.26, and 1.28%
responding for AFEVi > 10, 15, 20%. The percents of the population for AFEVi > 10, 15, 20% at
the end of the daily max 8-hr average exposure period where the concomitant 8-hr average EVR
is > 13 are 6.67%, 2.09%, and 0.84%. 15.17% of the population never have any instances of
EVR > 13 and 0.41% have at least one occurrence of AFEVi > 10% while never having EVR >
13 for any 8-hr period. 4.46% (not among the 15.17%) have instances of AFEVi > 10% but none
of those instances with concomitant EVR > 13. The 11.54% responding is made up of 6.67% of
6-31
-------
the population with instances of AFEVi > 10% concomitant with EVR > 13 and 4.87% with
instances of AFEVi > 10% not concomitant with EVR > 13.
Table 6-9 and Table 6-10 summarize the pertinent results for the Atlanta and three Los
Angeles simulations. Looking at the first rows of these tables shows that these models have
similar corresponding results. The broader scope of activity/exposure patterns encompassed by
the MSS model, beyond the 8-hr average EVR > 13 restriction of the E-R model, contributes
from a third to a half to the total MSS model risk and to a large part explains the differences
between the models for ages 18-35. The difference between the MSS and E-R models is larger
for school-aged children than for adults ages 18-35 due to the increased EVR and time spent
outdoors in children compared to adults.
6-32
-------
Table 6-9. Comparison of MSS Model and E-R Model of Previous Reviews for Atlanta, Mar 1-Oct 30, 2006, ages 18-35.
Component of results
profiles with instances of AFEVi > cutoff
concomitant with 8-hr EVR > 13
profiles with instances of AFEVi > cutoff never
concomitant with 8-hr EVR > 13
Final result of each model
MSS model
AFEVi > 10%
6.7%
4.8%
1 1 .5%
E-R model
AFEVi > 10%
5.0%
5.0%
MSS model
AFEVi > 15%
2.1%
1 .2%
3.3%
E-R model
AFEVi > 15%
1 .8%
1 .8%
MSS model
AFEVi > 20%
0.8%
0.5%
1 .3%
E-R model
AFEVi > 20%
0.9%
0.9%
Table 6-10. Comparison of MSS Model and E-R Model of Previous Reviews for Los Angeles, Jan 1-Dec 31, 2006, ages 18-35.
Component of results
profiles with instances of AFEVi > cutoff
concomitant with 8-hr EVR > 13
profiles with instances of AFEVi > cutoff never
concomitant with 8-hr EVR > 13
Final result of each model
MSS model
AFEVi > 10%
7.9%
6.5%
14.4%
E-R model
AFEVi > 10%
6.2%
6.2%
MSS model
AFEVi > 15%
2.6%
1 .8%
4.4%
E-R model
AFEVi > 15%
2.6%
2.6%
MSS model
AFEVi > 20%
1 .2%
0.8%
2.0%
E-R model
AFEVi > 20%
1 .4%
1 .4%
6-33
-------
Figure 6-10 compares the E-R function to the response curve of the MSS model restricted
to 8-hr average EVR > 13 and shows that these curves are very close. The E-R model has a
higher response for the low and high ranges of exposure concentrations, while the MSS model is
higher in the mid-range of exposures.
I
o
S,
a
24%
23%'
22%'
21%
20%
19%
18%
17%
16%
14%
13%
12%
11%
10%
9%
7%
6%
5%'
4%
3%
1%-
o%-
0
10
20
AFEV1>=10% (MSS)
AFEV1>=10% (E-R)
30 40 50
Ozone exposure (ppb)
AFEV1>=15% (MSS)
AFEV1>=15% (E-R)
60
70
80
AFEV1>=20% (MSS)
AFEV1>=20% (E-R)
Figure 6-10. Comparison of E-R and MSS Model (restricted to 8-hr average EVR > 13)
Response Functions (Atlanta 2006 base case, ages 18-35).
Another element of the difference between the models derives from the distribution of
EVR in the clinical studies the E-R approach is based on and how this compares to the
distribution of EVR in the APEX simulations. Most of the clinical studies are conducted with a
target EVR of 20 L/min-m2 BSA and the actual EVRs vary somewhat around this value. The
rationale for the cutpoint of 13 L/min-m2 BSA is described in EPA's responses to comments on
the 1996 proposed rule on the NAAQS for Os (Federal Register, 1996) as "for the 8-hr health
risk assessment the range (based on being within 2 standard deviations of the mean) of EVRs
observed in the subjects who participated in the study [McDonnell et al., 1991] was 13-27 liters
6-34
-------
per minute per meter squared [BSA] (L/min-m2)." Figure 6-11 shows the distribution of EVRs >
13 for the Atlanta simulation and is clearly shifted much lower than the distribution of EVR in
the clinical studies. This could lead to an overestimation of the percent of responders by the E-R
method, since higher EVRs lead to higher lung function decrements and it is applying an E-R
function based on EVRs around 20 to a population with median EVRs around 14.5.
EVR
13.5
14.5
15.5
16.5
17.5
18.5
19.5
20.5
21.5
22.5
23.5
1
1
1
1
1
1
|
H
n
]
24.5 P
25.5 1
26.5 I
27.5
28.5
29.5
30.5
31.5
32.5
33.5
34.5
35.5
PCT.
34.78
22.39
14.63
9.64
6.36
4.20
2.77
1.80
1.17
0.77
0.51
0.34
0.22
0.15
0.10
0.06
0.04
0.03
0.01
0.01
0.00
0.00
0.00
CUM.
PCT.
34.78
57.17
71.80
81.44
87.80
92.00
94.77
96.56
97.74
98.51
99.02
99.37
99.59
99.74
99.83
99.90
99.94
99.97
99.98
99.99
99.99
100.00
100.00
200000
400000
600000
FREQUENCY
800000
1000000
1200000
Figure 6-11. Distribution of Daily Maximum 8-hr Average EVR For Values of EVR > 13
(L/min-m2) (midpoints on vertical axis) (Atlanta 2006 base case, ages 18-35).
6-35
-------
6.4 EVALUATION OF THE MSS MODEL
6.4.1 Summary of Published Evaluations
McDonnell et al. (2010) performed a detailed evaluation of their model using two
methods: (1) cross-validation and (2) comparison of an independent data set against the
predictions of the model.
The cross-validation was based on the data set of 15 EPA studies from which their
original model was developed (McDonnell et al., 2007). This data set has 541 subjects, each with
multiple measurements during single experiments. Subjects were omitted from the data set, one
at a time, the model refit to the reduced data set, and the resulting parameters used to predict the
FEVi decrements for the omitted subject. The authors then compare the mean predictions and
mean observed values for each subject and presented these results in a scatter plot (Figure Ib,
McDonnell et al., 2010). The observations exhibit much more variability than the predictions; for
observed values of 20%, predicted values range from around 2 to 19%; and all observed values
above 20% are underpredicted (the observed values range from -20 to 60%, while the predicted
values range from 0 to 20%). These features result from the omission of the inter- and intra-
individual variability terms (Ui and Sijk) in the MSS model (equation 6-3), which are accounted
for in the risk estimates in this chapter.
Model predictions were compared against an independent data set of seven clinical
studies with a total of 204 subjects (McDonnell et al., 2010). Graphs of predicted and observed
study means vs. time show fair to good model fit. The authors do not present overall fit statistics
that are directly commensurate with the statistics of interest in this risk assessment: the
proportions of people with FEVi decrements greater than 10, 15, and 20%.
McDonnell et al. (2012) do compare observed and predicted proportions of people with
FEVi decrements greater than 10, 15, and 20% and provide the corresponding scatter plots
(Figure 4). They find the model to be unbiased, with the slopes of the observed vs. predicted
lines for 10, 15, and 20% to be around 1.0 and the R2 respectively 0.78, 0.73, and 0.67. The
higher observed proportions of people with FEVi decrements greater than 10, 15, and 20%
tended to be substantially underpredicted.
6.4.2 Children
A clinical study with children (ages 8-11; mean, 10 years; n=22), exposed to 120 ppb Os
over 2.5 hours at heavy exertion levels was done by McDonnell et al. (1985). This study could be
used to fit the model for children if all of the measurements of FEVi and ventilation rates were
available. The paper lists the end-of exposure FEVi responses for each individual (but not
ventilation rates), which we use to compare with the MSS model with the age term extension
described in Section 6.2.4. The numbers of subjects with clean-air adjusted responses greater
6-36
-------
than 10%, 15%, and 20% are respectively 4, 2, and 1, corresponding to 18.2%, 9.1%, and 4.5%
of the number of subjects. We ran the MSS 2010 model using the mean and standard deviation of
the ventilation rates reported in the paper. Resting ventilation rates were assumed to be 10.4
L/min (Avol et al., 1985) and BSA to be 1.08 m2 (U.S. EPA, 2011). Details of this comparison
can be found in Appendix 6D.
Table 6-11 compares the results of this simulation with the results of McDonnell et al.
(1985). The agreement is fairly good. Due to the limited sample size of 22 subjects from only
one study and the assumptions made in running the MSS model, this does not provide
confirmation that the age term extension is correct; on the other hand, this comparison does not
indicate that there is a problem with the age term extension. Information is not available that
would allow us to provide respectable confidence intervals for these estimates.
Table 6-11. Comparison of Responses from the MSS 2010 Model with Responses from
McDonnell et al. (1985).
Percent
responding
> 10% FEVi decrement
MSS model
18.4%
McDonnell
etal. (1985)
18.2%
(4 subjects)
> 15% FEVi decrement
MSS model
6.8%
McDonnell
etal. (1985)
9.1%
(2 subjects)
> 20% FEVi decrement
MSS model
2.3%
McDonnell
etal. (1985)
4.5%
(1 subject)
6.4.3 Threshold vs. Non-Threshold Models
The difference between the results of the MSS threshold and non-threshold models is
minor, with the threshold version estimates of lung function decrements almost identical to the
no-threshold version for the Atlanta 2006 recent air quality base case, as can be seen by
comparing Table 6-12 with Table 6-13, and inspection of Figures 6-4 and 6-5. This is consistent
with the logistic form of the model, where the impact of exposures to low concentrations on risk
is small.
6-37
-------
Table 6-12. Percents of the Population by Age Group with One or More Days during the
Os Season with Lung Function (FEVi) Decrements more than 10,15, and 20% (Atlanta
2006 base case), MSS Threshold Model, Ambient Monitor Data.5
Age
Group
5 to 18
19 to 35
36 to 55
AFEVi >
10%
31%
11%
3.7%
AFEVi >
15%
13%
3.1%
0.60%
AFEVi >
20%
6.4%
1 .3%
0.14%
Table 6-13. Percents of the Population by Age Group with One or More Days during the
Os Season with Lung Function (FEVi) Decrements more than 10,15, and 20% (Atlanta
2006 base case), MSS No-Threshold Model, Ambient Monitor Data.
Age
Group
5 to 18
19 to 35
36 to 55
AFEVi >
10%
31%
11%
3.8%
AFEVi >
15%
13%
3.1%
0.60%
AFEVi >
20%
6.6%
1 .3%
0.15%
6.5 CHARACTERIZATION OF UNCERTAINTY
In the controlled human exposure study based risk assessment, there are two broad
sources of uncertainty to the risk estimates. One of the most important sources of uncertainty is
the estimation by APEX of the population distribution of individual time series of Os exposures
and ventilation rates. The uncertainty regarding these estimated exposures is discussed in
Chapter 5; they are not discussed further here.
5 In the first draft HREA, monitor-level air quality was provided as input to the APEX model. As discussed in
Chapter 5, tract-level air quality was used in APEX for this HREA. Monitor-level air quality is used for the APEX
simulations here, since these simulations take less time to run. This does not affect the analyses here, since the two
air quality formats yield very similar results (see Appendix 6F).
6-38
-------
In this section, uncertainties associated with the second broad source of uncertainty in the
risk calculation are discussed, namely, uncertainties in the lung function risk model. The specific
sources of uncertainty covered are:
• Statistical model form
• Convergence of APEX results
• Application of model for all lifestages
• Application of model for asthmatic children
• Interaction between Os and other pollutants
6.5.1 Statistical Model Form
The MSS model is a 2-compartment model, the form of which is based on physical
considerations. It accommodates these key features of human exposure studies: (1) FEVi
responses increase with increasing Os concentration, ventilation rate, and duration of exposure if
the other two variables are held constant, (2) the effect of each of these three variables depends
on the levels of the other two variables, (3) FEVi responses depend on age, (4) certain
individuals are consistently more responsive to Cb exposure, and (5) Os-induced FEVi
decrements improve within a few hours of cessation of exposure (McDonnell et al., 2007). These
considerations support the form of the model, as do model evaluations that have been performed
(Section 6.4.1). Although the model does not have good predictive ability for individuals
(psuedo-R2 0.28), it does better at predicting the proportion of individuals with FEVi decrements
> 10, 15, and 20% (psuedo-R2s of 0.78, 0.74, 0.68) (McDonnell et al., 2012).
The clinical studies that these models' estimates are based on were conducted with young
adult volunteers rather than randomly selected individuals, so it may be that selection bias has
influenced the model parameter estimates.
The parameter estimates are not very precise, partly as the result of correlations between
the parameter estimates. Table 6-14 gives 95 percent confidence intervals for each of the
parameter estimates as percents of the estimates, based on the standard errors reported by
McDonnell et al. (2012). Figure 6-12 shows how much the modeled number of children with one
or more FEVi decrements > 10% changes when each parameter is increased by five percent
(keeping the other parameters fixed at their estimates). The MSS model is also sensitive to the
exposure concentrations, but we have not quantified that sensitivity. The scenario modeled is the
Los Angeles 2006 recent conditions base case. The physiological parameter MET, a measure of
the level of exertion for a given activity (see Appendix 6E), is also included here for comparison.
MET is a key variable in calculating ventilation rates and is specified by a distribution for each
activity. Here we have shifted all MET distributions by +5% of their means.
6-39
-------
Table 6-14. MSS Threshold Model Estimated Parameters with Confidence Intervals.
parameter
estimate
standard
error
95% conf.
interval
P1
10.916
0.8446
±1 5%
P2
-0.2104
0.31
±289%
P3
0.01506
0.00333
±43%
P4
13.497
4.734
±69%
P5
0.003221
0.000207
±13%
P6
0.8839
0.0647
±14%
P9
59.284
10.192
±34%
var(U)
0.9373
0.0824
±17%
var(E)
17.0816
1.1506
±13%
from McDonnell et al. (2012).
The most influential parameter in Figure 6-12 is Pe, the power to which ventilation rate is
raised in the MSS model, which makes sense since inhaled dose is equal to the ventilation rate
multiplied by the exposure. An increase of five percent in Pe leads to 27, 40, and 47 percent
increases respectively in the modeled number of children with FEVi decrements > 10, 15 and
20%. The next most influential parameter is the variance of E, the intra-individual variability
term. The least influential parameter is $2, the slope of the age term.
Since the MSS model parameter estimates are to some extent correlated, if we were to
increase a parameter by five percent and hold it fixed while refitting the model, the remaining
parameters would shift to compensate, and the sensitivity calculated from that model's results
would be much less than the corresponding sensitivity in Figure 6-12. As a result of this, the
sensitivities in Figure 6-12 overestimate the sensitivities which account for parameter
correlation. On the other hand, we are only increasing each parameter by five percent, which is
much less than the 95% confidence intervals given in Table 6-14. The uncertainty in the risk
estimates resulting from parameter uncertainty may be more or less than is indicated in Figure
6-12.
We are unable to properly estimate the true sensitivities or quantitatively assess the
uncertainty of the MSS model, since all of the clinical data used by McDonnell and coworkers to
fit the MSS model are not available to EPA (in particular, some studies by Adams and coworkers
are not available to EPA).
6.5.1.1 Age term significance
As discussed in Section 6.5.3 below, there are uncertainties in extrapolating the MSS
model down to age 5 from the age range of 18 to 35 to which the model was fit. Further
considerations indicating that the uncertainty of the extension to children of the MSS model
could be substantial are that the age coefficient $2 = -0.21 (s.e. 0.31) in the MSS model is not
statistically significantly different from zero; and when the MSS model is fit to the U.C. Davis
clinical data the age term is positive, $2 = +0.19 (0.60), although also not statistically
significantly different from zero (McDonnell et al., 2012). Note that, in the previous section, $2
was found to be the least influential model parameter.
6-40
-------
2-
5-
ra
o;
7
8
9-
10-
11
varE
beta2
( 5%) 0% 5%
10% 15% 20% 25% 30%
Sensitivity
Figure 6-12. Sensitivity (Percent Change) of Population With One or More FEVi
Decrements > 10% to a 5% Increase in Individual MSS Model Parameter Estimates.
6-41
-------
6.5.1.2 The inter- and intra-individual variability terms U and £
The MSS model has random variables for inter-individual (between-individual)
variability (U) (ISA, p. 6-16) and intra-individual (within-individual) variability (s) (see equation
6-3). The variable U reflects the responsiveness of an individual to Os and each individual
simulated in APEX is assigned a value of U drawn from a Gaussian distribution with mean zero
and variance estimated when fitting the MSS model. In order to estimate Var(U), it is necessary
to have repeated clinical trials using the same individuals. There are few of these in the data used
to fit the MSS model, making it harder to reliably estimate Var(U). Furthermore, the method
used for adjusting for filtered air (FA) exposures in the data used to fit the MSS model does not
use the subject-specific adjustments that are typically used in these clinical studies (ISA, p.6-4,
6-5). Rather, the mean FA response across a study is used to adjust the Os responses of each
subject in the study (McDonnell et al., 2012, page 621). This adds additional uncertainty to the
estimate of Var(U).
The estimate of Var(U) in the MSS threshold model is 0.9373 (standard deviation
0.9681). Since the actual values of U are bounded, we truncate the distribution of U at ±2
standard deviations (±1.93), a convention we use for the distributions of several physiological
variables input to APEX in the physiology input file. The results of sensitivity analyses of the
MSS model predictions to the estimate of Var(U) are given in Table 6-15. Comparing the base
case simulation to the simulation without any limits on U, we find that the results are not very
sensitive to truncation of the distribution of U. Increasing or decreasing the value of the standard
deviation of U, o(U), does change the risk results significantly. As expected, the extreme case of
setting all values of U to zero results in large changes in risk.
Table 6-15. Sensitivity Analysis of the Inter-individual Variability Term U using the MSS
Threshold Model. Percents of the population aged 5 to 18 with one or more and 6 or more
days during the Os season with lung function (FEVi) decrements more than 10,15, and
20% (Atlanta 2006 base case, ambient monitor data).
Description
st. dev. (U), ± limits on U
base: a(U) = 0.9681, ±1.93
a(U) -50%= 0.4840, ± 0.9680
a(U) +50%= 1 .4522, ± 2.9044
a(U)=0.9681 , no limits on U
a(U) = 0 (all U = 0)
1 or
more
AFEVi
>10%
31 .77%
27.22%
35.53%
32.53%
25.26%
1 or
more
AFEVi
>15%
12.66%
4.85%
20.05%
14.18%
0.94%
1 or
more
AFEVi
>20%
6.55%
0.82%
13.70%
7.97%
0.00%
6 or
more
AFEVi
>10%
9.55%
3.64%
15.67%
10.86%
0.84%
6 or
more
AFEVi
>15%
3.08%
0.21%
8.31%
4.24%
0.00%
6 or
more
AFEVi
>20%
1 .22%
0.01%
5.20%
2.20%
0.00%
6-42
-------
The MSS model estimated intra-individual variability Var(s) has two basic components:
(1) the intra-individual variability of the true response to Os (both within-day and between-day)
and (2) measurement error. These cannot be distinguished based on the available data. We are
assuming that all of this variability is due to the true response, which will (absent other
uncertainties) tend to overestimate the response to Os. We have conducted sensitivity analyses
setting all e to zero and also reducing the standard deviation of s by 50%. These results of these
are given in Table 6-16. Setting all e to zero reduces the percent of children ages 5 to 18 with one
or more FEVi decrements > 10% from 32% to 18%, for the scenario modeled (Atlanta, 2006 Os
season). Reducing the standard deviation of s by 50% reduces these percents from 32% to 20%.
The assumption of no measurement error in Var(s) has the potential to significantly affect the
risk results.
The intra-individual variability term e in equation 6-3 is assumed by the MSS threshold
model to have a Gaussian distribution with mean zero and estimated standard deviation 4.135.
As done for U, we truncate this variability term distribution at ±2 standard deviations (±8.27). To
look at the effect of truncating the intra-individual variability distribution, we conducted
simulations using the threshold model with this constraint relaxed. When fitting the MSS
threshold model, the actual values of the intra-individual variability term range from about -20
to 30 (McDonnell et al., 2012, Figure 3 A). Truncating the distribution of s at ±12, 16, and 20
gives the results in Table 6-16, for ages 18 to 35.
We see that this constraint has a very large effect on the results for percents of the
population with FEVi decrements > 10 and 15% and less of an effect for 20%. The percent of
children with one or more FEVi decrements > 10% increases from 31% to 92% when increasing
the truncation point from 8.27 to 20. The estimates of six or more FEVi decrements > 10% are
more stable, and therefore more reliable, increasing from 10% to 18% when increasing the
truncation point from 8.27 to 20. Note that if the truncation limit is > 10, then there would be a
nonzero probability of response >10% AFEVi at zero Os concentration (which would not be
inconsistent with the observed clinical data).
Clearly the intra-individual variability Var(s) in the MSS model is a key parameter and is
influential in predicting the proportions of the population with FEVi decrements > 10 and 15%.
The assumption that the distribution of this term is Gaussian is convenient for fitting the model,
but is not accurate. The extent to which this mis-specification affects the estimates of the
parameters of the MSS model and its predictions is not clear.
6-43
-------
Table 6-16. Sensitivity Analysis of the Intra-individual Variability Term £ using the MSS
Threshold Model. Percents of the population aged 5 to 18 with one or more and 6 or more
days during the Os season with lung function (FEVi) decrements more than 10,15, and
20% (Atlanta 2006 base case, monitors air quality).
Description
Base case, a(e) = 4.135
Truncation of e at ±8.27
Truncation of e at ±12
Truncation of e at ±16
Truncation of e at ±20
O(E) = 0 (all e = 0)
a(£) - 50% = 2.0665
Truncation of e at ±4.133
1 or
more
AFEVi
>10%
31 .77%
91.71%
91.71%
91.71%
17.72%
20.11%
1 or
more
AFEVi
>15%
12.66%
15.16%
19.00%
19.00%
9.38%
10.04%
1 or
more
AFEVi
>20%
6.55%
7.15%
7.21%
7.23%
5.21%
5.50%
6 or
more
AFEVi
>10%
9.55%
18.19%
18.19%
18.19%
6.07%
6.63%
6 or
more
AFEVi
>15%
3.08%
3.41%
3.44%
3.44%
2.41%
2.60%
6 or
more
AFEVi
>20%
1 .22%
1 .29%
1 .29%
1 .29%
1.01%
1 .06%
McDonnell et al. (2013) have introduced another version of their model which assumes
that Var(s) increases with median response. So, with a fixed ventilation rate, Var(s) will be larger
for higher exposure concentrations and smaller for lower exposure concentrations. Simulations
using their preferred model (Model 3) yield risk results higher than results based on the threshold
model used for this HREA; those results are not presented here.
6.5.2 Convergence of APEX Results
APEX accounts for several sources of variability by drawing random variables from
specified distributions. Some variables are drawn once for each simulated individual (e.g., age,
location of residence), some are drawn every day or every hour for each simulated individual,
and others are drawn more frequently, at the event level (e.g., activity). Increasing the number of
individuals simulated in an APEX run increases the accuracy of the modeled variability and the
results of the APEX runs are more reproducible. In order to assess the number of individuals to
simulate to achieve convergence of APEX results, we perform multiple APEX runs with
identical inputs except for the random number seed, and look at the variability of the results of
these model runs. Table 6-17 summarizes the results of 40 APEX simulations of the Atlanta
2006 base case with 200,000 simulated individuals. For each of these measures, the range of
results over the 40 APEX runs is less than one percent. This analysis of the convergence of
APEX results shows that modeling 200,000 simulated individuals is adequate for reasonable
convergence of the FEVi risk measures.
6-44
-------
Table 6-17. Convergence results for the Atlanta 2006 base case with 200,000 simulated
individuals. Percents of the population by age group with one or more days (and six or
more days) during the Os season with lung function (FEVi) decrements more than 10,15,
and 20%. Minimum and maximum values and ranges over 40 APEX runs.
Age group
AFEVi > 10%
Min
Max
Range
AFEVi > 15%
Min
Max
range
AFEVi > 20%
Min
Max
range
1 or more days in the season
5 to 18
19 to 35
36 to 55
31.3%
11.1%
3.54%
32.1%
1 1 .5%
3.79%
0.88%
0.39%
0.25%
12.4%
3.00%
0.55%
12.9%
3.26%
0.68%
0.49%
0.26%
0.13%
6.21%
1.11%
0.13%
6.71%
1 .32%
0.20%
0.50%
0.22%
0.07%
6 or more days in the season
5 to 18
19 to 35
36 to 55
9.28%
1 .09%
0.22%
9.73%
1 .25%
0.30%
0.45%
0.16%
0.08%
2.80%
0.15%
0.01%
3.18%
0.21%
0.03%
0.38%
0.06%
0.02%
1.11%
0.03%
0.00%
1 .37%
0.06%
0.01%
0.27%
0.03%
0.01%
6.5.3 Application of Model for All Lifestages
The E-R functions derived from controlled human exposure studies involving 18-35 year
old subjects were used to estimate responses for school-aged children (ages 5-18). This was in
part justified by the findings of McDonnell et al. (1985) who reported that children 8-11 years
old experienced FEVi responses similar to those observed in adults 18-35 years old when both
groups were exposed to 120 ppb Os at an EVR of 32-35 L/min/m2. In addition, a number of
summer camp studies of school-aged children exposed in outdoor environments in the Northeast
also showed Os-induced lung function changes similar in magnitude to those observed in
controlled human exposure studies using adults, although the studies may not directly
comparable. The MSS model predicts increasing responsiveness with younger participants in the
age range of 18-35 years, as shown in Figure 6E-4 (Appendix 6E), which might indicate that
responsiveness would continue to increase as age decreases from 18. In extending the MSS
model to children, we fixed the age term in the model at its highest value, the value for age 18. If
continuing the MSS model trend were to accurately describe continued increased response in
children, then the fixed age term for children may have underestimated the effects on children,
and particularly younger children. On the other hand, if FEVi responses for children are similar
to those observed in adults 18-35 years old, as the evidence suggests, then our approach to
extending the age term (using the most responsive age of the range of 18-35) would overestimate
the response to children (see Table 6E-3 in Appendix 6E).
6-45
-------
In considering extending the MSS model to ages older than 36, we note that, in general,
Os responsiveness steadily declines for people aged 35-55, with people >55 eliciting minimal
responsiveness (ISA, section 6.2.1.1). As described in Section 6.2.4, we extended the age term
from the value at 36 linearly to zero at age 55, and set it to zero for ages above 55 (see Table
6-2). The uncertainty of this extrapolation may be substantial, but these age groups are not the
primary focus in the clinical risk assessment.
6.5.4 Application of Model for Asthmatic Children
The risk assessment used the same E-R relationship, developed from data collected from
healthy study subjects, and applied it to all people, children, and asthmatic children. Based on
limited evidence from a few human exposure studies, it is likely that subjects having asthma are
at least as sensitive to acute effects of Os as other subjects not having this health condition (ISA,
page 6-20 to 6-21). An analysis by Romieu et al. (2002) indicated a larger Os-associated
decrement in FEVi among children with moderate to severe asthma than among all children with
asthma (ISA, page 6-54). This suggests that the lung function decrements presented in this
assessment for asthmatic children may be underestimated. The magnitude of influence this
element might have on our risk estimates remains unknown at this time. In addition, asthmatic
children may have less reserve lung capacity to draw upon when faced with decrements, and
therefore a >10% decrement in lung function may be a more adverse event in an asthmatic child
than a healthy child.
6.5.5 Interaction Between Ozone and Other Pollutants
Because the controlled human exposure studies used in the risk assessment involved only
Os exposures, it was assumed that estimates of Os-induced health responses would not be
affected by the presence of other pollutants (e.g., SO2,PM2.5, etc). The magnitude of influence
that potential interactions might have on our risk estimates remains unknown at this time.
6.5.6 2000 vs. 2010 Population Demographics
The population demographics used in the exposure modeling are based on the 2000
Census. These are tract-level counts of population by age and sex, tract-level employment
probabilities by age and sex, and tract-to-tract commuting patterns. We are modeling the years
2006 to 2010, and some uncertainty will result from using demographics for a different year than
being modeled. To assess the extent of this uncertainty, APEX runs were conducted for seven
study areas using the 2010 Census and the results compared with APEX runs based on 2000
Census data. Table 6-18 presents the percents of children (ages 5 to 18) with one or more FEVi
decrements > 10, 15, 20%, for the 2010 modeling year, existing standard scenario, for APEX
runs based on 2000 and 2010 Census data. Table 6-19 gives the relative changes of the results in
6-46
-------
Table 6-18 from using 2000 demographics to using 2010 demographics. The largest relative
changes occur in Cleveland and Detroit. For example, in Detroit, 1.845% of children have at
least one AFEVi > 20% using 2000 Census data, while that number is 1.723% using 2010
Census data. This is a relative reduction of 6.6 percent.
Table 6-18. Comparison of 2000 vs. 2010 Census Demographics. Percents of all School-
aged Children (ages 5 to 18) with One or More Days during the Os Season with Lung
Function (FEVi) Decrements more than 10,15, and 20% (7 study areas, 2010 existing
standard air quality).
Study Area
Atlanta
Boston
Cleveland
Detroit
Houston
Los Angeles
Sacramento
2000 Census Demographics
AFEVi >10%
17.57%
14.32%
14.20%
16.50%
16.34%
16.01%
1 1 .25%
AFEVi >15%
4.55%
3.76%
3.43%
4.62%
4.61%
3.60%
2.44%
AFEVi >20%
1 .74%
1 .40%
1 .24%
1 .85%
1 .83%
1 .20%
0.85%
2010 Census Demographics
AFEVi >10%
17.08%
14.07%
13.79%
15.93%
16.23%
15.57%
1 1 .22%
AFEVi >15%
4.40%
3.67%
3.24%
4.46%
4.54%
3.49%
2.56%
AFEVi >20%
1 .67%
1 .42%
1.17%
1 .72%
1 .83%
1.21%
0.83%
Table 6-19. Comparison of 2000 vs. 2010 Census Demographics. Changes from using 2000
Demographics to using 2010 Demographics of Percents of all School-aged children (ages 5
to 18) with One or More Days during the Os Season with Lung Function (FEVi)
Decrements more than 10,15, and 20% (7 study areas, 2010 existing standard air quality).
Study Area
Atlanta
Boston
Cleveland
Detroit
Houston
Los Angeles
Sacramento
AFEVi >10%
(2.8%)
(1 .8%)
(2.9%)
(3.5%)
(0.7%)
(2.8%)
(0.3%)
AFEVi >15%
(3.2%)
(2.6%)
(5.5%)
(3.6%)
(1.4%)
(3.1%)
5.1%
AFEVi >20%
(4.3%)
1 .4%
(5.9%)
(6.6%)
0.0%
0.4%
(1 .8%)
6-47
-------
6.5.7 Qualitative Assessment of Uncertainty
EPA staff have identified key sources of uncertainty with respect to the lung function risk
estimates. These are: the physiological model in APEX for ventilation rates, the Os exposures
estimated by APEX, the MSS model applied to ages 18 to 35, and extrapolation of the MSS
model to children ages 5 to 18. The first two of these are discussed in Chapter 5. At this time we
do not have quantitative estimates of uncertainty for any of these. Table 6-20 provides a
qualitative assessment of the uncertainty resulting from each of these key sources. The primary
source of uncertainty is the MSS model, applied to ages 18 to 35.
6-48
-------
Table 6-20. Summary of Qualitative Uncertainties of Key Modeling Elements in the Os Lung Function Risk Assessment
Source
Description
Potential influence of
uncertainty on risk
estimates
Direction Magnitude
Knowledge-
Base
uncertainty*
Comments
The physiological
model in APEX for
ventilation rates
The physiological model in APEX takes
into account the population distribution
of individual physiological
characteristics and activities and
models minute-by-minute ventilation
rates for each simulated individual
using a series of physiological
relationships known with varying
degrees of certainty.
Over
Low-
Medium
Low-Medium
Ventilation rates are a key input to the MSS model.
Figure 6E-3 in Appendix 6E gives an overview of the
physiological model in APEX for ventilation rates.
Comparisons with ventilation rates reported in the literature
show fairly good agreement with APEX ventilation rates
(Section 5.4.4).
Os exposures
The Os exposures estimated by APEX
and their uncertainties are discussed in
Chapter 5.
Both
Low-
Medium
Low
exposures are a key input to the MSS model.
The McDonnell-
Stewart-Smith (MSS)
FEV-i model for ages
18 to 35
The MSS model is integrated into
APEX and predicts FEV-i decrements
for each simulated individual.
Both
Medium-
High
Low
There is a good conceptual foundation for the structure of
this model, but the variability in measurements of FEV-i and
estimated parameters of the model introduce uncertainty
into the model predictions of large FEV-i decrements. The
estimated parameters have fairly wide confidence intervals
(Table 6-14) and the risk results are sensitive to varying the
parameters (Figure 6-12).
The most influential parameter is (36, the power to which
ventilation rate is raised in the MSS model. An increase of
five percent in (36 leads to a 27 percent increase in the
modeled number of children with FEV-i decrements > 10%.
(The 95 percent confidence interval of this parameter
estimate is ±14%)
We are unable to quantitatively assess the uncertainty of the
MSS model, since all of the clinical data used by McDonnell
and coworkers to fit the MSS model are not available to
EPA (in particular, some studies by Adams and coworkers).
6-49
-------
Source
Description
Potential influence of
uncertainty on risk
estimates
Direction Magnitude
Knowledge-
Base
uncertainty*
Comments
Inter-individual
(between-individual)
variability in the MSS
FEVi model
Inter-individual variability (Var(U))
reflects the variability in
responsiveness of individuals to Os.
Both
Medium
Low
In order to estimate Var(U), it is necessary to have repeated
clinical trials using the same individuals. There are few of
these in the data used to fit the MSS model, making it
difficult to reliably estimate Var(U). Furthermore, the method
used for adjusting for filtered air (FA) exposures in the data
used to fit the MSS model does not use the subject-specific
adjustments that are typically used in these clinical studies.
Rather, the mean FA response across a study is used to
adjust the Os responses of each subject in the study. This
adds additional uncertainty to the estimate of Var(U).
If Var(U) is underestimated, then the estimates of the
number of people experiencing FEVi decrements > 10, 15,
20% will be overestimated (absent other sources of
uncertainty). This is because higher Var(U) gives more
between-individual variability and less within-individual
variability, so more responsive individuals are more likely to
see repeated occurrences of high AFEV-i and less
responsive individuals are more likely to see no occurrences
ofhighAFEV-i.
There are few clinical data for population with diseased
lungs (i.e., asthma) and the MSS model does not account
for the increase in inter-individual variability that would
reflect this or the increased variability due to genetic factors.
This increase in inter-individual variability would not be
offset with a decrease in intra-individual variability, and thus
would tend to increase the estimates of numbers of people
at risk for large lung function decrements. This is a key
element missing from the MSS model.
6-50
-------
Source
Description
Potential influence of
uncertainty on risk
estimates
Direction Magnitude
Knowledge-
Base
uncertainty*
Comments
Intra-individual (within-
individual) variability in
the MSS FEVi model
The intra-individual variability Var(£)
has two basic components: (1) the
intra-individual variability of the true
response to Os (both within-day and
between-day) and (2) measurement
error. These cannot be distinguished
based on the available data. We are
assuming that all of this variability is
due to the true response, which will
(absent other uncertainties) tend to
overestimate the response to Os.
Over
Medium-
High
Low
The variability term £ is assumed by the MSS model to have
a Gaussian distribution with mean zero and estimated
standard deviation 4.135. Since the actual values are
bounded, we truncate the variability term distribution at ±2
standard deviations (±8.27). To look at the effect of
truncating the variability distribution, we conducted
sensitivity analyses with the variability term truncated at
different levels. We find that this constraint has a large effect
on estimates of percents of the population with FEVi
decrements > 10 and 15% and less of an effect for 20%.
The percent of children with FEVi decrements > 10%
increases from 31% to 92% when increasing the truncation
point from 8.27 to 20. The estimates of six or more FEVi
decrements > 10% are more stable, and therefore more
reliable, increasing from 10% to 18% when increasing the
truncation point from 8.27 to 20. If the truncation point used
is too low, then (absent other uncertainties) this will tend to
underestimate the response to Os.
Clearly the intra-individual variability Var(£) in the MSS
model is a key parameter and is influential in predicting the
proportions of the population with FEVi decrements > 10
and 15%. The assumption that the distribution of this term is
Gaussian is convenient for fitting the model, but is not
accurate. The extent to which this mis-specification affects
the estimates of the parameters of the MSS model and its
predictions is not clear.
Extrapolation of the
MSS model to children
(ages 5 to 18)
The MSS model is based on studies
with subjects ranging in age from 18 to
35 years; therefore prediction for
children involves assumptions for
extrapolation of the MSS model for
individuals <18 years of age.
Both
Medium
Low
Summer camp studies and one clinical study of children
indicate that FEVi responses for children are similar to
those observed in adults 18-35 years old. See discussion in
Section 6.5.3.
* Refers to the degree of uncertainty associated with our understanding of the phenomenon, in the context of assessing and characterizing its uncertainty
6-51
-------
6.6 KEY OBSERVATIONS
This lung function risk assessment evaluated risks of lung function decrements due to Os
exposure for three age study groups: school-aged children ages 5 to 18, young adults ages 19 to
35, and adults ages 36 to 55. Lung function risks were estimated for 15 study areas: Atlanta,
Baltimore, Boston, Chicago, Cleveland, Dallas, Denver, Detroit, Houston, Los Angeles, New
York, Philadelphia, Sacramento, St. Louis, and Washington, DC. We estimated the percent of
the study groups experiencing a reduction in lung function for three different levels of impact,
10, 15, and 20% decrements in FEVi. These levels of impact were selected based on the
literature discussing the adversity associated with these types of lung function decrements (U.S.
EPA, 2012, Section 6.2.1.1; Henderson, 2006).
• Individual versus Population-based Exposure-Response Model: Two models were
used, one based on application of an individual level E-R function (the MSS model
introduced in this review) and one based on application of a population level E-R
function consistent with the model used in the previous Os review which applies
probabilistic population-level E-R relationships for lung function decrements (measured
as percent reductions in FEVi) associated with 8-hr moderate exertion exposures. The
MSS model is preferred, due to its ability to model individual exposures for a wide range
of exposure times and levels of exercise as well as for exposures with time-varying
exposures and levels of exercise (Section 6.2.4; ISA pages 6-15 to 6-16). The MSS model
risk estimates are significantly higher than those estimated using the E-R approach. For
lung function responses greater than 10, 15, and 20% the MSS model gives results
typically a factor of three higher than the E-R model for school-aged children. Both
models give higher responses for higher concentrations, compared to lower
concentrations, as can be seen in Figures 6-6, 6-9, and 6-10.
• Study Group Variability in Estimated Risk: When considering the study groups
characterized by ages, school-aged children had the greatest percent of persons
experiencing FEVi decrements (e.g., Table 6-4). Adults >55 years old have minimal Os-
induced lung function risk considering any of the air quality scenarios. The MSS model
was applied to estimate lung function risk for outdoor workers (ages 19-35) in Atlanta for
one year (2006). The proportion of outdoor workers with FEVi decrements > 15% ranges
from 3.6 to 5.3 times the proportion of the general population (ages 19-35) with FEVi
decrements > 15% across the different standards simulated. The proportion of outdoor
workers with multiple occurrences of FEVi decrements > 15% is much greater than for
the general population.
6-52
-------
Year-to-Year Variability in Estimated Risk: Based on the MSS model, the percents of
population estimated to experience lung function responses greater than 10, 15, and 20%,
associated with Cb exposure while engaged in various levels of exertion, vary
considerably for different years and study areas under the recent air quality scenarios and
also for the existing and alternative standard scenarios (Figure 6-7 and Figure 6-8, Table
6-4 and Table 6-5). For instance, the estimates for > 10% FEVi decrement for all school-
aged children for recent Os concentrations range across study areas and years from 11 to
31 percent, and range from 11 to 22 percent after simulating just meeting the existing
standard. The estimates for > 15% FEVi decrement for all school-aged children for recent
Os concentrations range across study areas and years from 2 to 12 percent, and range
from 2 to 6 percent after simulating just meeting the existing standards. The estimates for
> 20% FEVi decrement for recent Os concentrations range across study areas and years
from 1 to 6 percent, and range from 1 to 3 percent after simulating just meeting the
existing standards.
Impact of Alternative Air Quality Scenarios: Figure 6-13 displays the risks and the
incremental increases in risk for increasing standard levels, where risk is taken to be the
highest value for each study area (over years) of the percent of school-aged children with
FEVi decrement > 10%. The risks in this figure for Washington, DC, for example, are
about 9.6% for the alternative standard level of 60 ppb and 13.4% for the alternative
standard level of 65 ppb. The length of the orange bar is the incremental risk (3.8%) in
going from the 60 ppb to the 65 ppb alternative standards. This figure shows that there
are significant increases in incremental risk for all 15 study areas in the progression of
alternative standard levels from 60 ppb to the level of the existing standard, 75 ppb. The
pattern of reductions for lung function decrements larger than 15 and 20% are similar. As
discussed in Section 4.3.1, the New York 60 ppb alternative standard was not modeled
and the risk for NY for that scenario would not necessarily be zero. Figure 6-14 displays
the risks and the incremental increases in risk for increasing standard levels, where risk is
taken to be the mean value for each study area (over years) of the percent of school-aged
children with FEVi decrement > 10%.
6-53
-------
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St Louis
Washington
0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20% 22% 24%
percent of school-aged children with FEV1 decrement > 10%
standard level (ppb) i i 60 i i 65 i i 70 i • 75
Figure 6-13. Lung Function Risk Results, Incremental Increases In Risk For Increasing Standard Levels: Percent of All
School-aged Children With FEVi Decrement > 10%, Highest Value For Each Study area Over Years6
6 New York level 60 was not modeled. We do not know what the percent risk would be for NY under the 60 ppb alternative standard, but it would not necessarily
be zero.
6-54
-------
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St Louis
Washington
0%
2%
18% 20%
4% 6% 8% 10% 12% 14% 16%
percent of school-aged children with FEV1 decrement > 10%
standard level (ppb) I I 60 I I 65 I I 70 I I 75
Figure 6-14. Lung Function Risk Results, Incremental Increases In Risk For Increasing Standard Levels: Percent of All
School-aged Children With FEVi Decrement > 10%, Mean Value For Each Study Area Over Years7
7 New York level 60 was not modeled. We do not know what the percent risk would be for NY under the 60 ppb alternative standard, but it would not necessarily
be zero.
6-55
-------
6.7 REFERENCES
Adams, W.C. 2002. Comparison of chamber and face-mask 6.6-hour exposures to ozone on
pulmonary function and symptoms responses. Inhalation Toxicology. 14:745-764.
Adams, W.C. 2003. Comparison of chamber and face mask 6.6-hour exposure to 0.08 ppm
ozone via square-wave and triangular profiles on pulmonary responses. Inhalation
Toxicology. 15:265-281.
Adams, W.C. 2006. Comparison of chamber 6.6-h exposures to 0.04-0.08 ppm ozone via square-
wave and triangular profiles on pulmonary responses. Inhalation Toxicology. 18:127-136.
American Thoracic Society. 2000. What constitutes an adverse effect of air pollution?. American
Journal of Respiratory and Critical Care Medicine. 161:665-673.
Bittar, E. Edward, editor. 2002. Pulmonary Biology in Health and Disease. Springer, New York.
Box, G.E.P. and G.C. Tiao. 1973. Bayesian Inference in Statistical Analysis. Wesley Publishing
Co.: Wiley Classic Library, ISBN 0-471-57428-7.
Burmaster, D.E. 1998. Lognormal distributions for skin area as a function of body weight. Risk
Analysis. 18(l):27-32.
Dietert, R.R.; R.A. Etzel; D. Chen, et al. 2000. Workshop to identify critical windows of
exposure for children's health: immune and respiratory systems work group summary."
Environ Health Perspect. 108(S3):483-490.
Esmail, S.; Y. Bhambhani and S. Brintnell. 1995. Gender differences in work performance on the
Baltimore therapeutic equipment work simulator. American Journal of Occupational
Therapy. 49:405-411.
Federal Register. 1996. Responses to Significant Comments on the 1996 Proposed Rule on the
National Ambient Air Quality Standards for Ozone (December 13, 1996; Fed. Reg. 61
FR 65716).
Folinsbee, L.J., et al. 1988. Pulmonary function and symptom responses after 6.6-hour exposure
to 0.12 ppm ozone with moderate exercise. Journal of the Air Pollution Control
Association. 38:28-35.
Graham, S.E. and T. McCurdy. 2005. Revised Ventilation Rate (VE) Equations for Use in
Inhalation-oriented Exposure Models. Washington, DC: EPA Office of Air and
Radiation. (EPA document number EPA/600/X-05/008, Appendix A).
.
Hazucha, M.J.; LJ. Folinsbee and P.A. Bromberg. 2003. Distribution and reproducibility of
spirometric response to ozone by gender and age. Journal of Applied Physiology.
95(5)1917-1925.
Hislop, A. A. 2002. Airway and blood vessel interaction during lung development. Journal of
Anatomy. 201:325-334. Available at:
.
Horstman, D.H. et al. 1990. Ozone concentration and pulmonary response relationships for 6.6-
hour exposures with five hours of moderate exercise to 0.08, 0.10, and 0.12 ppm.
American Review of Respiratory Disease. 142:1158-1163.
6-56
-------
Isaacs, K.; G. Glen; T. McCurdy and L. Smith. 2008. Modeling energy expenditure and oxygen
consumption in human exposure models: accounting for fatigue and EPOC. Journal of
Exposure Science and Environmental Epidemiology. 18:289-298.
Isaacs, K. and L. Smith. 2005. New Values for Physiological Parameters for the Exposure Model
Input File Physiology, text memorandum submitted to the U.S. Environmental Protection
Agency under EPA Contract EP-D-05-065. NERL WA 10. Alion Science and
Technology. Found in US EPA. (2009). Risk and Exposure Assessment to Support the
Review of the SO 2 Primary National Ambient Air Quality Standard. (EPA document
number EPA-452/R-09-007, August 2009).
.
Kim, C.S.; N.E. Alexis; A.G. Rappold; H. Kehrl; MJ. Hazucha; J.C. Lay; M.T. Schmitt; M.
Case; R.B. Devlin; D.B. Peden; D. Diaz-Sanchez. 2011. Lung function and inflammatory
responses in healthy young adults exposed to 0.06 ppm ozone for 6.6 hours. American
Journal of Respiratory and Critical Care Medicine. 183:1215-1221.
Linn, W.S.; R.D. Buckley; C.E. Spier; R.L. Blessey; M.P. Jones; D.A. Fischer and J.D. Hackney.
1978. Health effects of ozone exposure in asthmatics. The American Review of
Respiratory Disease. 117(5)835-843.
Lunn, D.; C. Jackson; N. Best; A. Thomas and D. Spiegelhalter. 2012. The BUGS Book-A
Practical Introduction to Bayesian Analysis. CRC Press: Chapman and Hall.
McCurdy, T.; G. Glen; L. Smith; Y. Lakkadi. 2000. The National Exposure Research
Laboratory's Consolidated Human Activity Database. Journal of Exposure Analysis and
Environmental Epidemiology. 10:566-578.
McDonnell, W.F.; R.S. Chapman; M.W. Leigh; G.L. Strope; A.M. Collier. 1985. Respiratory
responses of vigorously exercising children to 0.12 ppm ozone exposure. American
Review of Respiratory Disease. 132:875-879.
McDonnell, W.F.; H.R. Kehrl; S. Abdul-Salaam; P.J. Ives; L.J. Folinsbee; R.B. Devlin; J.J.
O'Neil; D.H. Horstman. 1991. Respiratory response of humans exposed to low levels of
ozone for 6.6 hours. Archives of Environmental Health. 46(3): 145-150.
McDonnell, W.F. et al. 1993. Predictors of individual differences in acute response to ozone
exposure. American Review of Respiratory Disease. 147:818-825.
McDonnell, W.F. et al. 1997. Prediction of ozone-induced FEVi changes. American Journal of
Respiratory and Critical Care Medicine. 156:715-722.
McDonnell, W.F.; P.W. Stewart and M.V. Smith. 2007. The temporal dynamics of ozone-
induced FEVi changes in humans: an exposure-response model. Inhalation Toxicology.
19:483-494.
McDonnell, W.F.; P.W. Stewart and M.V. Smith. 2010. Prediction of ozone-induced lung
function responses in humans. Inhalation Toxicology. 22(2): 160-8.
McDonnell, W.F.; P.W. Stewart; M.V. Smith; C.S. Kim andE.S. Schelegle. 2012. Prediction of
lung function response for populations exposed to a wide range of ozone conditions.
Inhalation Toxicology. 24:619-633.
6-57
-------
McDonnell, W.F.; P.W. Stewart and M.V. Smith. 2013. Ozone exposure-response model for
lung function changes: an alternate variability structure. Inhalation Toxicology. 25:348-
353.
Narayanan, M.; J. Owers-Bradley; C.S. Beardsmore; M. Mada; I. Ball; R. Garipov; K.S. Panesar;
C.E. Kuehni; B.D. Spycher; S E. Williams; M. Silverman. 2012. Alveolarization
continues during childhood and adolescence: new evidence from helium-3 magnetic
resonance." American Journal of Respiratory and Critical Care Medicine. 185(2):86-
191. Available at: .
Passannante, A.N.; MJ. Hazucha; P.A. Bromberg; E. Seal; L. Folinsbee; G. Koch. 1998.
Nociceptive mechanisms modulate ozone-induced human lung function decrements.
Journal of Applied Physiology. 85:1863-1870.
Romieu, I; JJ. Sienra-Monge; M. Ramirez-Aguilar; M.M. Tellez-Rojo; H. Moreno-Macias; N.I.
Reyes-Ruiz; B.E. Del Rio-Navarro; M.X. Ruiz-Navarro; G. Hatch; R. Slade; M.
Hernandez-Avila. 2002. Antioxidant supplementation and lung functions among children
with asthma exposed to high levels of air pollutants. American Journal of Respiratory
and Critical Care Medicine. 166:703-709.
Samet, J.M. 2011. Clean Air Scientific Advisory Committee (CASAC) Response to Charge
Questions on the Reconsideration of the 2008 Ozone National Ambient Air Quality
Standards. (Report number EPA-CASAC-11-004), dated March 30, 2011. Available at:
.
Schelegle, E.S.; C.A. Morales; W.F. Walby; S. Marion; R.P. Allen. 2009. 6.6-hour inhalation of
ozone concentrations from 60 to 87 ppb in healthy humans. American Journal of
Respiratory and Critical Care Medicine. 180:265-272.
Schelegle, E.S.; W.C. Adams; W.F. Walby and S. Marion. (2012). Modeling of individual
subject ozone exposure response kinetics. Inhalation Toxicology. 24(7):401-415.
Schofield, W.N. 1985. Predicting basal metabolic rate, new standards and review of previous
work. Clinical Nutrition. 39C(suppl):5-41.
Seal, Jr., E.; W.F. McDonnell; D.E. House. 1996. Effects of age, socioeconomic status, and
menstrual cycle on pulmonary response to ozone. Archives of Environmental and
Occupational Health. 51:132-137.
U.S. EPA. 1996. Review of National Ambient Air Quality Standards for Ozone: Assessment of
Scientific and Technical Information - OA QPS Staff Paper. Research Triangle Park, NC:
EPA Office of Air Quality Planning and Standards. (EPA document number EPA/452/R-
96-007). Available at: .
U.S. EPA. 2007a. Review of National Ambient Air Quality Standards for Ozone: Policy
Assessment of Scientific and Technical Information - OA QPS Staff Paper. Research
Triangle Park, NC: EPA Office of Air Quality Planning and Standards. (EPA document
number EPA-452/R-07-007). Available at:
.
6-58
-------
U.S. EPA. 2007b. Ozone Population Health Risk Assessment for Selected Urban Areas.
Research Triangle Park, NC: EPA Office of Air Quality Planning and Standards. (EPA
document number EPA-452/R-07-009). Available at:
.
U.S. EPA. 2009. Metabolically Derived Human Ventilation Rates: A Revised Approach Based
Upon Oxygen Consumption Rates. Washington, DC: EPA National Center for
Environmental Assessment. (EPA document number EPA/600/R-06/129F). Available at:
.
U.S. EPA. 2011. O 3 National Ambient Air Quality Standards: Scope and Methods Plan for
Health Risk and Exposure Assessment. Research Triangle Park, NC: EPA Office of Air
Quality Planning and Standards. (EPA document number EPA-452/P-11-001). Available at:
.
U.S. EPA. 2012a. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
Documentation (TRIM. Expo / APEX, Version 4.4) Volume I: User's Guide. Research
Triangle Park, NC: EPA Office of Air Quality Planning and Standards. (EPA document
number EPA-452/B-12-001a). Available at:
.
U.S. EPA. 2012b. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
Documentation (TRIM. Expo / APEX, Version 4.4) Volume II: Technical Support Document.
Research Triangle Park, NC. (EPA document number EPA-452/B-12-001b). Available at:
.
U.S. EPA. 2013. Integrated Science Assessment of Ozone and Related Photochemical Oxidants.
EPA National Center for Environmental Assessment. (EPA document number EPA/600/R-
10/076F, 2013). Available at:
.
WinBUGS, version 1.4.3. Available at:
.
6-59
-------
This page left intentionally blank
6-60
-------
7 CHARACTERIZATION OF HEALTH RISK BASED ON
EPIDEMIOLOGICAL STUDIES
This chapter provides an overview of the methods used to estimate health risks in
selected urban areas based on application of results of epidemiology studies. Section 7.1.1
discusses the basic structure of the risk assessment, identifying the modeling elements and
related sources of input data needed for the analysis and presenting an overview of the approach
used in calculating health effect incidence using concentration-response (C-R) functions based
on epidemiological studies. Section 7.2 discusses air quality considerations. Section 7.3
discusses the selection of model inputs including: (a) selection of urban study areas, (b) selection
of epidemiological studies and specification of C-R functions, (c) specification of baseline health
effect incidence and prevalence rates, and (d) estimation of population (demographic) counts.
Section 7.4 describes how uncertainty and variability are addressed in the risk assessment,
including specification of the sensitivity analyses completed for the risk assessment and how
these differ from the core risk estimates. Section 7.5 summarizes the risk estimates that are
generated, including both the core estimates and sensitivity analyses. Finally, Section 7.6
provides an assessment of overall confidence in the risk assessment together with a set of key
observations regarding the risk estimates generated.
7.1 GENERAL APPROACH
7.1.1 Basic Structure of the Risk Assessment
This risk assessment involves the estimation of the incidence of specific health effect
endpoints associated with exposure to ambient Os for defined populations located within a set of
urban study areas. Because the risk assessment focuses on health effect incidence experienced by
defined populations, it represents a form of population-level risk assessment and does not
estimate risks to individuals within the population. Furthermore, because it models risk for
populations within a set of selected urban study areas, it is not intended to provide an estimate of
national-scale risk.1
The general approach used in this component of the Os risk assessment relies on C-R
functions based on effect estimates and model specifications obtained from epidemiological
studies. Because these studies derive effect estimates and model specifications using averages of
ambient air quality data from fixed-site, population-oriented monitors, uncertainty arising from
the application of these functions in an Os risk assessment is decreased if, in modeling risk, we
1 Chapter 8 provides a limited assessment of national risk focused on the mortality burden associated with 2007
ambient Os concentrations.
7-1
-------
also use ambient air quality data at fixed-site, population-oriented monitors to characterize
exposure. Therefore, we developed a composite monitor for each urban study area to represent a
surrogate population exposure by averaging Os concentrations across the monitors in that study
area to produce a single composite hourly time-series of values. The Os metrics used in
evaluating risk are derived from the composite monitor hourly time-series distribution (see
sections 7.2 and Chapter 4 for additional detail on the characterization of ambient Os levels).2
The general Os health risk model used here, illustrated in Figure 7-1, combines Os air
quality data, C-R functions, baseline health incidence and prevalence data, and population data
(all specific to a given urban study area) to derive estimates of the annual incidence of specified
health effects for that urban study area attributable to Os exposure. Twelve urban study areas
were selected to provide coverage for the types of urban Os scenarios likely to exist across the
U.S. (see section 7.3.1). Chapter 8 provides an assessment of the degree to which risk-related
attributes in the 12 selected urban areas are representative of other urban areas in the U.S.
This assessment estimates health risk associated with recent ambient Os conditions and
for air quality adjusted to just meet the existing 8-hour (8-hr) Os standard, and additional
estimates of risk if alternative standards are just met, with an emphasis on reductions in risk
between just meeting the existing standard and just meeting alternative standards (the full set of
risk estimates, including simulation of risk under current conditions is presented in Appendix
7B). The alternative standard levels evaluated are 70, 65 and 60 ppb (expressed using the
existing form and averaging time of the Os standard).
We simulated just meeting the existing and alternative Os standards by adjusting hourly
Os concentrations measured over the Os season using a model-based adjustment methodology
that estimates Os sensitivities to precursor emissions changes.3 These sensitivities, which
estimate the response of Os concentrations to reductions in anthropogenic NOx and VOC
emissions, are developed using the Higher-order Decoupled Direct Method (HDDM) capabilities
in the Community Multi-scale Air Quality (CMAQ) model. More details on the HDDM-
adjustment approach is presented in Chapter 4 of this Health Risk and Exposure Assessment
(HREA) and in Simon et al. (2013).
As discussed in Chapters 2 and 3, in modeling risk we employ continuous non-threshold
C-R functions relating Os exposure to health effect incidence. The use of non-threshold functions
2 This holds for all air quality metrics used in modeling short-term mortality and morbidity endpoints. However, the
air metric used in modeling long-term mortality is based on a seasonal mean of maximum hourly values derived
for each Os monitor within an urban study area with those individual averages then combined to generate a single
seasonal average composite monitor value for each study area (see section 7.2 for more detail).
3 In the first draft of this HREA, we used a statistical quadratic rollback approach to simulate just meeting the
existing O3 standards. In that first draft, we proposed using the model based approach used in the current
document, and received support for the model based approach from CASAC (Frey and Samet, 2012).
7-2
-------
reflects the discussion of the relevant studies in the Ch ISA (U.S. EPA 2013a, section 2.5.4.4).
However, also consistent with the conclusions of the Os ISA, we recognize that the evidence
from the studies indicates less confidence in specifying the shape of the C-R function at Os
concentrations towards the lower end of the distribution of data used in fitting the curve due to
the reduction in the number of data points available. The Os ISA noted that the studies indicate
reduced certainty in specifying the shape of the C-R function specifically for short-term Os-
attributable respiratory morbidity and mortality, in the range generally below 20 ppb (for both 8-
hr maximum and 24-hr metrics) (Os ISA, section 2.5.4.4). However, care needs to be taken in
interpreting this range of reduced confidence indicated in the studies and applying it to the
interpretation of risk estimates generated for a specific urban study area. This is because there is
considerable heterogeneity in the effect of Os on mortality across urban study areas (Cb ISA,
section 6.6.2.3). Additionally, it is likely that levels of confidence associated with C-R functions
(including ranges of reduced confidence in specifying the function) also vary across urban study
areas reflecting underlying differences in factors impacting the exposure-response relationship
for Os, such as demographic differences and exposure measurement error. For these reasons, the
<20 ppb range discussed in the Os ISA should be viewed as a more generalized range to be
considered qualitatively or semi-quantitatively, along with many other factors, when
interpreting the risk estimates rather than as a fixed, bright-line.4
Based on comments we received from CAS AC on the 1st draft HREA (Frey and Samet,
2012), we are no longer including estimates of risk down to the lowest measured level (LML).5
Instead, through the use of heat map tables, we focus on providing estimates of total risk, and the
distribution of risk over concentrations of Os.6 Coupled with information about what the studies
indicate about the C-R function at lower Os concentrations, this provides for a more complete
understanding of confidence in estimated risk than simply truncating risk at the LML.
In modeling risk for all health endpoints included in the analysis, for recent Os conditions
and just meeting the existing standard, we estimated total risk (down to zero). For meeting the
4 This range of reduced confidence in the shape of the C-R function is most appropriately applied to area-wide
averages (i.e., composite monitor values) of the type often used in epidemiological studies rather than to the range
of O3 associated with a particular monitor. This reflects the fact that the observations presented in the O3 ISA are
themselves based on consideration for epidemiological studies which use composite monitor values.
5 Based on their November 19, 2012 letter commenting on the 1st draft HREA, CASAC recommended against
inclusion of risk estimates based on the LML in the core analysis due to the fact that there is little difference
between these estimates and risk estimates based on total O3 exposure and that LML information is not available
for many of the epidemiological studies used in the HREA (Frey and Samet, 2012). However, they recommended
a more limited exploration of the LML and its implications for risk for one or more areas. In response, we have
included coverage for LML as part of our discussion of the heat maps results presented in section 7.5.1.
6 Heat map tables illustrate the distribution of estimated O3-related deaths across daily O3 levels for each urban study
and allow a quick visual comparison of trends (in the distribution of total O3 risk as we as risk reductions) across
ambient O3 ranges both within and across study areas (see section 7.5).
7-3
-------
existing and alternative standards, we estimated both total risk as well as the difference in risk,
representing the degree of risk reduction associated with just meeting the existing and alternative
standard levels. When calculating risk differences, we focus on comparing total risk after just
meeting each alternative standard with total risk after just meeting the existing standards. We
also evaluate the incremental change in risk from meeting increasingly lower alternative standard
levels. Risk results are presented in terms of absolute numbers and changes in the Cb-attributable
incidence of mortality and morbidity, and in terms of the percent of baseline mortality and
morbidity attributable to Os. We also provide risks per 100,000 population (to normalize risks
across urban areas with different size populations to facilitate comparisons).
As with previous NAAQS risk assessments, for this analysis we have generated two
categories of risk estimates, including a set of core (or primary) estimates and an additional set of
sensitivity analyses. The core risk estimates utilize C-R functions based on epidemiological
studies for which we have relatively greater overall confidence and which provide the best
coverage for the broader Os monitoring period (rather than focusing only on the summer season).
Although it is not strictly possible to assign quantitative levels of confidence to these core risk
estimates due to data limitations, they are generally based on inputs having higher overall levels
of confidence relative to risk estimates that are generated using other C-R functions. Therefore,
emphasis is placed on the core risk estimates in making observations regarding total risk and risk
reductions associated with recent conditions and after just meeting the existing and alternative
standard levels. By contrast, the sensitivity analysis results typically reflect application of C-R
functions covering a wider array of design elements which can impact risk (e.g., length of
season, co-pollutants models, lag structures, statistical modeling methods etc). The sensitivity
analysis results provide insights into the potential impact of these design elements on the core
risk estimates, thereby informing our characterization of overall confidence in the core risk
estimates.7 We have significantly expanded our sensitivity analysis relative to that completed for
earlier drafts of the HREA to address a wider range of modeling elements which can impact the
core risk estimates. Details of the design of the core and sensitivity analyses (including modeling
element composition) for each of the health effect endpoints categories covered in this risk
assessment are presented in section 7.4.3 and briefly summarized below.
7 In presenting both the core and sensitivity analysis, we include both point estimates and 95th percentile confidence
intervals (CIs). The 95th percentile CIs reflect the statistical fit of the underlying effect estimates and therefore
reflect the statistical power of the epidemiological studies supplying the effect estimates. Often in comparing
sensitivity analysis with core risk estimates, we focus not only on the point estimates, but also on the confidence
intervals since these inform our understanding of confidence in the respective risk estimates.
7-4
-------
For short-term exposure related mortality, our core analysis is based on application of C-
R functions obtained from the Smith et al. (2009) epidemiological study (see section 7.3.2). In
addition, we have completed an expanded array of sensitivity analyses which provide coverage
for a number of modeling elements including: (a) time period reflected in risk modeling (summer
season versus full monitoring period), (b) peak Os metric (8-hr maximum versus 8-hr mean) (c)
use of regional versus national-based Bayesian adjustment in deriving effect estimates,8 (d) use
of single (Os-only) versus co-pollutant (Os and PMio) models,9 (e) application of alternative C-R
functions based on Zanobetti and Schwartz (2008) (see section 7.3.2) and (f) size of the urban
study area (CBSA versus smaller multi-county study area)10 (see sections 7.4.3 and 7.5.3 for
additional detail on the sensitivity analyses completed). In addition to these sensitivity analyses,
we have considered alternative methods for adjusting air quality to just meet the existing and
alternative standards (NOx-only versus combination of VOC andNOx reductions). Additional
sensitivity analyses exploring lag structure may also provide useful information, but are not
possible due to the lack of availability of Bayes adjusted estimates for alternative lag structures.
For short-term exposure morbidity, we have effect estimates covering a wide range of
design elements including co-/single-pollutant models and lag structure. However, we were not
in a position to differentiate between these alternative model forms in terms of overall
confidence and have therefore included all of these estimates in the core analysis. This range of
risk estimates can also be viewed as a sensitivity analysis where there is no clear "core" estimate
and instead, the full range of risk estimates is considered to provide the best overall picture of
risk for a specific endpoint (see section 7.3.2 and 7.4.3).
8 Short-term Os-attributable mortality in this analysis is modeled using Bayesian-adjusted effect estimates. This
approach involves adjustment of each city's effect estimate using a prior distribution reflecting the Os-mortality
relationship seen across the broader set of cities considered in the epidemiological study. For the sensitivity
analysis, we compare the use of a national prior distribution (the core approach) for the Bayesian adjustment with
use of a regional prior.
9 The co-pollutants model results are limited by the reduced number of days with co-pollutants sampling (either 1 in
3 or 1 in 6) which makes it difficult to evaluate the statistical significance of these results in view of the large
posterior standard deviations (Smith et al., 2009). This increased uncertainty associated with the estimates
prevents these results from being treated as part of the core analysis. Never the less, they provide perspective on
the potential magnitude of risk associated with co-pollutants modeling and as such make an important
contribution as a sensitivity analysis for short-term O3-attributable mortality.
10 Core based statistical areas (CBSAs) are U.S. geographic areas defined by the Office of Management and Budget
(OMB). They include an urban center of at least 10,000 people combined with adjacent urban and suburban areas
that are socioeconomically tied to the urban center by commuting. CBS As tend to be significantly larger than the
study areas used in the epidemiological studies providing effect estimates. We have used risk estimates based on
CBS As in the core analysis in order to better represent the changes in risk that could be experienced in the broader
urban areas and to avoid the introduction of known bias into the risk assessment. We have included risk estimates
based on the smaller study areas from the original epidemiological studies as sensitivity analyses (see discussion
later in this section for additional detail).
7-5
-------
Our analysis also includes estimates of long-term exposure related respiratory mortality,
including a core estimate based on a co-pollutant model11 (with PIVh.s) together with sensitivity
analyses exploring regional heterogeneity in the effect estimate and application of a national-
level estimates focusing only on Os, analyses based on findings reported by Jerrett et al. (2009)
(see section 7.5.3). We have also included a sensitivity analysis exploring the impact of potential
thresholds in the C-R relationship on estimates of long-term exposure-related mortality that were
evaluated in Jerrett et al. (2009). The approach used and added study author interpretation of the
published model results are further described in Sasser (2014) and in section 7.3.2.
We modeled all core risk estimates using study areas based on the core-based statistical
area (CBS A) regardless of whether the epidemiological studies providing the effect estimates
used the CBSA spatial definition or a different spatial study area definition. The decision to use
CBSA-based study areas in all core simulations reflects our desire to better represent the changes
in risk that could be experienced in the urban areas and avoid introducing substantial known bias
into the risk estimates. As discussed in Chapter 4 (section 4.3.1.2), most nonattainment Os
monitors are not located in the center of the urban area, but instead are sited in the surrounding areas,
reflecting the transport and atmospheric chemistry governing Os formation. The monitors in the
urban core areas are usually most affected by local sources of NOX and experience lower
concentrations of Os since the NO is titrating Os in these areas. For these monitors, simulating just
meeting the existing and alternative standard levels can result in an increase in low level or mean
Os concentrations, while areas further out from the urban core experience the expected reduction
in the highest Os levels. Had we focused risk estimates on the smaller urban core areas used in
some of the epidemiological studies, we would not have fully captured the changes in risk
estimated to be experienced by the broader urban area since we would have been focusing only
on those areas experiencing net increases in Os (when simulating just meeting the existing and
alternative standard levels). By modeling risk for the core analysis using the more inclusive
CBSA study areas, we insure that risk estimates will include consideration both for the relatively
smaller core urban areas experiencing increases in Os as well as the broader urban and suburban
area experiencing risk reductions associated with reductions in the highest Os concentrations. We
will also insure that, to a greater extent, the analysis includes the county with the design value
monitor in the assessment of risk (see section 7.2).
There is a degree of uncertainty introduced through application of effect estimates to
study areas (i.e., CBSAs) that do not match those used in the underlying epidemiological studies.
11 The seasonal average metrics used in the long-term mortality estimate are not very sensitive to the reduced
number of days with co-pollutant monitoring, and as such it is appropriate to use the co-pollutant model in
generating the core risk estimates.
7-6
-------
This uncertainty should be viewed within the context of the overall larger uncertainty associated
with transferring effect estimates from the context of the epidemiological studies to the context
of the risk assessment. The epidemiological studies used in modeling short-term exposure-related
endpoints generate effect estimates based on day to day variation in Os and health effects, using
the area wide average Os concentrations. Area wide Os averaging masks the specific population
distribution of Os exposures which reflects the times and durations of exposures to Os measured
at individual monitors in an urban area. We apply those effect estimates to the air quality
scenarios of just meeting existing and alternative standards, where we are shifting the entire
distribution of daily Os concentrations, and altering the relationships between Os concentrations
at different monitors, and thus likely altering the relationship between area wide average Os and
the population distribution of Os exposures. By doing so, we introduce an additional source of
exposure measurement error, which goes beyond the impact that measurement error has on the
effect estimate, and introduces additional uncertainty into the estimates of risk associated with
simulating meeting existing and alternative standards.
Our decision to use the CBSA to define the spatial extent of each urban study area
reflects the greater weight we place on minimizing biases relative to minimizing uncertainty,
although we strive to minimize both where possible. The sensitivity analysis related to using
epidemiological study-based spatial definitions for urban areas shows clearly that using the
smaller urban areas biases downward the risk reductions across an urban area. Thus, to avoid this
bias in risk estimates we accept a measure of increased uncertainty associated with the
application of effect estimates to study areas that are larger than those used in some of the
original epidemiological studies providing those effect estimates.
Using the CBSA definitions of urban areas can partially address the bias caused by
focusing only on urban core areas. However, it does not address this bias fully in some areas
because of the unevenness in monitoring throughout urban areas. In some urban areas the
monitors are more evenly distributed across the CBSA, while in other areas they are not. For
example, in some urban areas, there is a high density of monitors in the urban core counties, with
less density of monitors in surrounding counties also in the CBSA. Because we use a simple
average (to match the averaging used in the epidemiology studies) of monitors across the CBSA,
this means that Os concentrations in areas where there are more monitors (e.g. in urban core
counties) will get a higher weight in the average Os concentrations relative to Cb concentrations
in other parts of the CBSA. To the extent that the area with the higher density of monitors
experiences increases in Os while the remaining area experiences decreases in Os, the overall
average Os concentrations applied to populations in the entire CBSA will be weighted more
towards Os increases, which will attenuate the overall risk reduction that may be associated with
meeting alternative Os standards. We are not able to determine the magnitude of this remaining
7-7
-------
bias; however, it is expected to be higher in locations with a high percentage of total CBSA
monitors concentrated in urban core counties.
The risk assessment reflects consideration for five years of recent air quality data from
2006 through 2010, with these five years reflecting two three-year averaging periods that share a
common overlapping year (i.e., 2006-2008 and 2008-2010 - see section 7.2). We selected these
two averaging periods to provide coverage for a more recent time period with relatively elevated
Os levels (2006-2008) and recent time period with relatively lower Os levels (2008-2010). For
the HREA, we model risk for the middle year of each three-year averaging period in order to
provide estimates of risk for a year with generally higher Os levels (i.e., 2007) and a year with
generally lower Os levels (i.e., 2009). In modeling risk, we matched the population data used in
the risk assessment to the year of the air quality data. For example, when we used 2007 air
quality data, we used 2007 population estimates. For baseline incidence and prevalence, rather
than interpolating rates for the two specific years modeled in the risk assessment, we selected the
closest year for which we had existing incidence/prevalence data (i.e., for simulation year 2007,
we used available data for 2005 and for simulation year 2009, we used data from 2010). The
calculation of baseline incidence and prevalence rates is described in section 7.3.4.
The risk assessment procedures described in more detail below are diagramed in Figure
7-1. To estimate the change in incidence of a given health effect resulting from a given change in
ambient Os concentrations in an assessment location, the following analysis inputs are necessary:
• Air quality information including: (1) O3 air quality data from each of the simulation
years included in the analysis (2007 and 2009) from population-oriented monitors in
the assessment location (these are aggregated to form composite monitor values used
to represent population exposure), and (2) a method for adjusting the air quality data
to simulate just meeting the existing or alternative suite of O3 standards. (These air
quality inputs are discussed in more detail in Chapter 4).
• C-R function(s): which provide an estimate of the relationship between the health
endpoint of interest and Os concentrations (for this analysis, C-R functions used were
applied to urban study areas matching the assessment locations from the
epidemiological studies used in deriving the functions, in order to increase overall
confidence in the risk estimates generated - see section 7.3.2). For Os,
epidemiological studies providing information necessary to specify C-R functions are
readily available for Os-related health effects associated with short-term exposures
(Section 7.1.2 describes the role of C-R functions in estimating health risks associated
with Cb). In addition, the Jerrett et al. (2009) study provided a C-R function for
modeling mortality risks associated with longer-term exposures to Os.
7-8
-------
• Population information (baseline health affects incidence and prevalence rates
and population): The baseline incidence provides an estimate of the incidence rate
(number of cases of the health effect per year or day, depending on endpoint, usually
per 10,000 or 100,000 general population) in the assessment location corresponding
to recent ambient Os levels in that location. The baseline prevalence rate describes the
prevalence of a given disease state or conditions (e.g., asthma) within the population
(number of individuals with the disease state/condition, usually per 10,000 or 100,000
general population). To derive the total baseline incidence or prevalence per year, this
rate must be multiplied by the corresponding population number (e.g., if the baseline
incidence rate is number of cases per year per 100,000 population, it must be
multiplied by the number of 100,000s in the population) (Section 7.3.3 summarizes
considerations related to the baseline incidence and prevalence rates and population
data inputs to the risk assessment).
In addition to the inputs described above, it is also necessary to specify the spatial extent
of the study areas that will be modeled. These study areas definitions determine the composition
of (a) the composite monitor values (which specific set of monitors are used in constructing the
composite monitor, reflecting the area-wide average across monitors for each study area), (b) the
specific set of effect estimates that will be used (matching the study areas to the specific set of
effect estimates in the epidemiological studies being used to support modeling of endpoints), (c)
the baseline incidence data and (d) the population demographic (count) data for each study area.
As mentioned earlier, for this HREA we have modeled 12 urban study areas and have used the
CBS A spatial definition to specify the extent of each of these urban areas (see section 7.3.1 for
additional details on study area selection).
7-9
-------
Air Quality Inputs (Chapter4)
Composite Monitor
03Melrics for recent
conditions in Urban Case
Study Areas
Composite Monitor
03 Metrics in Urban Case
Study Areas after just meeting
Existing and alternative standards
Concentration-Response Functions
Daily location-specific
baseline health incidence
BenMAP
Calculate daily changes in ozone
between just meeting current
standards and alternative standards
Compute day by day
ozone attributable
incidence of mortality
and morbidity
Compute day by day
changes in incidence of
mortality and morbidity
attributable to meeting
alternative standards
Estimates of * attributable
Incidence and change
in % attributable incidence
of mortality and morbidity
for the modeling period
Estimates of ozone-
attnbulable incidence of
mortality and morbidity
and change in ozone
attributable incidence
for the modeling period
Figure 7-1. Flow Diagram of Risk Assessment for Short-term Exposure Studies.
7-10
-------
This risk assessment was implemented using the EPA's Environmental Benefits Mapping
and Analysis Program (BenMAP, version 4.0) (U.S. EPA, 2013b). This GIS-based computer
program draws upon a database of population, baseline incidence/prevalence rates and effect
coefficients to automate the calculation of health impacts. For this analysis, the standard set of
effect coefficients and health effect incidence data available in BenMAP has been augmented to
reflect the latest studies and data available for modeling Os risk. EPA has traditionally relied
upon the BenMAP program to estimate the health impacts avoided and economic benefits
associated with adopting new air quality rules. For this analysis, EPA used the model to estimate
Os-related risk for the suite of health effects endpoints described in section 3.2. There are three
primary advantages to using BenMAP for this analysis, as compared to the procedure for
estimating population risk followed in the last review. First, once we have configured the
BenMAP software for this particular Ch analysis, the program can produce risk estimates for an
array of modeling scenarios across a large number of urban areas. Second, the program can more
easily accommodate a variety of sensitivity analyses. Third, BenMAP allowed us to complete the
national assessment of Os mortality described in Chapter 8, which plays in important role in
assessing the representativeness of the urban study area analysis.
7.1.2 Calculating Ozone-Related Health Effects Incidence
The C-R functions used in the risk assessment are empirically estimated associations
between average ambient concentrations of Os and the health endpoints of interest (e.g.,
mortality, hospital admissions (HA), emergency department (ED) visits). This section describes
the basic method used to estimate changes in the incidence of a health endpoint associated with
changes in Os, using a "generic" C-R function of the most common functional form.
Although some epidemiological studies have estimated linear C-R functions and some
have estimated logistic functions, most of the studies used a method referred to as "Poisson
regression" to estimate exponential (or log-linear) C-R functions in which the natural logarithm
of the health endpoint is a linear function of Os:
(1)
where x is the ambient Os level, y is the incidence of the health endpoint of interest at Os
level x, p is the coefficient relating ambient Os concentration to the health endpoint, and B is the
incidence at x=0, i.e., when there is no ambient Cb. The relationship between a specified ambient
Os level, xo, for example, and the incidence of a given health endpoint associated with that level
(denoted as yo) is then
(2)
7-11
-------
Because the log-linear form of a C-R function (equation (1) is by far the most common
form, we use this form to illustrate the "health impact function" used in the Os risk assessment.
If we let xo denote the baseline (upper) Os level, and xi denote the lower Os level, and yo
and yi denote the corresponding incidences of the health effect, we can derive the following
relationship between the change in x, Ax= (xo- xi), and the corresponding change in y, Ay, from
equation (I).12
*y = (y0-yi) = y0V-e-/&x]- (3)
Alternatively, the difference in health effects incidence can be calculated indirectly using
relative risk. Relative risk (RR) is a measure commonly used by epidemiologists to characterize
the comparative health effects associated with a particular air quality comparison. The risk of
mortality at ambient Os level xo relative to the risk of mortality at ambient Os level xi, for
example, may be characterized by the ratio of the two mortality rates: the mortality rate among
individuals when the ambient Os level is xo and the mortality rate among (otherwise identical)
individuals when the ambient Os level is xi. This is the RR for mortality associated with the
difference between the two ambient Os levels, xo and xi. Given a C-R function of the form
shown in equation (1) and a particular difference in ambient Os levels, Ax, the RR associated
with that difference in ambient Os, denoted as RRAx, is equal to epAx. The difference in health
effects incidence, Ay, corresponding to a given difference in ambient Os levels, Ax, can then be
calculated based on this RRAx as:
(4)
Equations (3) and (4) are simply alternative ways of expressing the relationship between
a given difference in ambient Os levels, Ax > 0, and the corresponding difference in health
effects incidence, Ay.13 These health impact equations are the key equations that combine air
quality information, C-R function information, and baseline health effects incidence information
to estimate ambient Os health risk.
12 If Ax < 0 - i.e., if Ax = (xi- xo) - then the relationship between Ax and Ay can be shown to be
Ay = (yl - y0) = y0 [e pta - 1]. If Ax < 0, Ay will similarly be negative. However, the magnitude of Ay will be the
same whether Ax>OorAx<0- i.e., the absolute value of Ay does not depend on which equation is used.
13 When calculating total risk associated with a specific air quality scenario, Ax is the total O3 concentration
associated with a given study area (as noted earlier in section 7.1.1, we are not incorporating thresholds, such as
LMLs into this analysis).
7-12
-------
7.2 AIR QUALITY CONSIDERATIONS
Air quality data are discussed in detail in Chapter 4 of this report. Here we describe those
air quality considerations that are directly relevant to the estimation of health risks in the
epidemiology based portion of the risk assessment. As described in section 7.1.1, the risk
assessment uses composite (area-wide average) monitor values derived for each urban study area
as the basis for characterizing population exposure in modeling risk. The use of composite
monitors reflects consideration for the way ambient Os data are used in the epidemiological
studies providing the C-R functions (see section 7.1.1). For the short-term exposure related
health endpoints, the composite monitor values derived for this analysis include hourly time-
series for each study area (where the Cb value for each hour is the average of measurements
across the monitors in that study area reporting values for that hour). Once these composite
monitor hourly time series are constructed, we can then extract short-term peak Os metrics
needed to model specific health effects endpoints. For short-term Os-attributable endpoints,
reflecting consideration for available evidence in the published literature (see section 7.3.2), we
have focused the analysis on short-term peak Os metrics including 1-hr maximum, 8-hr mean
and 8-hr maximum. The 24-hr average has been deemphasized for this analysis, although it is
still used in risk modeling when use of C-R functions based on this metric allow us to cover a
specific health effect endpoint/location of particular interest (see section 7.3.2).14
For modeling mortality risk associated with long-term Os-attributable we construct
seasonally-averaged maximum hourly Os values (see section 7.3.2). The derivation of composite
monitor distributions used in modeling this health effect endpoint is different than that used for
short-term Os-attributable endpoints. Specifically, for the long-term Os-attributable endpoint we
first construct the seasonally-averaged peak Os metric for each monitor within a given study area
and then average those monitor-specific metric values together to generate a single composite
value to use in generating risk estimates for that study area.
In applying effect estimates obtained from epidemiological studies we attempted to
match the modeling period (e.g. Os monitoring season) associated with each epidemiology study.
This increases overall confidence in the risk compared with using a single more generalized
specification of the modeling period. As discussed earlier, we modeled all health effect endpoints
for the core analysis using a CBSA-based study area. The use of the CBSA-based study areas
addresses potential bias that would have occurred had we focused the risk assessment on the
smaller core urban study areas (see section 7.1.1). Table 7-1 identifies (a) the counties associated
with the CBS A definition for each of the 12 urban study areas, (b) the number of Os monitors
14 In order to provide estimates of respiratory-related hospital admission for Los Angeles, we did include a C-R
function based on Linn et al., 2000, which utilizes a 24-hr average exposure metric.
7-13
-------
associated with each CBSA (and a flag for whether the design value monitor is contained within
the CBSA), (c) the number of monitors associated with the smaller Smith et al., 2009-based
study areas, and (d) the specific Os modeling period for each study area. A map showing the
counties and monitors for these 12 urban areas can be found in Chapter 4 (Figure 4-5, Section
4.3.2.2).
Table 7-1. Information on the 12 Urban Study Areas in the Risk Assessment.
Study Area
Atlanta
Baltimore
Boston
Cleveland
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Counties Associated with the CBSA
Definition
Barrow, Bartow, Butts, Carroll, Cherokee
Clayton, Cobb, Coweta, Dawson, DeKalb,
Douglas, Fayette, Forsyth, Fulton,
Gwinnett, Haralson, Heard, Henry,
Jasper, Lamar, Meriwether, Newton,
Paulding, Pickens, Pike, Rockdale,
Spalding, Walton
Anne Arundel, Baltimore, Carroll, Harford,
Howard, Queen Anne's, Baltimore
Essex, Middlesex, Norfolk, Plymouth,
Suffolk, Rockingham, Strafford
Cuyahoga, Geauga, Lake, Lorain, Medina
Adams, Arapahoe, Broomfield, Clear
Creek, Denver, Douglas, Elbert, Gilpin,
Jefferson, Park
Lapeer, Livingston, Macomb, Oakland,
St. Clair, Wayne
Austin, Brazoria, Chambers, Fort Bend,
Galveston, Harris, Liberty, Montgomery,
San Jacinto, Waller
Los Angeles, Orange
Bergen, Essex, Hudson, Hunterdon,
Middlesex, Monmouth, Morris, Ocean,
Passaic, Somerset, Sussex, Union,
Bronx, Kings, Nassau, New York,
Putnam, Queens, Richmond, Rockland,
Suffolk, Westchester, Pike
New Castle, Cecil, Burlington, Camden,
Gloucester, Salem, Bucks, Chester,
Delaware, Montgomery, Philadelphia
El Dorado, Placer, Sacramento, Yolo
Bond, Calhoun, Clinton, Jersey,
Macoupin, Madison, Monroe, St. Clair,
Franklin, Jefferson, Lincoln, St. Charles,
St. Louis, Warren, Washington, St. Louis
#of03
Monitors
within the
CBSA3
13(3)
7(1)
11* (2)
10* (4)
16(6)
8(4)
22(17)
21* (17)
22(7)
15(4)
17(8)
17(2)
Required Os Monitoring Season
March - October
April - October
April - September
April - October
March - September
April - September
January - December
January- December
April - October
April - October
January - December
April - October
a column presents the number of monitors within each CBSA, whether the design value
(denoted with an "*") and the number of monitors within the smaller Smith et al., 2009-
parenthesis).
falls outside of the CBSA
•based study area (in
7-14
-------
We estimate risk associated with recent Os conditions as well as risk associated with
simulating just meeting the existing and alternative standards. While the derivation of composite
monitor hourly Os distributions (and associated peak exposure metrics) for recent conditions is
relatively straightforward, the generation of these estimates for the scenarios of just meeting the
existing and alternative standards is more complex. The procedures for adjusting air quality to
just meet the existing and alternative Os standards are presented in Chapter 4 and Chapter 4
appendices.
Summary statistics for the air metrics used in modeling risk for each of the 12 urban
study areas under recent conditions and for air quality adjusted to just meets the existing and
alternative standard levels are presented in Chapter 4 (see section 4.3.3.2, Figures 4-9 (2007) and
4-10 (2009)).
7.3 SELECTION OF MODEL INPUTS AND ASSUMPTIONS
7.3.1 Selection of Urban Study Areas
This analysis focuses on modeling risk for a set of urban study areas, reflecting the goal
of providing risk estimates that have greater overall confidence due to the use of location-
specific data when available for these urban locations. In addition, given the greater availability
of location-specific data, a more rigorous evaluation of the impact of uncertainty and variability
can be conducted for a set of selected urban study areas than would be possible for a broader
regional or national-scale analysis. We considered the following factors in selecting the 12 urban
study areas included in this analysis:
• Air Quality Data: An urban area has reasonably comprehensive monitoring data for
the period of interest (2006-2010) to support the risk assessment. This criterion was
evaluated qualitatively by considering the number of monitors within the CBS A of
the prospective urban areas. Locations with one or two monitors would be excluded
since they had relatively limited spatial coverage in characterizing Os levels.
• Elevated Ambient Os Levels: Because we are interested in evaluating the potential
magnitude of risk reductions associated with just meeting the existing and alternative
Os standard levels, we focus on study areas with elevated ambient Os levels at or
above the existing standard, such that just meeting alternative Os standard levels
would result in some degree of risk reduction.
• Location-specific C-R Functions: Given the health endpoints selected for inclusion
in the analysis (see section 7.3.2), there are epidemiological studies of sufficient
quality available for these urban study areas to provide the C-R functions necessary
for modeling risk. This criterion primarily applies to short-term epidemiological
studies since the associated health effect endpoints are the primary focus of the
7-15
-------
HREA. Short-term Os-attributable epidemiological studies often include city-specific
effect estimates, and in some cases are multi-city studies that provide estimates for
multiple cities.
• Baseline Incidence Rates and Demographic Data: The required urban area-specific
baseline incidence rates and population data are available for a recent year for at least
one of the health endpoints.
• Geographic Heterogeneity: Because Os distributions and population characteristics
vary geographically across the U.S., we selected urban study areas to provide
coverage for regional variability in factors related to Os risk including variability in
the spatial pattern of Os in the urban area, population exposure (differences in
residential housing density, air conditioning use and commuting patterns),
demographic characteristics (baseline incidence rates, socio-economic status) and
variability in effect estimates. The degree to which the set of urban study areas
provided coverage for regional differences across the U.S. in many of these Os risk-
related factors was evaluated as part of the representativeness analysis presented in
Chapter 8.
Application of the above criteria resulted in the selection of 12 urban study areas for
inclusion in the risk assessment including:
• Atlanta, GA
• Baltimore, MD
• Boston, MA
• Cleveland, OH
• Denver, CO
• Detroit, MI
• Houston, TX
• Los Angeles, CA
• New York, NY
• Philadelphia, PA
• Sacramento, CA
• St. Louis, MO
The specific set of counties used in defining each of the 12 urban study areas based on
the CBSA is presented in Table 7-1.
7-16
-------
7.3.2 Selection of Epidemiological Studies and Specification of Concentration-Response
Functions
Once the set of health effect endpoints to be included in the risk assessment has been
specified, the next step was to select the set of epidemiological studies that will provide the
effect estimates and model specifications used in the C-R functions. This section describes the
approach used in completing these tasks and presents a summary of the epidemiological studies
and associated C-R functions specified for use in the risk assessment.
In Chapter 2, section 2.5 we identified the set of health effect categories and associated
endpoints to be included in this assessment, based on review of the evidence provided in the Os
ISA (U.S. EPA, 2013a). The selection of specific health effect endpoints to model within a given
health effect endpoint category is an iterative process involving review of both the strength of
evidence (for a given endpoint) as summarized in the Os ISA together with consideration for the
available epidemiological studies supporting a given endpoint and the ability to specific key
inputs needed for risk modeling, including effect estimates and model forms. Ultimately,
endpoints are only selected if (a) they are associated with an overarching effect endpoint
category selected for inclusion in the risk assessment and (b) they have sufficient
epidemiological study support to allow their modeling in the risk assessment. Health effect
endpoints selected for inclusion in HREA include:
Short-term Os-attributable effects:
• Mortality (likely to be a causal relationship)
o All-cause (non-accidental)
o Cardiovascular
o Respiratory
• Respiratory effects (causal relationship)
o ED (for asthma, wheeze, all respiratory symptoms)
o HA (for COPD, asthma, all respiratory}15
o Respiratory symptoms
15 Regarding chronic obstructive pulmonary disease (COPD)-related HA, the Os ISA states that "Although limited in
number, both single- and multi-city studies consistently found positive associations between short-term Os
exposures and asthma and COPD hospital admissions." (U.S. EPA, 2013a, p. 6-128). It is also important to point
out that when modeling of COPD-related HA is limited to the summer months (as was done for the HREA),
available effect estimates have tighter confidence intervals and are generally positive, which increases overall
confidence in the resulting risk estimate (see U.S. EPA, 2013a, Figure 6-19).
7-17
-------
Long-term Os-attributable effects:
• Respiratory effects, focusing on respiratory-related mortality (likely causal
relationship).16
We selected epidemiological studies to support modeling of the health effect endpoints
listed above by applying a number of criteria including:17
• The study was peer-reviewed, evaluated in the Os ISA, and judged adequate by EPA
staff for purposes of inclusion in the risk assessment. We considered the following
criteria: whether the study provides C-R relationships for locations in the U.S.,
whether the study has sufficient sample size to provide effect estimates with a
sufficient degree of precision and power, and whether adequate information is
provided to characterize statistical uncertainty.
• Preference for multicity studies given that they typically have greater power and
reflect patterns of Os related health effects over a range of urban areas (and regions)
which can display variability in key risk-related factors such as exposure
measurement error. In the case of short-term Os-attributable mortality, we also
favored those multi-city studies for which we could obtain Bayesian-adjusted city-
specific estimates from the study authors, since these incorporate both city-specific
effect information with information from the broader array of cities included in the
study. In those instances where we did not have multi-city studies (e.g., with many of
the short-term respiratory-related morbidity endpoints) we use single-city studies.
• The study design is considered robust and scientifically defensible, particularly in
relation to methods for covariate adjustment, including treatment of confounders, as
well as treatment of effect modifiers. For example, if a given study used ecological-
defined variables (e.g., smoking rates) as the basis for controlling for confounding,
concerns may be raised as to the effectiveness of that control.
16 The O3 ISA classifies long-term Os-attributable respiratory health effects, including respiratory-
related mortality, as having a likely causal classification. By contrast, it classifies long-term O3-
attributable total mortality as having a suggestive of a causal relationship classification (O3
ISA, Chapter 1). We have focused on modeling long-term Os-attributable respiratory-related
mortality given the greater support for this health endpoint relative to total mortality.
17 In addition to the criteria listed here, we also attempted to include studies that provide coverage for populations
considered particularly at-risk for a particular health (e.g., children, individuals with preexisting disease).
However, a study would have to meet the criteria listed here (in addition to providing coverage for an at-risk
population) in order for that study to be used to derive C-R functions.
7-18
-------
• The study is not superseded by another study (e.g., if a later study is an extension or
replication of a former study, the later study would effectively replace the former
study), unless the earlier study has characteristics that are clearly preferable (e.g.,
inclusion of co-pollutants models, or use of a peak exposure metric of interest).
We applied the above criteria and selected the set of epidemiological studies presented in
Table 7-2 for use in specifying C-R functions (Table 7-2 also describes elements of the C-R
functions specified using each epidemiological study, as discussed below).
As part of methods refinement for this risk assessment, we considered studies that
utilized more sophisticated and potentially representative exposure surrogates in characterizing
population exposure (e.g., using population-weighted Os monitor values instead of equally-
weighted monitors, linking exposures in individual counties or U.S. Census tracts to the nearest
monitor, rather than using a composite monitor value to represent the entire study area).
However, analysis conducted by EPA demonstrated that use of the simpler composite monitor
approach (as used for other short-term Os-attributable morbidity endpoints) generated risk
estimates that were very close to those generated using the population-weighted Os metric.
Therefore, in order to conserve time and resources, we modeled this endpoint using the more
generalized composite monitor-based metric. And finally, a number of the long-term Os-
attributable morbidity studies originally considered for modeling this endpoint category did
involve more complex Cb metrics (e.g., Akinbami et al., 2010; Meng et al., 2010; Moore et al.,
2008). However, limitations in the study-level data required to support risk assessment prevents
us at this point from completing a quantitative risk assessment for this category of health
endpoints with a reasonable degree of confidence.18
Based on additional evaluation of the literature, we have substituted Smith et al. (2009)
for Bell et al. (2004) as a source of Bayes-adjusted city-specific effect estimates to support
modeling short-term Os-attributable mortality. This decision reflects a number of factors. Smith
et al. (2009) includes a wider range of simulations exploring sensitivity of the mortality effect to
different model specifications including (a) regional versus national Bayes-based adjustment, (b)
co-pollutants models considering PMio, and (c) all - year versus Os-season based estimates. This
is contrasted with the Bell et al. (2004) study which does not provide this degree of model
exploration. In obtaining the city-specific Bayes-adjusted effect estimates for the Smith et al.,
18 However, these limitations do not prevent the use of this evidence from informing consideration of the levels of
exposure at which specific types of health effects may occur (i.e., the evidence analysis, which is an important
aspect of the O3 NAAQS review). Rather, these limitations only prevent the quantitative estimation of risk with a
reasonable degree of confidence.
7-19
-------
2009 study from the study authors, we were provided with estimates reflecting this range of
alternative model specifications which allowed us to incorporate them into both the core and
sensitivity analysis portions of the HREA (see section 7.4.3). In addition, the Smith et al. (2009)
study does not use the trimmed mean approach employed in the Bell et al. (2004) study in
preparing Os monitor data. We have a number of concerns regarding the trimmed mean approach
including (1) the potential loss of temporal variation in the data when the approach is used (this
could impact the size of the effect estimate) and (2) a lack of complete documentation for the
approach which prevents us from fully reviewing the technique and using it in preparing Os
metrics for the HREA. Given these concerns, we view it as advantageous that the Smith et al.
(2009) study does not use the trimmed mean approach.
With the exception of the trimmed mean approach, the Smith et al. (2009) study was
intended to reproduce the results of the Bell et al. (2004) analysis. Thus, the core risk results
based on Smith et al. (2009) are functionally comparable to the 1st draft HREA estimates based
on Bell et al. (2004), while the alternative models provided in Smith et al. (2009) allow for an
expanded set of sensitivity analyses. The comparability of the Smith et al. (2009) and Bell et al.
(2004) estimates is confirmed by the graphical comparison in Smith et al. (2009) of mortality
effect estimates (for the 24-hr Os metric) with matching effect estimates from Bell et al. (2004).
This comparison demonstrates the close match of the two studies (for this particular scenario).
Reflecting the points made above, in modeling short-term Os-attributable mortality, we
have included a core analysis based on the national-Bayesian adjusted city-specific effect
estimates (reflecting the full Os monitoring period in each city) obtained from Smith et al.
(2009). As sensitivity analyses, we have included effect estimates obtained from Smith et al.
(2009) which reflect application of co-pollutants models (including PMio), Bayes adjustment
using a regional prior,19 and a shorter fixed Os measurement period (April-October). In the 1st
draft HREA, we had also included national Bayes-adjusted effect estimates (reflecting a fixed
June-August period) obtained from Zanobetti and Schwartz (2008) as part of the core analysis.
However, here we decided to include these as part of the sensitivity analysis here because these
effect estimates cover a more limited warm-weather period and consequently will generate only
partial characterizations of mortality risk (since they exclude risk occurring during the non-
summer months).
We have also included estimates of respiratory-related mortality associated with long-
term Os exposures (summer season mean of daily maximum 1-hr Os) based on effect estimates
obtained from Jerrett et al. (2009). The decision to model long-term Os-attributable mortality
19 With application of a regional prior within Bayesian adjustment, city-specific effect estimates are adjusted
towards the regional value rather than a national value as is the case with the application of a national prior.
7-20
-------
reflects consideration for evidence supporting a likely to be a causal relationship for long-term
Os-attributable respiratory effects, including mortality (Os ISA, section 2.5.2). The Jerrett et al.
(2009) study was the first study to explore the relationship between long-term Os exposure and
respiratory mortality (rather than focusing on cardiopulmonary mortality). Jerrett et al. (2009)
does have a number of strengths including (a) the study is based on the 1.2 million participant
American Cancer Society cohort drawn from all 50 states, DC, and Puerto Rico (included Os
data from 1977 [5 years before enrollment in the cohort began] to 2000); (b) inclusion of co-
pollutant models that controlled for PIVfo.s; and (c) exploration of the potential for a threshold
concentration associated with the long-term mortality endpoint. However we also recognize a
few factors which impact overall interpretation of the long-term exposure-related respiratory
mortality estimates based on this study. First, while the Jerrett et al., (2009) study is a relatively
strong study in terms of overall design, it is a single study for this particular health endpoint
which means that the underlying support for the quantitative risk estimation for this endpoint is
limited in comparison with other endpoints modeled (e.g., short-term exposure-related mortality
risk that uses numerous studies). In addition, there is uncertainty about the existence and location
of a threshold in the C-R function relating mortality and long-term Os concentrations, and that
uncertainty may have a large impact on our quantitative risk estimates.
The exploration of potential thresholds presented in the Jerrett et al., (2009) study
deserves additional discussion.20 In their memo clarifying the results of their study (see Sasser,
2014), the authors note that in terms of goodness of fit, long-term health risk models including
Os clearly performed better than models without Os, indicating the improved predictions of
respiratory mortality when Os is included. In exploring different functional forms, they report
that the model including a threshold at 56 ppb had the lowest log-likelihood value of all models
evaluated (i.e., linear models and models including thresholds ranging from 40-60 ppb), and thus
provided the best overall statistical fit to the data. However, they also note that it is not clear
whether the 56 ppb threshold model is a better predictor of respiratory mortality than when using
a linear (no-threshold) model for this dataset. Using one statistical test, the model with a
threshold at 56 ppb was determined to be statistically superior to the linear model. Using another,
more stringent test, none of the threshold models considered were statistically superior to the
linear model. Under the less stringent test, although the threshold model produces a statistically
superior prediction than the linear model, there is uncertainty about the specific location of the
20 The approach we developed to explore the potential for thresholds related to long-term exposure-related mortality
was presented in a memorandum to CASAC which was also released to the public (Sasser, 2014). That
memorandum describes additional data obtained from the authors of Jerrett et al. (2009) to support modeling of
potential thresholds and also lays out our proposed approach for exploring the impact of potential thresholds on
estimates of long-term exposure-related mortality.
7-21
-------
threshold, if one exists. This is because the confidence intervals on the model predictions
indicate that a threshold could exist anywhere from 0 to 60 ppb. The authors conclude that
considerable caution should be exercised in using any specific threshold, particularly when the
more stringent statistical test indicates there is no significantly improved prediction. Based on
this additional information from the authors (Sasser, 2014), we have chosen to reflect the
uncertainty about the existence and location of a potential threshold by estimating mortality
attributable to long-term Os exposures using a range of threshold-based effect estimates as
sensitivity analyses. Specifically, we generate additional long-term risk results using unique risk
models that include a range of thresholds from 40 ppb to 60 ppb in 5 ppb increments, while also
including a model with a threshold equal to 56 ppb, which had the lowest log likelihood value for
all models examined.21
Other limitations associated with the Jerrett et al. (2009) study include possible exposure
misclassification and uncontrolled confounding by temperature, which are endemic to most long-
term epidemiological studies. It is also important to note that, while Jerrett et al. (2009) found
negative associations between Os exposure and cardiovascular mortality when controlling for
PM2.5, null or negative associations for Os are consistent with the evidence that PM2.5 is the
pollutant most strongly associated with cardiovascular disease (U.S. EPA, 2009).
We based the core estimate for long-term exposure-related mortality on a co-pollutant
model (with PIVh.s) obtained from Jerrett et al. (2009). This reflects the observation that the
seasonal average metrics used in the long-term exposure-related mortality are not very sensitive
to the reduced number of days with co-pollutant monitoring, and as such it is appropriate to
include the co-pollutant model as the core estimate. We also include three sensitivity analyses for
long-term Os-attributable respiratory mortality including: (a) application of regionally-
differentiated effect estimates (although these do not include a co-pollutants model
specification), (b) application of a single pollutant (Os-only) national-based effect estimate and
(c) application of the suite of effect estimates reflecting potential thresholds at 40, 45, 50, 55, 56
and 60 ppb (effect estimates for both the core and sensitivity analyses are presented in Appendix
7A).
The effect estimates used in modeling long-term Os-attributable mortality (see Table 7-2)
utilize a seasonal average of peak (1-hr maximum) measurements. These long-term exposure
metrics can be viewed as long-term exposures to daily peak Os over the warmer months, as
compared with annual average levels such as are used in long-term PM exposure calculations.
21 There is a separate effect estimate (and associated standard error) for each of the fitted threshold models estimated
in Jerrett et al. (2009). As a result, the sensitivity of estimated mortality attributable to long-term O3
concentrations is affected by both the assumed threshold level (below which there is no effect of O3) and the effect
estimate applied to Os concentrations above the threshold.
7-22
-------
This increases the need for care in interpreting these long-term Os-attributable mortality
estimates together with the short-term Os-attributable mortality estimates, in order to avoid
double counting. It is also important to keep in mind that our estimates of short-term Cb-
attributable mortality are for all-causes, while estimates of long-term Os-attributable mortality
are focused on respiratory-related mortality. This further limits the ability to compare estimates
of long-term and short-term exposure related mortality.
Once the set of epidemiology studies described above was selected, the next step was to
specify C-R functions for use in the risk assessment. Several factors were considered in
identifying the effect estimates and model forms used in specifying C-R functions for each
endpoint. These factors are described below:
• Ozone Exposure Metric: In the risk assessment performed for the previous Os
NAAQS review, for short-term exposure, we included C-R functions based on 24-hr
averages as well as a number of peak Os measurements. However, given that the
existing Os NAAQS standard uses an 8-hr average form and given that many of the
clinical studies involving Os also utilize shorter exposures (on the order of 2 to 8
hours - see Os ISA, section 6.2.1.1), we wanted to see if the latest epidemiological
studies for Os also supported use of an 8-hr averaging time in modeling risk. Several
epidemiological studies completed since the last review provide limited support for
stronger associations between health endpoints and peak Os metrics (i.e., 1-hr
maximum, 8-hr maximum and 8-hr means) relative to 24-hr averages. Specifically, a
study of respiratory ED visits in Atlanta (Darrow et al., 2011) found stronger
associations with peak metrics (including 1-hr and 8-hr maximum measurements)
compared with 24-hr averages (see Os ISA, section 6.2.7.3 and Figure 6-17, U.S.
EPA, 2013a). Similarly, for short-term exposure-related mortality, there are also a
limited number of epidemiologic studies that have compared mortality associations
with peak Os metrics and the 24-hr average metric. Although the Os ISA recognizes
that 24-hr exposure metrics when used in time series studies may result in smaller risk
estimates, ultimately it concludes that "Overall, the evidence from time-series and
panel epidemiologic studies does not indicate that one exposure metric is more
consistently or strongly associated with mortality or respiratory-related health effects"
(U.S. EPA, 2013a, section 2.5.4.2). Based on consideration for the evidence
summarized in the Os ISA, we have decided to focus on peak exposure metrics
because of the limited evidence that these metrics may be associated with higher risk
estimates relative to the 24-hr exposure metric. However, we recognize that, as
summarized in the Os ISA, there is only weak support for differentiating between
these two categories of short-term exposure metric.
7-23
-------
Table 7-2. Overview of Epidemiological Studies Used in Specifying C-R Functions.
Epidemiological
Study
(stratified by Oy
attributable
health
end points)
Health
Endpoints
Location
(urban
study
area(s)
covered)
Exposure Metric
(and modeling
period)
Additional Study Design Details
Notes Regarding Application in the Analysis
Short-term O3-attributable mortality
Smith et al., 2009
Zanobetti and
Schwartz (2008)
Non-accidental,
respiratory,
cardiovascular
Non-accidental,
respiratory,
cardiovascular
95 large
urban
communities
(provides
coverage for
all 12 urban
study areas)
48 U.S. cities
(provides
coverage for
the 12 urban
study areas)
24-hr avg, 8-hr
max, 1-hr max.
April through
October and all
year
8-hr max. June-
August
Adjusting for time-varying
confounders (PM, weather,
seasonality). Lag structure included
0, 1 , 2 and day 3 lag as well as 0-6
day distributed lag. Age range: all
ages.
Effect controlled for season, day of
week, and temperature. Lag structure
included 0-3 day, 0-20 and 4-20 day).
Age range: all ages
Focused on the 8-hr max-based metric C-R functions
for the HREA (see text discussion later in this section).
Obtained Bayes-adjusted city-specific effect estimates
for non-accidental all-cause mortality from Dr. Smith
(personal communication, Dr. Richard L. Smith,
January 15, 2013) reflecting consideration for the
following modeling elements: (a) regional- versus
national-prior Bayes model adjustment, (b) single
pollutant versus co-pollutants (PMio) models, and (c)
full monitoring period versus summer only (April-
October). For the core analysis, we focused on the
single pollutant (O3-only) model covering the full
monitoring period. The co-pollutants model (with PM10)
was included as a sensitivity analysis (see section
7.4.3).
Obtained Bayes-adjusted city-specific effect estimates
for non-accidental, respiratory and cardiovascular from
Dr. Zanobetti (personal communication, Dr. Antonella
Zanobetti, January 5, 2012). These effect estimates
reflect a 0-3 day distributed lag and are based on 8-hr
mean O3 levels measured between June and August.
Estimates were generated for each study area using
this constrained warm-season period.
Short-term O3-attributable morbidity - HA for respiratory effect)
Medina-Ramon et
al.,2006.
Linn et al., 2000
Linetal., 2008
HA: COPD,
pneumonia
HA:
unscheduled
for pulmonary
illness
HA: respiratory
disease
36 cities
(provides
coverage for
all 12 urban
study areas)
LA only
NY State
(used to
cover NYC)
8-hr mean, warm
(May-September),
cool (October-
April), all year
24-hr mean, LA O3
season (all year),
winter, spring,
summer and
autumn
1-hr max (for
10am-6pm
interval), warm
season (April-
October)
Distributed lag (0-1 day). Age range:
a 65yrs. Controlled for day of the
week and weather (including
temperature).
Lag 0. Age range: all ages. Used
subgroup analysis to explore the
effect of temporal variation, weather
and autocorrelation on O3 effect.
Lag 0, 1, 2, 3. Age range: <18yrs.
Models adjusted for the confounding
effects of demographic
characteristics, particulate matter
(PMio), meteorological conditions,
Generated risk estimates based on warm season for
COPD only (May-September).
Included effect estimate based on 24-hr avg metric (for
summer) since this provided additional coverage for HA
in L.A. Modeled using air quality for June-August.
Used 1-hr max metric applied to the city-specific O3
season for NY (April-October).
7-24
-------
Epidemiological
Study
(stratified by Oy
attributable
health
endpoints)
Katsouyanni et al.
(2009)
Silverman et al.
(2010)
Health
Endpoints
HA:
cardiovascular
disease,
chronic
obstructive
pulmonary
disease,
pneumonia, all
respiratory
HA: asthma
(ICU and non-
ICU)
Location
(urban
study
area(s)
covered)
1 4 cities
(provides
coverage for
Detroit only)
NYC
Exposure Metric
(and modeling
period)
1-hr max. Summer
only and all year
8-hr max. Warm
season (April-
August)
Additional Study Design Details
day of the week, seasonality, long-
term trends, and different
lag periods of exposure.
Lag 0-1 day. Age range: a 65yrs.
Models accounted for seasonal
patterns, but also, for weekend and
vacation effects, and for epidemics of
respiratory disease. The data were
also analyzed to detect potential
thresholds in the concentration-
response relationships.
Includes control for PM25. Lag 0-1
day. Age range: children 6-18yrs.
The model adjusted for temporal
trends, weather, and day of the
week.
Notes Regarding Application in the Analysis
C-R function applied only for all respiratory endpoint.
Used June-August-based composite monitor.
Applied C-R function (for O3 and O3 with control for
PM25) to the city-specific O3 season for NY (slightly
longer than the modeling period used in the study).
Short-term O3-attributable morbidity- ED and ER visits (respiratory)
Ito et al,(2007)
Tolbertetal.
(2007)
Strickland et al.
(2010)
Darrow et al.
(2011)
ED: asthma
ED: all
respiratory
ER: respiratory
ED: all
respiratory
NYC
Atlanta
Atlanta
Atlanta
8-hr max. Warm
season (April-
September)
8-hr max. Summer
(March-October)
8-hr max (based on
population
weighted average
across monitors).
Warm season (May
to October) and
cool (November to
April)
8-hr max, 1-hr max,
2-4hr avg for
summer (March-
October).
Includes models controlling for SO2,
NO2, CO and PM2.5. Lag: 0, 1 , and
distributed lag (0-1 day). Age range:
all ages. Model adjusts for temporal
trends, weather terms, day-of-week
and other pollutants.
Includes models controlling for NO2,
CO, PM10, and NO2/ NO2. Age range:
all ages. Model controls for temporal
trends, temperature, other pollutants.
Lag: average of 0-2 day, distributed
lag 0-7 day. Age range: 5-17yrs.
Model controls for seasonal trends
and meteorology.
Lag: 1day. Age range: all ages. The
study used a time series analysis
similar to case-crossover with
crossover matching based on daily
temperature (rather than day of the
week) to provide control for this key
risk-related factor.
Applied C-R functions (for O3 alone and O3 with control
for listed pollutants) to the city-specific O3 season for
NY (slightly longer than the modeling period used in the
study).
Applied C-R functions (for O3 alone and O3 with control
for listed pollutants) to the city-specific O3 season for
Atlanta.
Included effect estimates based on both lag structures
and used composite monitor values for city-specific O3
season.
Used city-specific O3 season-based composite monitor
values.
7-25
-------
Epidemiological
Study
(stratified by Oy
attributable
health
endpoints)
Health
Endpoints
Location
(urban
study
area(s)
covered)
Exposure Metric
(and modeling
period)
Additional Study Design Details
Notes Regarding Application in the Analysis
Short-term O3-attributable morbidity - respiratory symptoms
Gent et al. (2003)
Respiratory
symptoms:
wheeze,
persistent
cough, chest
tightness,
shortness of
breath
Long-term O3-attributable respirator]
Jerrett et al.
(2009)
Respiratory,
cardiovascular,
cardiopulmonar
y
Springfield,
MA (study
used to
cover
Boston)
1-hr max, 8-hr max
Lag: 0 and 1 day. Age range:
asthmatic children <12 yrs. Model
adjusted for temperature.
Included effect estimates for different symptoms based
on both 8-hr max and 1-hr max metrics (for city-specific
O3 season composite monitor values for Boston). The
study area (which focuses on Springfield and the
northern portion of Connecticut) does not encompass
Boston. However, we are willing to accept uncertainty
associated with using effect estimates from this study
to provide coverage for Boston given the goal of
providing coverage for this morbidity endpoint.
However, there is increased uncertainty associated
with modeling for this endpoint.
/ mortality
96
metropolitan
statistical
areas
(provides
coverage for
all 12 study
areas)
Seasonal average
(i.e., Apr-Sep) of
the peak daily
max1-hr values.
>30 yrs of age, includes national-
level and regional effect estimates
(only national-level estimate has co-
pollutants modeling considering
PM2.5along with O3). Modeling
included consideration for a range of
potential confounders evaluated at
both the ecological level and
personal level.
Included national co-pollutants model-based effect
estimates in core analysis and single-pollutant model
regional effect estimates and national effect estimates
as sensitivity analyses. Also included effect estimates
reflecting potential thresholds as sensitivity analyses.
7-26
-------
Single-and Multi-pollutant Models (pertains to both short-term and long-term
exposure studies): Epidemiological studies often consider health effects associated
with ambient Os using both single-pollutant and co-pollutant models. To the extent
that any of the co-pollutants present in the ambient air may have contributed to health
effects attributed to Os in single pollutant models, risks attributed to Os may be
overestimated or underestimated if C-R functions are based on single pollutant
models. This would argue for inclusion of models reflecting consideration of co-
pollutants. Conversely, in those instances where co-pollutants are highly correlated
with Os, inclusion of those pollutants in the health impact model can produce unstable
and statistically insignificant effect estimates for both Os and the co-pollutants.
Furthermore, there are often significant differences in sampling frequencies for each
pollutant included in co-pollutants models, which can lead to a loss of statistical
power in co-pollutants models (relative to single pollutant models). These last points
could argue for inclusion of a model based exclusively on Os. Given that single and
multi-pollutant models each have potential advantages and disadvantages, to the
extent possible, given available information we have included both types of C-R
functions in the risk assessment.
Multiple Effect Estimates within a Given CBSA-based Study Area: All health
endpoints, including short-term Cb-attributable mortality are modeled using CBSA-
based study areas. In the case of both Smith et al. (2009) and Zanobetti and Schwartz
(2008), these CBSA-based study areas are larger than the study areas used in these
epidemiological studies to derive effect estimates. Furthermore, for some of the
CBSA-based urban study areas, several of the smaller study areas evaluated in the
epidemiological study fall within a single larger CBSA-based study area. For
example, with the Smith et al. (2009) study, multiple effect estimates are available for
the CBSA-defined study areas of Los Angeles and New York. Specifically, the Smith
et al. (2009) study provides separate effect estimates for (a) Santa Anna/Anaheim and
Los Angeles study areas, both of which fall within the larger CBSA-based Los
Angeles study area and (b) New York, Jersey City and Newark study areas, all of
which fall within the larger CBSA-defined New York study area (see Table 7-3). This
raises the question of how to specify the effect estimate for these larger CBSA-based
study areas when there are multiple effect estimates available from the
epidemiological study. For this analysis, in those instances where there are multiple
effect estimates, we have decided to use the effect estimate that represents the largest
number of residents within each CBSA-based study area. There is uncertainty
7-27
-------
associated with this decision which is discussed both in section 7.4.2 and section
7.5.3 (as part of the air quality-related sensitivity analysis discussion).
Table 7-3. CBSA-based Study Areas with Multiple Effect Estimates from the Smith et al.
(2009) Study
CBSA
Study Area
New York
Los Angeles
Smith et al.,
2009 (smaller)
Study Areas
with CBSA-
based Study
Area"
New York, NY
Jersey City, NJ
Newark, NJ
Santa
Ana/Anaheim,
CA
Los Angeles, CA
Population
Totals
9,100,000
630,000
780,000
3,000,000
9,800,000
Mortality
Effect
Estimate
0.0009
0.0001
0.0005
0.0002
0.0001
Comments
New York study area dominates from
a population standpoint, so that effect
estimate was chosen to represent the
entire CBSA. An additional 8.3
million people live in portions of the
New York CBSA not covered by the
Smith et al., 2009 study areas.
Los Angeles dominates from a
population standpoint, so that effect
estimate was chosen to represent the
entire CBSA. In this case, the full
CBSA-based study area is covered by
the Smith et al., 2009-based subareas.
a Source: obtained from Dr. Smith (personal communication, Dr. Richard L. Smith, January 15, 2013)
• Single-city Versus Multi-city Studies: All else being equal, we judge C-R functions
estimated in the assessment location as preferable to a function estimated in some
other location, to avoid uncertainties that may exist due to differences associated with
geographic location. There are several advantages, however, to using estimates from
multi-city studies versus studies carried out in single cities. Multi-city studies are
applicable to a variety of settings, since they estimate a central tendency across
multiple locations. Multi-city studies also tend to have more statistical power and
provide effect estimates with relatively greater precision than single-city studies due
to larger sample sizes, reducing the uncertainty around the estimated health
coefficient. By contrast, single-city studies, while often having lower statistical power
and varying study designs which can make comparison across cities challenging,
reflect location-specific factors such as differences in underlying health status, and
differences in Os exposure-related factors such as air conditioner use and patterns of
urban residential density. There is a third type of study design that generates Bayes-
adjusted city-specific effect estimates, thereby combining the advantages of both city-
specific and multi-city studies. Bayes-adjusted city-specific estimates begin with a
city-specific effect estimate and shrink that towards a multi-city mean effect estimate
7-28
-------
based on consideration for the degree of variance in both estimates. We have elected
to place greater confidence on these types of Bayesian-adjusted effect estimates when
they are available. Otherwise, given the advantages for both city-specific and multi-
city effect estimates, we have used both types when available.
Multiple Lag Models: Based on our review of evidence provided in the Os ISA, we
believe there is increased confidence in modeling both short-term Os-attributable
mortality and respiratory morbidity risk based on exposures occurring up to a few
days prior to the health effect, with less support for associations over longer exposure
periods or effects lagged more than a few days from the exposure (see Os ISA, section
2.5.4.3). Consequently, we have favored C-R functions reflecting shorter lag periods
(e.g., 0, 1 or 1-2 days). With regard to the specific lag structure (e.g., single day
versus distributed lags), the Cb ISA notes that epidemiological studies involving
respiratory morbidity have suggested that both single day and multi-day average
exposures are associated with adverse health effects (see Os ISA, section 2.5.4.3).
Therefore, when available both types of lag structures where considered in specifying
C-R functions for short-term Os-attributable mortality and morbidity.
Seasonally-differentiated Effects Estimates: The previous Os Air Quality Criteria
Document (AQCD) (published in 2006) concluded that aggregate population time-
series studies demonstrate a positive and robust association between ambient Os
concentrations and respiratory-related hospitalizations and asthma ED visits during
the warm season (see Os ISA, section 2.5.3.1). The Os ISA notes that recent studies of
short-term Os-attributable respiratory mortality in the U.S. suggest that the effect is
strengthened in the summer season (Os ISA, section 2.5.3.1). In addition, many of the
key epidemiological studies discussed in the Os ISA exploring both short-term
exposure related mortality and morbidity have larger (and more statistically
significant) effect estimates when evaluated for the summer Os season, relative to the
full year (see Os ISA, Figures 6-20 and 6-27). However, if we focus the assessment of
risk on the warm season, we bias our estimate by excluding potential effects
associated with cooler (non-summer) months. Given our desire to provide a more
complete picture of overall risk in each of the study areas, we have favored (for the
core analysis) effect estimates that cover the full Os monitoring period specific to
each study area, rather than the more limited warm (summer) period.
Shape of the Functional Form of the Risk Model (including threshold): The Os
ISA concludes that there is little support in the literature for a population threshold for
short-term Os-attributable effects. However, specifically in relation to mortality, the
Os ISA concludes that a national or combined analysis may not be appropriate to
7-29
-------
identify whether a threshold exists (see Os ISA, section 2.S.4.4).22 Given the above
general observation from the Os ISA regarding the low potential for thresholds, we
did not include C-R functions for any of the short-term Os-attributable health
endpoints modeled that incorporated a threshold.23 To reflect uncertainty about the
existence and location of a potential threshold in the C-R function for mortality
attributable to long-term Os exposures, we include a sensitivity analysis for thresholds
ranging from 40 to 60 ppb for the summer season average of daily maximum 1-hr Os
concentrations.
Application of the above criteria resulted in an array of C-R functions specified for the
risk assessment (see Table 7-2), including functions covering short-term Os-attributable
mortality and morbidity and long-term Cb-attributable mortality. In presenting the C-R functions
in Table 7-2, we have focused on describing key attributes of each C-R function (and associated
source epidemiological study) relevant to a review of their use in the risk assessment. More
detailed technical information including effect estimates and model specification is provided in
Appendix 7A. Specific summary information provided in Table 7-2 includes:
• Health endpoints: identifies the specific endpoints evaluated in the study. Generally
we included all of these in our risk modeling, however, when a subset was modeled,
we reference that in the "Notes" column (last column in the table).
• Location: identifies the specific urban areas included in the study and maps those to
the set of 12 urban study areas included in the risk assessment.
• Exposure metric: describes the exposure metric used in the study, including the
specific modeling period (e.g., Os season, warm season, full year). We developed two
categories of composite monitor values to match the modeling periods used in the two
short-term Os-attributable mortality studies providing C-R functions for the analysis.
For the remaining morbidity endpoints, we mapped specific C-R functions to
whichever of these two composite monitor categories most closely matched the
modeling period used in the underlying epidemiological study. This mapping (for
22 Specifically, given the multi-city nature of these mortality studies combined with the variability in Os and other
factors related to exposure and risk, the Os ISA concludes that these studies are not well positioned to evaluate the
potential for a threshold in the mortality effect.
23 While clinical studies have suggested the presence of a threshold for respiratory effects, these should not be used
to support specification of population-level thresholds for use in the epidemiological-based risk assessment
focusing on short-term exposure-related endpoints. The clinical studies focus on relatively small and clearly
defined populations of healthy adults which are not representative of the broader residential populations typically
associated with epidemiological studies, including older individuals and individuals with existing health
conditions which place them at greater risk for O3-related effects. Therefore, the clinical studies are unlikely to
have the power to capture population thresholds in a broader and more diverse urban residential population,
should those thresholds exist.
7-30
-------
morbidity endpoint C-R functions) is described in the "Notes" column (the seasons
reflecting in modeling for each C-R function are also presented in Appendix 7A).
• Additional study design details: this column provides additional information primarily
covering the lag structure and age ranges used in the study.
• Application notes: this column provides notes particular to the application of a
particular epidemiological study and associated C-R functions in the risk assessment.
7.3.3 Baseline Health Effect Incidence and Prevalence Data
As discussed earlier (section 7.1.2), the most common epidemiological-based health risk
model expresses the change in health risk (Ay) associated with a given change in Os
concentrations (Ax) as a percentage of the baseline incidence (y). To accurately assess the impact
of Os air quality on health risk in the selected urban areas, information on the baseline incidence
of health effects (i.e., the incidence under recent air quality conditions) in each location is
needed. In some instances, health endpoints are modeled for a population with an existing health
condition, necessitating the use of a prevalence rate. Where at all possible, we use county-
specific incidences or incidence rates (in combination with county-specific populations). In some
instances, when county-level incidence rates were not available, BenMAP can employ more
generalized regional rates (see BenMAP Guidance Manual for additional detail, Abt Associates,
Inc. 2010). For prevalence rates (which were only necessary for modeling respiratory symptoms
among asthmatic children using Gent et al. (2008) - see Table 7-2), we utilized a national-level
prevalence rate appropriate for the age group being modeled. A summary of available baseline
incidence data for specific categories of effects (and prevalence rates for asthma) is presented
below:
• Baseline incidence data on mortality: County-specific (and, if desired, age- and race-
specific) baseline incidence data are available for all-cause and cause-specific
mortality from CDC Wonder.24 The most recent year for which data are available
online is 2005 and this was the source of incidence data for the risk assessment.25
• Baseline incidence data for hospital admissions and emergency room (ER) visits:
Cause-specific hospital admissions baseline incidence data are available for each of
40 states from the State Inpatient Databases (SID). Cause-specific ER visit baseline
incidence data are available for 26 states from the State Emergency Department
Databases (SEDD). SID and SEDD are both developed through the Healthcare Cost
24 http://wonder.cdc.gov/mortsql.html
25 Note: For years 1999 - 2005, CDC Wonder uses ICD-10 codes; for years prior to 1999, it uses ICD-9 codes.
Since most of the studies use ICD-9 codes, this means that EPA will have to create or find a mapping from ICD-9
codes to ICD-10 codes if the most recent data available are to be used.
7-31
-------
and Utilization Project (HCUP), sponsored by the Agency for Healthcare Research
and Quality (AHRQ). In addition to being able to estimate State-level rates, SID and
SEDD can also be used to obtain county-level hospital admission and ER visit counts
by aggregating the discharge records by county.
• Asthma prevalence rates: state-level prevalence rates that are age group stratified are
available from the Centers for Disease Control and Prevention (CDC) Behavioral
Risk Factor Surveillance System (BRFSS) (U.S. CDC, 2010).
Incidence and prevalence rates are presented as part of the full set of model inputs
documented in Appendix 7A. The incidence rates and prevalence rates provided in Appendix 7A
are weighted average values for the age group associated with each of the C-R functions. These
weighted averages are calculated within BenMAP using more refined age-differentiated
incidence and prevalence rates originally obtained from the data sources listed in the bullets
above.
7.3.4 Population (demographic) Data
To calculate baseline incidence rates, in addition to the health baseline incidence data we
also need the corresponding population. We obtained population data from the 2010 U.S. Census
(http://www.census.gov/popest/counties/asrh/). These data are then used as the basis for back-
casting estimates for simulation years (in this case, 2007 and 2009) (see Appendix J of the
BenMAP User's Manual for additional detail, U.S. EPA, 2012b). Total population counts used in
modeling each of the health endpoints evaluated in the analysis (differentiated by urban study
area and simulation year) are provided as part model inputs presented in Appendix 7A.
7.4 ADDRESSING VARIABILITY AND UNCERTAINTY
An important component of a population risk assessment is the characterization of both
uncertainty and variability. Variability refers to the heterogeneity of a variable of interest within
a population or across different populations. For example, populations in different regions of the
country may have different behavior and activity patterns (e.g., air conditioning use, time spent
indoors) that affect their exposure to ambient Os and thus the population health response. The
composition of populations in different regions of the country may vary in ways that can affect
the population response to exposure to Os - e.g., two populations exposed to the same levels of
Os might respond differently if one population is older than the other. Variability is inherent and
cannot be reduced through further research. Refinements in the design of a population risk
assessment are often focused on more completely characterizing variability in key factors
affecting population risk - e.g., factors affecting population exposure or response - in order to
7-32
-------
produce risk estimates whose distribution adequately characterizes the distribution in the
underlying population(s).
Uncertainty refers to the lack of knowledge regarding the actual values of inputs to an
analysis. Models are typically used in analyses, and there is uncertainty about the true values of
the parameters of the model (parameter uncertainty) - e.g., the value of the coefficient for Os in a
C-R function. There is also uncertainty about the extent to which the model is an accurate
representation of the underlying physical systems or relationships being modeled (model
uncertainty) - e.g., the shapes of C-R functions. In addition, there may be some uncertainty
surrounding other inputs to an analysis due to possible measurement error - e.g., the values of
daily Os concentrations in a risk assessment location, or the value of the baseline incidence rate
for a health effect in a population.26 In any risk assessment, uncertainty is, ideally, reduced to the
maximum extent possible through improved measurement of key variables and ongoing model
refinement. However, significant uncertainty often remains, and emphasis is then placed on
characterizing the nature of that uncertainty and its impact on risk estimates. The characterization
of uncertainty can be both qualitative and, if a sufficient knowledgebase is available,
quantitative.
The selection of urban study areas for the Os risk assessment was designed to cover the
range of Os-related risk experienced by the U.S. population and, in general, to adequately reflect
the inherent variability in those factors affecting the public health impact of Os exposure.
Sources of variability reflected in the risk assessment design are discussed in section 7.4.1, along
with a discussion of those sources of variability which are not fully reflected in the risk
assessment and consequently introduce uncertainty into the analysis.
The characterization of uncertainty associated with risk assessment is often addressed in
the regulatory context using a tiered approach in which progressively more sophisticated
methods are used to evaluate and characterize sources of uncertainty depending on the overall
complexity of the risk assessment (WHO, 2008). Guidance documents developed by EPA for
assessing air toxics-related risk and Superfund Site risks (U.S.EPA, 2004 and 2001, respectively)
as well as recent guidance from the World Health Organization (WHO, 2008) specify multi-
tiered approaches for addressing uncertainty.
The WHO guidance, in particular, presents a four-tiered approach for characterizing
uncertainty (see Chapter 3, section 3.2.6 for additional detail on the four tiers included in the
26 It is also important to point out that failure to characterize variability in an input used in modeling can also
introduce uncertainty into the analysis. This reflects the important link between uncertainty and variability with
the effort to accurately characterize variability in key model inputs actually reflecting an effort to reduce
uncertainty about population means and population variability.
7-33
-------
WHO's guidance document). With this four-tiered approach, the WHO framework provides a
means for systematically linking the characterization of uncertainty to the sophistication of the
underlying risk assessment. Ultimately, the decision as to which tier of uncertainty
characterization to include in a risk assessment will depend both on the overall sophistication of
the risk assessment and the availability of information for characterizing the various sources of
uncertainty. We used the WHO guidance as a framework for developing the approach used for
characterizing uncertainty in this risk assessment.
The overall analysis in the Os NAAQS risk assessment is relatively complex, thereby
warranting consideration of a full probabilistic (WHO Tier 3) uncertainty analysis. However,
limitations in available information prevent this level of analysis from being completed at this
time. In particular, the incorporation of uncertainty related to key elements of C-R functions
(e.g., competing lag structures, alternative functional forms, etc.) into a full probabilistic WHO
Tier 3 analysis would require that probabilities be assigned to each competing specification of a
given model element (with each probability reflecting a subjective assessment of the probability
that the given specification is the "correct" description of reality). However, for many model
elements there is insufficient information on which to base these probabilities. One approach that
has been taken in such cases is expert elicitation; however, this approach is resource- and time-
intensive and consequently, it was not feasible to use this technique in the current Os NAAQS
review to support a WHO Tier 3 analysis.27
For most elements of this risk assessment, rather than conducting a full probabilistic
uncertainty analysis, we have included qualitative discussions of the potential impact of
uncertainty on risk results (WHO Tierl). As discussed in section 7.1.1, for this risk assessment,
we have also expanded the sensitivity analysis considerably to cover a range of model elements
(this represents a WHO Tier 2 analysis). The specific modeling elements covered in the
sensitivity analysis for each health effects endpoint together with the specification of the core
analysis is presented in section 7.4.3. As part of the sensitivity analysis, we have also completed
an influence analysis using estimated elasticities of response28 designed to determine which of
the input factors used in calculating risk are primarily responsible for inter-city variability in risk.
This influence analysis focuses on the response of core short-term exposure-related mortality risk
to inputs since this is one of the key risk metrics completed for the HREA (see section 7.5.3).
27 While a full probabilistic uncertainty analysis was not completed for this risk assessment, we were able to use
confidence intervals associated with effects estimates (obtained from epidemiological studies) to incorporate
statistical uncertainty associated with sample size considerations in the presentation of risk estimates.
28 Elasticities are a measure of sensitivity calculated as the percent change in the response variable for a one percent
change in the input variable.
7-34
-------
In addition to the qualitative and quantitative treatment of uncertainty and variability
which are described here, we have also completed an analysis to evaluate the representativeness
of the selected urban study areas against national distributions for key Os risk-related attributes
to determine whether they are nationally representative or more focused on a particular portion
of the distribution for a given attribute (see Chapter 8, section 8.2.1). In addition, we have
completed a second analysis addressing the representativeness issue, which identified where the
12 urban study areas included in this risk assessment fall along a distribution of national-level
short-term and long-term exposure-related mortality risk (see Chapter 8, section 8.2.2). This
analysis allowed us to assess the degree of which the 12 urban study areas capture locations
within the U.S. likely to experience elevated levels of risk related to Os exposure (for both short-
term and long-term Os-attributable mortality).
The remainder of this section is organized as follows. Key sources of variability which
are reflected in the design of the risk assessment, along with sources excluded from the design,
are discussed in section 7.4.1. A qualitative discussion of key sources of uncertainty associated
with the risk assessment (including the potential direction, magnitude and degree of confidence
associated with our understanding of the source of uncertainty - the knowledge base) is
presented in section 7.4.2. The design of the core analysis and sensitivity analysis completed for
each of the health effect endpoint categories modeled in the risk assessment is discussed in
section 7.4.3.
7.4.1 Treatment of Key Sources of Variability
The risk assessment was designed to cover the key sources of variability related to
population exposure and exposure response, to the extent supported by available data. Here, the
term key sources of variability refers to those sources that we believe have the potential to play
an important role in impacting population risk estimates generated for this risk assessment.
Specifically, we have concluded that these sources of variability, if fully addressed and
integrated into the analysis, could result in adjustments to the core risk estimates which might be
relevant from the standpoint of interpreting the risk estimates in the context of the Os NAAQS
review. The identification of sources of variability as "key" reflects consideration for sensitivity
analyses conducted for previous Ch NAAQS risk assessments (which have provided insights into
which sources of variability can influence risk estimates) as well as information presented in the
O3 ISA.
As with all risk assessments, there are sources of variability which have not been fully
reflected in the design of the risk assessment and consequently introduce a degree of uncertainty
into the risk estimates. While different sources of variability were captured in the risk
assessment, it was generally not possible to separate out the impact of each factor on population
7-35
-------
risk estimates, since many of the sources of variability are reflected collectively in a specific
aspect of the risk model. For example, inclusion of urban study areas from different regions of
the country likely provides some degree of coverage for a variety of factors associated with Os
risk (e.g., air conditioner use, differences in population commuting and exercise patterns,
weather). However, the model is not sufficiently precise or disaggregated to allow the individual
impacts of any one of these sources of variability on the risk estimates to be characterized.
Key sources of potential variability that are likely to affect population risks are discussed
below, including the degree to which they are captured in the design of the risk assessment:
• Heterogeneity in the Effect of Os on Health Across Different Urban Areas: A
number of studies cited in the Os ISA have found evidence for regional heterogeneity
in the short-term Os-attributable mortality effect (Smith et al., 2009 and Bell and
Dominici, 2008, Bell et al., 2004, Zanobetti an Schwartz 2008 - see Os ISA section
6.6.2.2). These studies have demonstrated that differences in effect estimates between
cities can be quite substantial (see Os ISA, Figures 6-32 and 6-33). Therefore, for the
short-term Os-attributable mortality endpoint modeled using Smith et al., 2009-based
effect estimates, we have included Bayes-adjusted city-specific effect estimates
reflecting application of both a regional- and national-prior, both of which are
intended to capture cross-city differences in effect estimates for the mortality
endpoint, while still reflecting input from the more stable regional, or national-level
signal. The national-prior based estimates are included in the core analysis since they
have greater overall power, while the regional-prior based estimates are included as
sensitivity analyses to explore the impact of using regional prior in developing the
Bayes-adjusted estimates (see section 7.4.3).29 For short-term morbidity endpoints,
typically we have used city-specific effect estimates; however, for most endpoints, we
only have estimates for a subset of the urban study areas (typically NY, Atlanta
and/or LA). Therefore, although our risk estimates do reflect the application of city-
specific effect estimates, because we do not have estimates for all 12 urban study
areas, we do not provide comprehensive coverage for heterogeneity in modeling the
respiratory morbidity endpoint category. Long-term Os-attributable mortality has
been shown to demonstrate regional heterogeneity. Specifically, Jerrett et al. (2009)
presented regional effect estimates that demonstrated considerable heterogeneity
ranging from essentially a no-effect (for the Northeast and Industrial Midwest) to
effects substantially larger than the national effect (Southeast, Southwest and Upper
29 Note, that in some instances, there may be insufficient variance between cities to generate city-specific estimates
using a regional prior, which compromises their use in the core analysis.
7-36
-------
Midwest) (see Table 4 in Jerrett et al., 2009). There are many potential explanations
for regional heterogeneity including differences in Os-attributable factors and
potential confounding, potential for the presence of (and regional differences in)
averting behavior, and variation in sample sizes which can impact stability of effect
estimates. For the core analysis, we use a national effect estimate in modeling long-
term exposure related mortality. Consideration of regional effect estimates are
included as a sensitivity analysis (see section 7.4.3 and 7.5.3).
Exposure Measurement Error Associated with Os Effect Estimates: Exposure
measurement error refers to uncertainty associated with using ambient monitor based
exposure surrogate metrics to represent the actual exposure of an individual or
population. As such, this factor can be an important contributor to variability in
epidemiological study results across locations, and uncertainty in results for any
specific city (Os ISA, p. Ixii). Exposure measurement error can result from a number
of factors (e.g., central site monitors not representing actual patterns of personal
exposure including activity patterns, presence of non-ambient sources of exposure for
the pollutant of interest) (Os ISA, Ixii). These factors can vary across urban study
areas (and even within urban study areas), thereby contributing to differences in the
nature and magnitude of exposure measurement error across locations and ultimately
to differences in effect estimates and associated confidence levels. Exposure
measurement error is related to heterogeneity in effect estimates, since regional
differences in effect estimates can result in part, from differences in exposure
measurement error as noted here.
Intra-urban Variability in Ambient Os Levels: The picture with regard to within
city variability in ambient Os levels and the potential impact on epidemiologic-based
effect estimates is somewhat more complicated. The Os ISA notes that spatial
variability in Os levels is dependent on spatial scale with Cb levels being more
homogeneous over a few kilometers due to the secondary formation nature of Os,
while levels can vary substantially over tens of kilometers. Community exposure may
not be well represented when monitors cover large areas with several sub-
communities having different sources and topographies as exemplified by Los
Angeles which displays significantly greater variation in inter-monitor correlations
than does, for example, Atlanta or Boston (see Os ISA, section 4.6.2.1). Despite the
potential for substantial variability across monitors the Os ISA notes that studies have
tended to demonstrate that monitor selection has only a limited effect on the
association of short-term Os exposure with health effects. The likely explanation for
this is that, while absolute values for a fixed point in time can vary across monitors in
7-37
-------
an urban area, the temporal patterns of Os variability across those same monitors
tends to be well correlated. Given that most of the short-term Os-attributable Os
epidemiological studies are time series in nature, the Os ISA notes that the stability of
temporal profiles across monitors within most urban areas means that monitor
selection will have little effect on the outcomes of an epidemiological study
examining short-term Os-attributable mortality or morbidity (see Cb ISA, section
4.6.2.1). For this reason, we conclude that generally intra-city heterogeneity in Os
levels is not a significant factor likely to impact estimates of short-term Os-
attributable risk. One exception is LA which, due to its size and variation in Os
sources and other factors impacting Cb patterns such as topography, displays
significant variation in ambient Os levels with a subsequent impact on risk. However,
in the case of LA (as with the other urban study areas), we model risk using
composite monitors which do not provide spatially-differentiated representations of
exposure and consequently, we do not address this source of variability in the risk
assessment. As discussed in the uncertainty section, short-term exposure mortality
effect estimates for the New York CBSA (Smith et al., 2009) display significant
variability. However, it is not clear which factors are primarily responsible for this
heterogeneity (e.g., differences in the urban structure, residential behavior, or ambient
Os levels within the CBSA). The potential for intra-city heterogeneity in Os levels to
affect risk is more pronounced with long-term Os-attributable mortality where the
relationship between annual trends in ambient Os (as represented using composite
monitor values) and annual mortality is compared between urban study areas in order
to derive effect estimates. Here, pronounced heterogeneity in Os levels within a given
city can result in exposure misclassification, if that heterogeneity is not well
represented by the composite monitor for that city. Different degrees of exposure
misclassification across urban study areas can introduce uncertainty into the overall
national-level effect estimate for long-term exposure-related mortality. Furthermore,
if that exposure measurement error has a regional trend, then measurement error can
potentially result in apparent regional heterogeneity in the effect estimates. The
degree to which there is true regional heterogeneity is made uncertain by the presence
of differential measurement error across regions.
Variability in the Patterns of Ambient Os Reduction across Urban Areas: The
simulated patterns of ambient Os concentrations across an urban area can vary based
on the methodology used to adjust ambient Os concentrations to represent just
meeting the existing or alternative suites of standards. For the 1st draft HREA, we
used a statistical approach called the "quadratic rollback" method for simulating just
7-38
-------
meeting the existing Os standard. Although the quadratic rollback method replicates
historical patterns of air quality changes better than some alternative methods, its
implementation relies on a statistical relationship instead of on a mechanistic
characterization of physical and chemical processes in the atmosphere. Because of its
construct as a statistical fit to measured Os values, the quadratic rollback technique
cannot capture spatial and temporal heterogeneity in Os response and also cannot
account for nonlinear atmospheric chemistry that causes increases in Cb as a result of
emissions reductions of certain Os precursors under some circumstances. As noted in
section 7.1.1, for this HREA, we have employed a model-based Os adjustment
methodology in the risk assessment for simulating Os concentrations under existing
and alternate standard levels. Use of this model-based approach allows the risk
assessment results to more fully account for non-linearities in Os formation and to
reflect spatial and temporal heterogeneity in Os response, including NOx titration
conditions under which a reduction in NOx causes an increase in Os concentrations, in
some core urban locations.
Demographics and Socioeconomic-status (SES)-related Factors: Variability in
population density, particularly in relation to elevated levels of Os has the potential to
influence population risk, although the significance of this factor also depends on the
degree of intra-urban variation in Os levels (as discussed above). In addition,
community characteristics such as pre-existing health status, ethnic composition, SES
and the age of housing stock (which can influence rates of air conditioner use thereby
impacting rates of infiltration of Os indoors) can contribute to observed differences in
Os-related risk (see Os ISA, section 2.5.4.5). Some of the heterogeneity observed in
effect estimates between cities in the multicity studies may be due to these
community characteristics, and while we cannot determine how much of that
heterogeneity is attributable to these factors, the degree of variability in effect
estimates between cities in our analysis should help to capture some of the latent
variability in these community characteristics.
Baseline Incidence of Disease: We collected baseline health effects incidence data
(for mortality and morbidity endpoints) from a number of different sources (see
section 7.3.3). Often the data were available at the county-level, providing a relatively
high degree of spatial refinement in characterizing baseline incidence given the
overall level of spatial refinement reflected in the risk assessment as a whole.
Otherwise, for urban study areas without county-level data, either (a) a surrogate
urban study area (with its baseline incidence rates) was used, or (b) less refined state-
level or national incidence rate data were used.
7-39
-------
7.4.2 Qualitative Assessment of Uncertainty
As noted in section 7.4, we have based the design of the uncertainty analysis carried out
for this risk assessment on the framework outlined in the WHO guidance document (WHO,
2008). That guidance calls for the completion of a Tier 1 qualitative uncertainty analysis,
provided the initial Tier 0 screening analysis suggests there is concern that uncertainty associated
with the analysis is sufficient to significantly impact risk results (i.e., to potentially affect
decision making based on those risk results). Given previous sensitivity analyses completed for
prior Os NAAQS reviews, which have shown various sources of uncertainty to have a potentially
significant impact on risk results, we believe that there is justification for conducting a Tier 1
analysis.
For the qualitative uncertainty analysis, we have described each key source of uncertainty
and qualitatively assessed its potential impact (including both the magnitude and direction of the
impact) on risk results, as specified in the WHO guidance. Similar to our discussion of
variability in the last section, the term key sources of uncertainty refers to those sources that the
we believe have the potential to play an important role in impacting population incidence
estimates generated for this risk assessment (i.e., these sources of uncertainty, if fully addressed
could result in adjustments to the core risk estimates which might impact the interpretation of
those risk estimates in the context of the Os NAAQS review). These key sources of uncertainty
have been identified through consideration for sensitivity analyses conducted for previous Os
NAAQS risk assessments, together with information provided in the final Os ISA and comments
provided by CASAC on the analytical plan for the risk assessment.
Table 7-4 includes the key sources of uncertainty identified for the epidemiological-based
risk portion of the Os HREA. For each source of uncertainty, we have (a) provided a description,
(b) estimated the direction of influence (over, under, both, or unknown) and magnitude (low,
medium, high) of the potential impact of each source of uncertainty on the risk estimates, (c)
assessed the degree of uncertainty (low, medium, or high) associated with the knowledge-base
(i.e., assessed how well we understand each source of uncertainty), and (d) provided comments
further clarifying the qualitative assessment presented.
The categories used in describing the potential magnitude of impact for specific sources
of uncertainty on risk estimates (i.e., low, medium, or high) reflect our consensus on the degree
to which a particular source could produce a sufficient impact on risk estimates to influence the
interpretation of those estimates in the context of the Os NAAQS review.30 Sources classified as
30 For example, if a particular source of uncertainty were more fully characterized (or if that source was resolved,
potentially reducing bias in a core risk estimate), could the estimate of incremental risk reduction in going from
7-40
-------
having a "low" impact would not be expected to impact the interpretation of risk estimates in the
context of the Os NAAQS review; sources classified as having a "medium" impact have the
potential to change the interpretation: and sources classified as "high" are likely to influence the
interpretation of risk in the context of the Os NAAQS review. Because this classification of the
potential magnitude of impact of sources of uncertainty is not based on our direct quantitative
assessments, we use qualitative judgments, in some cases informed by other relevant quantitative
analyses. Therefore, the results of the qualitative analysis of uncertainty are not useful for
making quantitative estimates of confidence, e.g. probabilistic statements about risk. However,
they can be used to support the interpretation of the risk estimates, including the assessment of
overall confidence in the risk estimates. In addition, they can also be used in guiding future
research to reduce uncertainty related to Os risk assessment. As with the qualitative discussion of
sources of variability included in the last section, the characterization and relative ranking of
sources of uncertainty addressed here is based on our consideration of information provided in
previous Os NAAQS risk assessments (particularly past sensitivity analyses), the results of risk
modeling completed for the current Os NAAQS risk assessment and information provided in the
Os ISA as well as earlier Os Criteria Documents. Where appropriate, in Table 7-4, we have
included references to specific sources of information considered in arriving at a ranking and
classification for a particular source of uncertainty.
the current to an alternative standard level change sufficiently to produce a different conclusion regarding the
magnitude of that risk reduction in the context of the Os NAAQS review?
7-41
-------
Table 7-4. Summary of Qualitative Uncertainty Analysis of Key Modeling Elements in the Os NAAQS Risk Assessment.
Source
Description
Potential Influence of
Uncertainty on Risk
Estimates
Direction
Magnitude
Knowledge-
Base
Uncertainty*
Comments
(KB: knowledge base, INF: influence of uncertainty on risk estimates)
A. Adjustment of recent air
quality measurements of
O3 to simulate just meeting
the existing and alternative
standard levels
See Chapter 4 for details
Both
Low-
Medium
Low-medium
See Chapter 4 for more details
B. UseofCBSA-based
study areas in modeling
risk (i.e., potential
mismatch between study
areas used in the HREA
and study areas used in
the epidemiological studies
providing the effect
estimates used in modeling
health effect endpoints)
If the set of monitors used in a
particular urban study area to
characterize population exposure as
part of an ongoing risk assessment do
not match the ambient monitoring data
used in the original epidemiological
study, then uncertainty can be
introduced into the risk estimates. This
uncertainty is balanced in part by the
reduction in bias that results from using
the expanded CBSA definition. (See
section 7.1.1 for more details.)
However, it should be noted that
because these epidemiological studies
occurred in the past, sometimes it can
be impossible to exactly match the
monitors used in the study using recent
air quality data given that monitors may
have moved to a different location or
there may not be measurements
available at specific monitors in the
more recent time period.
Both
Low-
medium
Low-medium
KB and INF: In modeling risk for the HREA, we used CBSA-based study
areas for all health effect endpoints. As discussed in section 7.1.1, the use
of the larger CBSA study areas allows us to better reflect how the change
in air quality affects risk across the entire urban area and to avoid
introducing known bias into the HREA by focusing risk estimates on that
subpopulation living in areas likely to experience potential increases in O3
(and excluding the larger population of urban and suburban areas likely to
experience reductions in ambient O3 levels). While the use of the larger
CBSA-based study areas addresses this source of known bias, it also
introduces uncertainty into the REA since we are no longer matching the
HREA study areas to the study areas in the epidemiological studies
providing the effect estimates used in modeling health effects endpoints.
Given available data, it is not possible at this point to reliably characterize
the degree of uncertainty introduced into the HREA by having this
mismatch in study areas. However, the potential bias avoided through the
use of the larger CBSA study areas (with its acknowledged uncertainty) is
substantial, as illustrated in the sensitivity analyses exploring spatial study
area (see section 7.5.3).
C. Application of C-R
functions based on a
specific temporal and
spatial pattern of
correlations between O3
monitors in an urban area
(as reflected in the
epidemiological study
providing the effect
estimates) to a simulated
change in the patterns of
those correlations when we
estimate risk in the HREA.
The effect estimates used in this risk
assessment reflect a specific spatial
and temporal pattern of ambient O3 (as
represented by the particular monitoring
network providing data for the
underlying epidemiological study).
However, if the spatial and temporal
pattern of O3 levels in the study areas
being modeled differ significantly from
the patterns in the original
epidemiological study (for those same
study areas), then uncertainty can be
introduced into the risk estimates.
Both
Low-
medium
Low-medium
KB and INF: With application of the HDDM adjustment approach, we
simulate potential changes in the spatial and temporal pattern of O3 for a
study areas when just meeting the existing and alternative standards
relative to patterns under recent conditions. This introduces uncertainty into
the application of the original effect estimates, since the exposure
surrogate represented by the composite monitor values may no longer
match that of the underlying epidemiological study. However, it is not
possible, given available data, to characterize quantitatively the magnitude
of this uncertainty. This is probably most true in the urban areas of New
York and Los Angeles where simulation meeting the existing and
alternative standards using the HDDM-adjustment approach relied on large
NOX reductions and there is very little day-to-day variability in the resulting
O3 concentrations.
D. Characterizing intra-
urban population exposure
in the context of
Exposure misclassification within
communities that is associated with the
use of generalized population monitors
Under
(generally)
Low-
medium
Medium
KB and INF: Despite the potential for substantial variability in O3 levels
across monitors (particularly in larger urban areas with greater variation in
sources and topography such as L.A.), the O3 ISA notes that studies have
7-42
-------
Source
Description
Potential Influence of
Uncertainty on Risk
Estimates
Direction Magnitude
Knowledge-
Base
Uncertainty*
Comments
(KB: knowledge base, INF: influence of uncertainty on risk estimates)
epidemiology studies
linking O3to specific health
effects
(which may miss important patterns of
exposure within urban study areas)
introduces uncertainty into the effect
estimates obtained from epidemiology
studies.
tended to demonstrate that monitor selection has only a limited effect on
the association of short-term O3 exposure with health effects (see O3 ISA,
section 4.6.2.1). However, this issue could be more of a concern in larger
urban areas which may exhibit greater variation in O3 levels due to diverse
sources, topography and patterns of commuting. Of particular interest is the
potential for population living in the vicinity of heavily-trafficked roadways to
experience different patterns of exposure relative to more generalized
urban populations (both for O3 and co-pollutants such as PM2.s).
E. Statistical fit of the C-R
functions
Exposure measurement error combined
with other factors (e.g., size of the effect
itself, sample size, control for
confounders) can affect the overall level
of confidence associated with the fitting
of statistical effect-response models in
epidemiological studies.
Both
Medium
(short-term
health
endpoints)
Medium
INF: For short-term mortality and morbidity health endpoints, there is
greater uncertainty associated with the fit of models given the smaller
sample sizes often involved, difficulty in identifying the etiologically relevant
time period for short-term O3 exposure, and the fact that models tend to be
fitted to individual counties or urban areas (which introduces the potential
for varying degrees of confounding and effects modification across the
locations). These studies can also have effects estimates that are not
statistically significant. For this risk assessment, in modeling short-term
mortality, we are not relying on location-specific models. Instead, we are
using city-specific effects estimates derived using Bayesian techniques
(these combine national-scale models with local-scale models). Exposure
measurement error (uncertainty associated with the exposure metrics used
to represent exposure of an individual or population) can also be an
important contributor to uncertainty in effect estimates associated both with
short-term and long-term O3-attributable studies (O3 ISA, p. 1xii). Together
with other factors (e.g., low data density), exposure measurement error can
result in the smoothing of epidemiologically-derived response functions and
the obscuring of thresholds should they exist (O3 ISA, p. Ixix). In addition,
exposure measurement error can vary across different populations even
within the same urban study area. For example a particular group could
have an activity pattern that results in central site monitors (in that urban
study area) being particularly poor at representing that group's exposure to
ambient O3. In this example, an effect estimate derived for that specific
population based on O3 exposure characterized using central site monitors
would have increased uncertainty relative to effect estimates generated for
other population with different activity patterns and lower levels of exposure
measurement error, (for long-term exposure-related mortality, see
discussion of overall confidence in section 7.3.2)
F. Shape of the C-R
functions
Uncertainty in predicting the shape of
the C-R function, particularly in the
lower exposure regions which are often
the focus in O3 NAAQS reviews.
Both
Medium
Low-medium
KB and INF: Studies reviewed in the O3 ISA that attempt to characterize the
shape of the O3 C-R curve for short-term exposure-related mortality along
with possible "thresholds" (i.e., O3 concentrations which must be exceeded
in order to elicit an observable health response) have indicated a generally
linear C-R function with no indication of a threshold (for analyses that have
examined 8-h max and 24-h avg O3 concentrations). However, the ISA
notes that the studies from which the C-R functions are derived indicate
there is less certainty in the shape of the C-R curve at the lower end of the
distribution of O3 concentrations (in the range below 20 ppb) due to the low
density of data in the studies in this range. To reflect uncertainty about the
7-43
-------
Source
Description
Potential Influence of
Uncertainty on Risk
Estimates
Direction Magnitude
Knowledge-
Base
Uncertainty*
Comments
(KB: knowledge base, INF: influence of uncertainty on risk estimates)
existence and location of a potential threshold in the C-R function for
mortality attributable to long-term O3 exposures, we include a sensitivity
analysis for thresholds ranging from 40 to 60 ppb for the summer season
average of daily maximum 1-hr O3 concentrations (see section 7.5.3).
G. Addressing co-
pollutants
The inclusion or exclusion of co-
pollutants which may confound, or in
other ways, affect the O3 effect,
introduces uncertainty into the analysis.
Both
Low-
medium
Medium
KB and INF: The O3 ISA notes that across studies, the potential impact of
PM indices on O3-mortality risk estimates tended to be much smaller than
the variation in O3-mortality risk estimates across cities. This suggests that
O3 effects are independent of the relationship between O3 and mortality.
However, interpretation of the potential confounding effects of PM on O3-
mortality risk estimates requires caution. This is because the PM-O3
correlation varies across regions, due to the difference in PM components,
complicating the interpretation of the combined effect of PM on the
relationship between O3 and mortality. Additionally, the limited PM or PM
component datasets used as a result of the every-3rd- and 6th-day PM
sampling schedule instituted in most cities limits (in most cases) the overall
sample size employed to examine whether PM or one of its components
confounds the O3-mortality relationship (O3 ISA, section 2.5.4.5).
H. Specifying lag structure
(short-term exposure
studies)
There is uncertainty associated with
specifying the exact lag structure to use
in modeling short-term O3-attributable
mortality and respiratory-related
morbidity.
Both
Low-
Medium
Low
KB and INF: The majority of studies examining different lag models suggest
that O3 effects on mortality occur within a few days of exposure. Similar,
studies examining the impact of O3 exposure on respiratory-related
morbidity endpoints suggests a rather immediate response, within the first
few days of O3 exposure (see O3 ISA, section 2.5.4.3). Consequently, while
the exact nature of the ideal lag models remains uncertain, generally, we
are fairly confident that they would be on the order of a day to a few days
following exposure.
I. Using studies from one
geographic area to cover
urban areas outside of the
study area
In the case of Gent et al., 2003 (used in
modeling asthma exacerbations in
Boston), we are using C-R functions
based on an epidemiological study of a
region (northern Connecticut and
Springfield) that does not encompass
the actual urban study area assessed
for risk (Boston).
Both
Medium
Low
INF: Factors related to O3exposure including commuting patterns, exercise
levels etc may differ between the region reflected in the epidemiological
study and Boston. If these differences are great, then applying the effect
estimate from the epidemiological study to Boston could be subject to
considerable uncertainty and potential bias.
J. Characterizing baseline
incidence rates
Uncertainty can be introduced into the
characterization of baseline incidence in
a number of different ways (e.g., error
in reporting incidence for specific
endpoints, mismatch between the
spatial scale in which the baseline data
were captured and the level of the risk
assessment).
Both
Low-
medium
Low
INF: The degree of influence of this source of uncertainty on the risk
estimates likely varies with the health endpoint category under
consideration. There is no reason to believe that there are any systematic
biases in estimates of the baseline incidence data. The influence on risk
estimates that are expressed as incremental risk reductions between
alternative standards should be relatively unaffected by this source of
uncertainty.
KB: The county level baseline incidence and population estimates at the
county level were obtained from data bases where the relative degree of
uncertainty is low.
* Refers to the degree of uncertainty associated with our understanding of the phenomenon, in the context of assessing and characterizing its uncertainty
(specifically in the context of modeling PM risk).
7-44
-------
7.4.3 Description of Core and Sensitivity Analyses
As discussed in section 7.1.1, this risk assessment includes a set of core (higher
confidence) risk estimates which are supplemented by sensitivity analyses. The sensitivity
analyses explore the potential impact that variation in specific model design elements can have
on the core risk estimates. This section specifies which design elements are included in both the
core and sensitivity analyses completed for each of the health effect endpoint categories included
in the risk assessment. We divided the sensitivity analyses into two categories: (a) those
involving air quality characterization and (b) those associated directly with the specification of
the C-R functions used in estimating risk. We recognize that there can be overlap between these
categories with some modeling elements (e.g., modeling period) affecting both the composite
monitor distribution as well as representing an element of C-R function specification. However,
we have retained these two categories to aid in the presentation and discussion of sensitivity
analysis results.31 The sensitivity analyses also included an initial influence analysis designed to
evaluate which of the model inputs are primarily responsible for inter-city variability
(heterogeneity) in risk. The influence analysis uses estimated elasticities of risk with respect to
the risk function input variables, focusing on the short-term exposure-related mortality endpoint
and associated input parameters since this is one of the key risk estimates generated for the
HREA (additional detail on how the influence analysis was conducted is presented in section
7.5.3).
Table 7-5 presents the alternative approaches for adjusting the Os distributions used in
the sensitivity analysis and also identifies the approaches used in the core analysis for each of the
study areas. The alternative air quality adjustment approaches examine the differences in
changes in air quality and risk when applying NOx-only versus NOx and VOC reductions in the
HDDM-adjustment approach. It should be noted that when NOx and VOC reductions were used
in the HDDM-adjustment approach in this sensitivity analysis, the same percent reduction for
both pollutants was used in the air quality adjustment for meeting the existing and alternative
standard in each urban area. More details on these alternative air quality adjustment approaches
are discussed in Chapter 4 and appendices.
31 In presenting both the core and sensitivity analyses, we include both point estimates and 95th percentile confidence
interval (CI), with the latter reflecting the statistical fit of the effect estimates (and hence the power of the
underlying epidemiological study). In comparing core and sensitivity analyses, we not only focus on point
estimates, but also on the CIs since they provide insights into differences in the degree of statistical support for the
effect estimates underlying the risk estimates and therefore, overall confidence in those estimates.
7-45
-------
Besides the approach used to adjust the distributions of Cb, another fact which has a
direct impact on composite monitor composition is the specification of the study area (since this
determines the mix of monitors that will be included in constructing the composite monitor). As
discussed in section 7.1.1, for the core analysis, we modeled all endpoints (for all study areas)
using CBSA-based study areas. For the sensitivity analysis (for the short-term Os-attributable
mortality endpoint), we included a smaller study area based on the original study area definition
used in the Smith et al., 2009 study.32
Table 7-6 presents the model elements included in sensitivity analyses exploring
alternative C-R function specifications. These sensitivity analyses were applied both to short-
term Os-attributable mortality and long-term Os-attributable mortality. As discussed in section
7.1.1, we were not able to differentiate between alternative C-R function specifications for short-
term Os-attributable morbidity endpoints and therefore included the full set of alternative C-R
function specifications in the core analysis. This results in a distribution of core risk estimates for
each endpoint which can be used to gain insights into the impact of different C-R function
specifications on risk. Because separate sensitivity analyses were not completed for short-term
Os-attributable morbidity endpoints, this category is not included in Table 7-6.
32 We did not include an alternative study area simulation as a sensitivity analysis for long-term exposure related
mortality since, for the core analysis, we are using a single (national) effect estimate in modeling this endpoint,
and consequently, the use of an effect estimate from a smaller study area to represent a somewhat larger area (as is
the case with short-term Os-attributable mortality) is likely to introduce less uncertainty.
7-46
-------
Table 7-5. Specification of the Core and Sensitivity Analyses (air quality simulation).
Study Area
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Core Simulation
(type of precursor reduced to
adjust Oa distribution)
NOx-only
NOx-only
NOx-only
NOx-only
NOx & VOC
NOx-only
NOx-only
NOx-lower bound*
NOx-lower bound* (exclude 60)
NOx-only
NOx-only
NOx-only
Sensitivity Analysis
Alternative modeling
approach not evaluated
NOx-only
NOx & VOC
NOx & VOC
NOx & VOC-lower bound*
NOx & VOC-lower bound*
NOx & VOC
NOx & VOC
Alternative modeling
approach not evaluated
A lower-bound fit of the HDDM-based Os sensitivities (reflecting a greater increment of Os reduction per
unit of VOC and/or NOX reduction) was required in simulation of the alternative standard levels.
Table 7-6. Specification of the Core and Sensitivity Analyses (alternative C-R function
specification).
Health Effect
Endpoint
Category
Modeling Elements Included
Core Analysis
Sensitivity Analysis
Short-term
03-
attributable
mortality
- Full monitoring period
(specific to each study area),
8-hr max metric, national-
Bayes adjusted, single
pollutant model.
- effect estimates obtained
from: Smith et al. (2009)
- summer (warm month), 8-hr mean, regional-Bayes
adjusted, multi-pollutant (with PMio).
- effect estimates obtained from Zanobetti and Schwartz
(2008) and Smith et al. (2009)
Long-term
03-
attributable
mortality
- Single national estimate,
two-pollutant model (PM2§),
long-term peak trend metric
(based on daily 1-hr max
values), CBSA-based study
area.
- effect estimates obtained
from Jerrett et al. (2009)
study
- Regional-differentiated effect estimates, single pollutant
model.
- National-level effect estimate, single pollutant model.
- Effect estimates reflecting consideration for potential
thresholds in the respiratory mortality effect (ranging from
40 to 60 ppb)
- All effect estimates for this sensitivity analysis were also
obtained from Jerrett et al., 2009 study (with the exception
of the threshold models - which were obtained through
direct communication with the authors; Sasser, 2014)
7-47
-------
7.5 URBAN STUDY AREA RESULTS
This section discusses risk estimates generated for the set of 12 urban study areas,
including both the core risk estimates and accompanying sensitivity analyses. In summarizing
risk estimates, this discussion focuses on results most relevant to two policy-related questions:
(a) to what extent is the existing Os standard protective of public health and, (b) what is the
nature and magnitude of additional public health protection provided by the suite of alternative
standards under consideration? Consequently, we focus on two types of risk estimates including
the magnitude of Os-attributable risk after simulation of just meeting the existing standard and
the degree of risk reduction potentially provided by each of the alternative standards relative to
just meeting the existing standard.33
This section is organized as follows. We begin by presenting the core risk estimates in
both tabular and graphical format at the end of this section. We then present key observations
about the risk estimates for just meeting the existing standard (for core risk) in section 7.5.1. Key
observations related to risk estimates for just meeting alternative standard levels, and for
estimates of risk changes comparing alternative standards to just meeting the existing standard
(again, for core risk) are presented in section 7.5.2. After presenting key observations related to
the core risk estimates, we then present key observations resulting from the sensitivity analyses
(section 7.5.3).
A number of details regarding the design of the core risk assessment should be kept in
mind when reviewing the core risk estimates presented in this section (see section 7.1.1 for
additional detail on these design elements):
• All risk estimates reflect application of a CBSA-based study area.
• Estimates are presented for two simulation years (2007 and 2009)
• Short-term Os-attributable mortality estimates are generated for all 12 urban study
areas, while most short-term Os-attributable morbidity estimates (depending on the
specific health endpoint) are generated for only a subset of urban study areas. Long-
term Os-attributable mortality is modeled for all 12 urban study areas.
• For all health effect endpoints, we model risk down to zero Os and do not include
either consideration for LML or alternative threshold levels.
There are several categories of risk metrics generated for the core mortality and
morbidity endpoints modeled in this analysis. Below we describe both the types of risk metrics
33 As part of this HREA, we have also generated estimates of risk under recent conditions as well as estimates of the
degree of risk reduction (relative to risk under recent conditions) associated with adjusting air quality to just meets
the existing standard. See Appendix 7B.
7-48
-------
generated for the core analysis and the specific types of tables and figures used in presenting
those metrics.
I. Core short-term Ch-attributable mortality estimates
• Table presenting estimates of Os-attributable mortality risk after just meeting
the existing standard and the estimated change in mortality associated with
meeting each of the alternative standard levels relative to the existing standard
(Table 7-7): These estimates include point estimates and 95th percentile confidence
intervals representing uncertainty associated with the statistical fit of the effect
estimates.
• Table presenting estimates of the percent of total mortality attributable to Os
after just meeting the existing standard and the percent reduction in Os-
attributable risk associated with each alternative standard (Table 7-8).
• Heat maps for mortality illustrating distribution across daily Os levels of total
Os-attributable risk after just meeting the existing standard and risk reductions
after meeting alternative standards (Figure 7-2 and Figure 7-3): Heat maps are
provided for each of the 12 urban areas. The color gradient in each figure reflects the
distribution of mortality (or the change in mortality) across the range of daily 8-hr Cb
levels and provides a visual tool to explore trends in mortality counts across daily Os
levels and between urban study areas. Visual patterns in the figures presenting total
risk and risk reduction are interpreted differently:
o For figures depicting total Ch-attributable risk, colors range from blue (lower
mortality count) to red (higher mortality count). As an example, with Figure 7-2,
top heat map (which presents total Os-attributable risk for the existing standard in
2007, based on Smith et al., 2009 C-R functions), if we focus on the first row
(Atlanta, GA), we see a value of 31 under the column 55-60 ppb. This value
reflects the fact that 31 of the 270 Os-attributable deaths estimated for Atlanta
after just meeting the existing 75 ppb standard in 2007 occurred on days with
composite monitor Os levels between 55 and 60 ppb. Similarly, in the same row,
we see that only two Cb attributable deaths occurred on days when the composite
monitor value was between 20 and 25 ppb. We also include the total Os-
attributable mortality (for each study area) in the final column marked "Total".
7-49
-------
o For figures depicting changes in risk associated with simulation of existing and
alternative standard levels, we see that the pattern is more complex since we can
have a combination of increases and decreases in risk in the heat maps, with
increases in risk identified as red to yellow and decreases in risk identified as
yellow to blue. Increases in risk are negative numbers, decreases are positive. In
addition, in the final three columns of each map, we provide estimates of the total
Os-attributable mortality, as well as the total broken down into the subtotals
across days with increases (negative) and days with decreases (positive) in that
mortality. The increase and decrease for a given study area should sum
(accounting for rounding in these subtotals) to the overall total for Os-attributable
deaths for that study area.34 Several factors can contribute to the patterns of
changes in Os-attributable risk reflected in these maps. For example, non-
linearities in Os formation can result in increases in Os on some days, even when
simulating air quality just meeting a lower alternative standard (see section 7.1.1).
In addition, simulation of alternative standard levels can result in a change in the
overall distribution of the composite monitor ambient Os distribution. Often, this
change will take the form of a shift in the upper tail of the distribution towards the
mean, given that simulated air quality just meeting alternative standard levels
targets higher Os days. If we look at Figure 7-2 at the second map (Decrease 75 to
70) and specifically at the row for Houston, we see that there is a -3 increase in
deaths distributed across 20-35 ppb days and a decrease in deaths of 8, primarily
distributed across 40-60 ppb days.
• Graphic plots of Os-attributable deaths per 100,000 population for just meeting
the existing and alternative standards (Figure 7-4): This plot provides estimates
that are adjusted for the size of the underlying urban population, thereby allowing the
mortality estimates and associated trends to be more readily compared across urban
study areas (consideration of absolute Os mortality is complicated by the role that
underlying urban population plays in driving total Os-attributable mortality - larger
study areas like Los Angeles and New York having substantially larger mortality
estimates primarily due to their higher underlying populations). These figures allow
us to evaluate the overall magnitude of risk reductions across standard levels and
determine the degree to which those trends differ for different study areas.
34 The "Change in Risk" totals at the right edge of each heat map may not sum exactly to the total (for a given row)
due to rounding.
7-50
-------
Tables summarizing Os-attributable morbidity counts, percent of baseline incidence and
percent reduction in Os-attributable risk for short-term Os-attributable morbidity (Table 7-9
through Table 7-11): Three categories of short-term Os-attributable morbidity effects were
modeled for the analysis (respiratory related HA, respiratory-related ER visits and asthma
exacerbations). As discussed in section 7.1.1, these morbidity effects were modeled for a
combination of all 12 urban study areas and a subset of those study areas depending on the
endpoint (see below). The C-R functions available for modeling many of these morbidity
endpoints included consideration for a number of design elements (e.g., co-pollutants and lag
structure). However, as noted earlier in section 7.1.1, for short-term exposure morbidity
endpoints with multiple C-R functions, we were not able to differentiate between C-R functions
in terms of overall confidence and consequently we could not identify a single core model.
Therefore, when we have multiple C-R functions reflecting different treatments of key design
elements such as lag structure, we consider the risk estimates that result from the full set of C-R
functions to represent a core range of risk. Each of the tables summarizing short-term Os-
attributable morbidity risk present several risk metrics including: (a) total Os-attributable counts
(after just meeting the existing standard), (b) reductions in Os-attributable counts (for each of the
alternative standard levels relative to just meeting the existing standard), (c) percent of baseline
incidence attributable to Os (after just meeting the existing standard) and (d) percent reductions
in Cb-attributable risk (for each of the alternative standard levels). In presenting these morbidity
risk estimates, we do not include 95th percentile confidence intervals (CIs) in order to conserve
space (CIs are included in the detailed tables in Appendix 7B). Specific tables summarizing these
morbidity risk estimates include:
o HA visits (for respiratory symptoms including asthma): Table 7-9 presents
estimates of HA (for respiratory symptoms, chronic lung disease and asthma).
Risk estimates are generated for a subset of the urban study areas for some of the
health endpoints (e.g., New York City for HA [chronic lung disease and asthma]),
while HA (respiratory-related) estimates cover all 12 urban study areas.
o ER visits (for respiratory symptoms including asthma): Table 7-10 presents
estimates of ER visits (for respiratory symptoms and asthma) specifically for New
York City and Atlanta based on C-R functions obtained from several
epidemiological studies.
o Asthma exacerbations: Table 7-11 presents estimates of asthma exacerbations
(including estimates for a range of symptoms) for Boston, the only urban study
area with C-R functions supporting modeling for this endpoint.
7-51
-------
• Graphic plots of Os-attributable respiratory-related HA per 100,000 residents
for the existing and alternative standard levels (Figure 7-5): This figure is
intended to complement Figure 7-4 which presents the same type of risk information
for short-term Os-attributable mortality. By plotting respiratory HA per 100,000, we
adjust for the underlying population which makes trends in risk more comparable
across urban study areas. We have only created this graphic for respiratory HA (based
on application of Medina-Ramon et al., 2006) since that is the only morbidity
endpoint modeled for all 12 urban study areas. As with the mortality figure, this
figure allows us to evaluate the overall magnitude of risk reductions across standard
levels and determine the degree to which those trends differ for different study areas.
III. Core long-term Ch-attributable mortality estimates
• Table presenting estimates of long-term Os-attributable mortality including total
risk after just meeting the existing standard and risk reductions based on
comparing risks after meeting alternative standards to risks after meeting the
existing standard (Table 7-12): Estimates presented in Table 7-12 reflect respiratory
mortality and include 95th percentile confidence intervals representing uncertainty
associated with the statistical fit of the effect estimates used. Estimates presented in
these tables allow for consideration for the magnitude of risk associated with just
meeting the existing standard and the pattern of risk reduction in meeting alternative
standards relative to the existing standard.
• Table presenting estimates of the percent of respiratory mortality attributable to
Os and percent reductions in Os-attributable risk for long-term Os-attributable
mortality (Table 7-13).
Graphic plots of Os-attributable deaths per 100,000 population for just meeting the
existing and alternative standards (Figure 7-6): This plot provides estimates that are adjusted
for the size of the underlying urban population, thereby allowing the mortality estimates and
associated trends to be more readily compared across urban study areas (consideration of
absolute Os mortality is complicated by the role that underlying urban population plays in
driving total Os-attributable mortality - larger study areas like Los Angeles and New York
having substantially larger mortality estimates primarily due to their higher underlying
populations). As with similar plots for short-term exposure-related mortality and morbidity, these
figures allow us to evaluate the overall magnitude of risk reductions across standard levels and
determine the degree to which those trends differ for different study areas.
7-52
-------
Table 7-7. Short-Term O3-attributable All-Cause Mortality, 2007 and 2009 Air Quality.
Smith et al. (2009) C-R Functions, Os season, CBSA-based study area, no threshold.
Study Area
Air Quality Scenario
Total Ozone-
Attributable Deaths
75ppb
Change in Ozone-Attributable Deaths
75-70 75-65 75-60
2007 Simulation Year
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
220
(-310-740)
230
(-130-570)
200
(-290-670)
270
(-25 - 550)
58
(-190-300)
520
(26-990)
580
(110-1000)
750
(-310-1800)
3200
(1900-4500)
920
(200-1600)
160
(-170-480)
350
(-86-770)
10
(-13-32)
7
(-4-17)
4
(-6-14)
8
(-1-18)
1
(-4-7)
18
(1-35)
4
(1-8)
26
(-11-62)
150
(92 - 220)
26
(6-46)
3
(-3-9)
15
(-4-33)
18
(-24-60)
14
(-8-35)
11
(-16-39)
20
(-2-41)
3
(-10-15)
33
(2-64)
9
(2-17)
52
(-22 - 130)
740
(440-1000)
56
(12-100)
6
(-6-17)
31
(-8-70)
28
(-39-95)
23
(-13-59)
18
(-25-62)
40
(-4-83)
5
(-17-27)
54
(3-110)
20
(4-37)
96
(-40-230)
NA
NA
86
(19-150)
10
(-11-31)
49
(-12-110)
2009 Simulation Year
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
200
(-280-670)
210
(-110-520)
180
(-260-610)
250
(-23-510)
56
(-180-290)
460
(23-880)
600
(110-1100)
770
(-320-1800)
3000
(1800-4200)
820
(180-1400)
160
(-170-490)
310
(-77-690)
7
(-10-24)
4
(-2-10)
-1
(1--2)
7
(-1-15)
0
(-1-1)
-17
(-1--33)
-1
(0--1)
25
(-10-60)
96
(57-130)
14
(3-25)
3
(-3-8)
7
(-2-15)
13
(-18-45)
9
(-5-23)
3
(-4-10)
18
(-2-37)
1
(-4-7)
-5
(0--10)
3
(1-6)
53
(-22 - 130)
500
(300-700)
33
(7-58)
5
(-6-17)
17
(-4-37)
19
(-26-64)
14
(-8-37)
8
(-11-27)
31
(-3-64)
5
(-15-25)
12
(1-23)
12
(2-22)
98
(-41 - 240)
NA
NA
51
(11-90)
9
(-10-28)
30
(-7-67)
NA: for NY, the model-based adjustment methodology was unable to adjust Os distributions such that they would meet the lower
alternative standard level of 60 ppb.
"0" counts denote non-zero estimates that round to zero.
7-53
-------
Table 7-8. Percent of Total All-Cause Mortality Attributable to O3 and Percent Change in
Os-Attributable Risk, 2007 and 2009 Air Quality. Smith et al. (2009) C-R functions, O3
season, CBSA-based study area, no threshold.
Study Area
Air Quality Scenario
% of Baseline
Incidence
75ppb
% Change in O3-Attributable
Risk
75-70
75-65
75-60
2007 Simulation Year
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
1.1
1.9
1.2
2.4
0.8
3.0
1.9
1.0
4.1
3.2
1.2
2.5
4
3
2
3
2
3
1
3
5
3
2
4
8
6
5
7
5
6
2
7
22
6
3
9
13
10
9
14
9
10
3
13
NA
9
6
14
2009 Simulation Year
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
1.0
1.8
1.1
2.3
0.8
2.7
1.9
1.1
4.0
3.0
1.2
2.3
3
2
-0.3
3
0.3
-4
-0.1
3
3
2
2
2
7
4
2
7
2
-1
0.5
7
16
4
3
5
9
7
4
12
8
3
2
13
NA
6
6
9
NA: for NY, the model-based adjustment methodology was unable to adjust Os distributions such that they would meet the lower
alternative standard level of 60 ppb.
7-54
-------
Figure 7-2. Heat Maps for Short-term O3-attributable Mortality, 2007 Air Quality
Adjusted to Just Meet the Existing Standard and Risk Reductions from Just Meeting
Alternative Standards. Smith et al. (2009) C-R functions. (See Key at bottom of figure).
Current Standard (75)
Study are a
Atlanta, GA
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Daily Shr Max Ozone Level (ppb)
0-5
0
0
0
0
0
0
0
0
0
0
0
5-10
0
0
0
0
0
0
0
0
0
0
0
10-15
0
0
0
0
0
0
0
0
0
0
0
15-20
0
0
0
0
0
0
0
2
0
1
20-25
2
2
0
1
14
0
24
0
1
2
25-30
4
11
0
5
42
0
113
25
8
6
30-35
15
26
25
1
33
107
0
341
46
23
15
35-40
20
29
3
56
124
10
625
115
43
52
40-45
34
33
55
4
97
126
204
851
157
29
53
45-50
43
33
50
9
116
81
268
545
175
29
61
50-55
52
20
27
12
59
42
233
418
155
17
60
55-60
31
12
25
15
41
42
27
268
122
9
38
60-65
12
17
19
10
44
2
8
45
2
24
65-70
5
5
8
3
16
0
3
0
1
23
70-75 >75
3
-T-
6
1
34
0
0
0
0
10
=F
0
0
14
0
0
0
0
3
Total
222
228
202
268
58
516
580
753
3,230
916
161
348
Decrease 75 to 70
Study are a
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Daily Shr Max Ozone Level (ppb)
0
0
0
0
0
0
0
0
0
0
0
0
510
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
30-35
0
0
0
0
0
0
-1 1 -1 1 -1
0
-1
0
±
0
0
0
0 0 1 0
0 | 0 | 0
35-40
0
0
0
0
0
0
0
0
14
0
0
1
40-45
1
1
0
1
0
2
2
4
31
2
1
2
45-50
50-55
55-60
2
2
0
2
1
2
1
29
6
0
2
60-65
1
1
1
1
0
0
0
6
4
0
2
65-70
0
0
0
1
0
0
0
0
2
0
2
70-75
0
0
0
0
0
0
0
0
0
0
1
>75
0
0
0
0
0
0
0
0
1
0
0
Total
Change in risk
Inc.
0
0
0
0
0
0
-3
0
-13
-2
0
0
Dec.
10
6
3
10
1
19
8
25
167
27
4
16
Decrease 75 to 65
Study are a
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Daily Shr Max Ozone Level (ppb)
0-5
0
0
0
0
0
0
0
0
0
0
0
0
5-10
0
0
0
0
0
0
0
0
0
0
0
0
10-15
0
0
0
0
0
0
0
0
0
0
0
0
15-20
0
0
0
0
0
0
0
0
0
0
0
0
20-25
0
0
0
0
0
0
-2
0
-1
0
0
0
25-30
0
0
0
0
0
0
-2
0
2
-1
0
0
30-35
1
0
0
0
0
0
2
0
35-40
1
0
1
1
0
0
0
0
98
0
1
2
40-45
2
2
1
4
0
3
4
8
172
5
1
4
45-50
4
2
2
4
0
7
4
20
156
11
2
6
50-55
5
3
2
3
1
5
3
21
156
13
1
6
55-60
1 3
4
60-65
1
2
3
1
5
0
1
22
9
0
3
65-70
1
1
1
0
2
0
0
0
4
0
3
70-75
0
1
1
0
5
0
0
0
1
0
2
>75
0
0
0
0
2
0
0
0
1
0
0
Total
18
14
20
3
33
9
52
735
56
6
31
Change in risk
Inc.
0
0
0
-1
0
-2
-8
0
-7
-4
-1
0
Dec.
18
15
12
20
3
35
16
52
742
60
6
31
Decrease 75 to 60
Study are a
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Daily Shr Max Ozone Level (ppb)
0-5
0
0
0
0
0
0
0
0
0
0
0
5-10
0
0
0
0
0
0
0
0
0
0
0
10-15
0
0
0
0
0
0
0
0
0
0
0
15-20
0
0
0
0
0
0
0
0
0
0
0
20-25
0
0
0
0
0
0
-2
0
0
0
0
25-30
0
0
0
0
0
0
-4
0
-1
-1
0
30-35
1
0
0
0
0
-1
-4
0
-1
-1
0
35-40
2
0
40-45
4
3
2
7
0
6
7
24
8
3
6
45-50
6
4
3
9
0
11
8
35
NA
17
3
9
50-55
7
5
3
6
1
8
6
29
19
2
10
55-60
5
6
2
6
2
7
7
4
20
1
7
60-65
13
0
5
65-70
1
1
1
2
1
4
0
1
6
0
5
70-75
1
1
2
2
0
7
0
0
1
0
2
>75
0
0
1
0
0
3
0
0
2
0
1
Total
28
23
18
40
5
54
20
96
86
10
49
Change in risk
Inc.
0
0
0
-2
0
-2
-11
0
-4
-2
0
Dec.
29
25
19
41
6
57
31
95
89
11
49
NA: for NY, the model-based adjustment methodology was unable to adjust Os distributions such that they would meet the lower
alternative standard level of 60 ppb.
Key: For current standard (75) which is an absolute risk metric expressed as total Os-attributable deaths, color gradient ranges
from blue (smallest Os-related mortality count) to red (highest Os-related mortality count). For estimates of decreases in risk,
color gradient ranges from red (increase in risk - negative cell values) to blue (reduction in risk - positive cell values).
7-55
-------
Figure 7-3. Heat Maps for Short-term O3-attributable Mortality, 2009 Air Quality
Adjusted to Just Meet the Existing Standard and Risk Reductions from Just Meeting
Alternative Standards. Smith et al. (2009) C-R functions. (See Key at bottom of figure).
Current Standard (75)
Study are a
Atlanta, GA
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Daily Shr Max Ozone Level (ppb)
0-5
0
0
0
0
0
0
0
0
0
0
0
5-10
0
0
0
0
0
0
0
0
0
0
0
10-15
1
0
0
0
1
0
0
0
0
0
1
15-20
0
0
0
7
5
0
7
2
0
5
20-25
3
0
5
0
41
12
1
5
25-30
13
16
1
21
0
246
38
10
14
30-35
26
28
2
36
1
489
118
28
22
35-40
28
42
3
53
10
407
93
30
44
40-45
29
46
6
89
96
168
724
162
32
42
45-50
50
12
116
77
196
538
130
24
63
50-55
35
30
72
297
314
151
18
53
55-60
25
40
31
91
201
67
14
60-65
8
36
23
5
64
50
3
65-70
1
0
6
0
0
0
0
70-75
0
0
0
17
3
0
0
0
0
0
>75
0
2
0
0
5
3
0
0
0
0
0
Total
201
183
249
56
456
595
770
3,031
822
162
310
Decrease 75 to 70
Study are a
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Daily Shr Max Ozone Level (ppb)
0
0
0
0
0
0
0
0
0
0
0
0
510
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
-1
0
0
0
0
0
0
0
0
0
0
0
-2
-1
0
-1
0
0
-1
0
0
-1
0
0
-2
0
-A
-1
0
0
0
0
0
0
0
-2
0
-16
-2
0
-1
30-35
0
0
-1
0
0
-3
0
-9
-2
0
0
35-40
1
0
0
1
0
-1
0
9
-1
0
0
40-45
1
1
0
1
0
1
3
26
3
1
1
45-50
2
1
0
2
0
1
6
36
4
1
2
50-55
2
1
0
2
0
0
2
12
26
6
1
2
55-60
60-65
1
0
0
0
0
2
1
0
7
3
0
1
65-70
0
0
0
0
0
0
0
0
0
0
0
0
70-75
0
0
0
0
0
1
0
0
0
0
0
0
>75
0
0
0
0
0
0
0
0
0
0
0
0
Total
Change in risk
Inc.
0
0
-3
0
0
-22
-9
0
-44
-6
0
-2
Dec.
9
Decrease 75 to 65
Study are a
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Daily Shr Max Ozone Level (ppb)
0-5
0
0
0
0
0
0
0
0
0
0
0
0
5-10
0
0
0
0
0
0
0
0
0
0
0
0
10-15
0
0
0
0
0
-1
0
0
0
0
0
0
15-20
0
0
0
0
0
-3
-1
0
-1
0
0
-1
20-25
-1
0
-1
0
0
-1
-4
0
-5
-2
0
-1
25-30
-1
0
-1
-1
0
-5
-4
0
-19
-3
-1
-1
30-35
0
0
-1
0
0
-4
-5
0
18
-3
0
0
35-40
1
0
0
2
0
-3
-1
0
60
-1
1
1
40-45
3
1
1
3
0
-2
2
6
122
8
2
2
45-50
4
8
50-55
3
2
2
4
0
1
25
13
1
5
55-60
3
2
0
3
1
4
8
7
1
5
60-65
1
1
0
1
0
4
0
6
0
1
65-70
0
0
0
1
0
0
0
0
0
0
1
70-75
0
0
1
0
0
0
0
0
0
0
>75
0
0
0
0
0
0
0
0
0
0
0
Total
13
9
3
18
1
3
53
500
33
5
17
Change in risk
Inc.
-2
-1
-4
-1
0
-15
0
-48
-11
-2
-4
Dec.
15
11
6
21
1
16
19
53
550
44
7
22
Decrease 75 to 60
Study are a
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Daily Shr Max Ozone Level (ppb)
0-5
0
0
0
0
0
0
0
0
0
0
0
5-10
0
0
0
0
0
0
0
0
0
0
0
10-15
0
0
0
0
0
-1
0
0
0
0
-1
15-20
0
0
0
0
0
-4
-2
0
-1
0
-2
20-25
-1
0
-1
0
0
-2
-6
0
-2
0
-1
25-30
-1
-1
-1
0
0
-6
-6
0
-4
-1
-1
30-35
0
0
-1
1
0
-4
-6
0
-3
-1
0
35-40
2
1
1
4
0
-2
0
1
0
1
3
40-45
4
2
1
5
0
1
4
19
12
3
4
45-50
5
5
3
8
0
8
7
26
NA
12
3
8
50-55
4
4
3
7
1
3
9
37
19
2
8
55-60
3
0
60-65
2
1
0
2
1
6
5
1
8
0
2
65-70
0
0
0
1
0
0
1
0
0
0
1
70-75
0
0
1
0
0
4
1
0
0
0
0
>75
0
0
0
0
0
1
1
0
0
0
0
Total
19
14
8
31
5
12
12
98
51
9
30
Change in risk
Inc.
-2
-2
-4
-1
0
-22
-22
0
-14
-2
-6
Dec.
21
16
11
33
5
32
35
97
65
11
34
NA: for NY, the model-based adjustment methodology was unable to adjust Os distributions such that they would meet the lower
alternative standard level of 60 ppb.
Key: For current standard (75) which is an absolute risk metric expressed as total Os-attributable deaths, color gradient ranges
from blue (smallest Os-related mortality count) to red (highest Os-related mortality count). For estimates of decreases in risk,
color gradient ranges from red (increase in risk - negative cell values) to blue (reduction in risk - positive cell values).
7-56
-------
2007 Simulation year
Trend in ozone-related mortality across standard
levels (deaths per 100,000)
-Atlanta, GA
• Baltimore, MD
Boston, MA
-Cleveland, OH
-Denver, CO
Detroit, Ml
•Houston, TX
-Los Angeles, CA
New York, NY
-Philadelphia, PA
-Sacramento, CA
-St. Louis, MO
75ppb
70ppb
65ppb
60ppb
2009 Simulation year
Trend in ozone-related mortality across standard levels
(deaths per 100,000)
I18
lie
| 12
£
| 10
-Atlanta, GA
-Baltimore, MD
Boston, MA
-Cleveland, OH
-Denver, CO
Detroit, Ml
-Houston, TX
-Los Angeles, CA
New York, NY
-Philadelphia, PA
-Sacramento, CA
St. Louis, MO
75ppb
70ppb
65ppb
60ppb
Figure 7-4. Short-Term O3-attributable All-Cause Mortality for 2007 (top panel) and 2009
(bottom panel) Air Quality Adjusted to Just Meet the Existing and Alternative Standards.
Smith et al. (2009) C-R functions.
7-57
-------
Table 7-9. Short-Term Os-attributable Morbidity Counts, Percent of Baseline and
Reduction in Os-attributable Risk, Respiratory-Related Hospital Admissions, 2007 and
2009 Air Quality.
Endpoint/Study Area/Descriptor
Air Quality Scenario
Total Ozone-
Attributable
Counts
75ppb
Change in Ozone-Attributable
Counts
75-70
75-65
75-60
Percent of
Baseline
75ppb
% Change in Ozone-Related
Risk
75-70
75-65
75-60
2007 Simulation Year
HA (respiratory); Detroit (Katsouyanni et al., 2009)
Ihr max, penalized splines
Ihr max, natural splines
190
180
10
9.8
18
18
29
28
HA (respiratory); NYC (Silverman and Ito, 2010; Lin et al., 2008)
HA(
HA Chronic Lung Disease (Lin)
HA Asthma (Silverman)
HA Asthma, PM2.5 (Silverman)
140
490
360
7.9
33
23
34
140
98
NA
2.8
2.7
3.3
27.7
20.2
5
5
10
10
15
15
5
5
5
23
21
22
NA
respiratory); LA (Linn etal., 2000)
Ihr max penalized splines
480
11
23
36
HA (COPD less asthma); all 12 study areas (Medina- Ramon, et al., 2006)
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
55
40
58
37
18
71
57
110
200
97
15
43
3
1
1
1
1
2
1
5
13
3
1
2
5
3
3
3
1
4
2
10
57
7
1
4
8
5
6
6
2
7
3
15
NA
11
2
7
2.4
2.5
2.6
2.2
2.4
2.9
2.5
2.1
2.7
2.2
2.5
2.5
2.6
2
5
7
5
4
2
4
3
3
1
4
6
3
3
5
9
7
6
8
6
6
3
9
28
7
7
10
15
12
9
17
11
10
6
13
NA
11
11
15
2009 Simulation Year
HA (respiratory); Detroit (Katsouyanni et al., 2009)
HA(
HA(
Ihr max, penalized splines
Ihr max, natural splines
170
160
2.8
2.7
10
9.8
20
19
respiratory); NYC (Silverman and Ito, 2010; Lin et al., 2008)
HA Chronic Lung Disease (Lin)
HA Asthma (Silverman)
HA Asthma, PM2. 5 (Silverman)
140
470
350
respiratory); LA (Linn etal., 2000)
Ihr max penalized splines
500
5.9
28
20
25
110
79
NA
11
23
37
HA (COPD less asthma); all 12 study areas (Medina- Ramon, et al., 2006)
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
52
37
53
36
18
64
63
120
190
88
16
41
3
1
0
1
0
-3
0
5
8
2
1
2
4
2
1
3
1
-1
1
10
40
4
1
3
6
3
2
5
2
1
3
16
NA
6
2
5
2.5
2.4
3.2
27.3
19.9
2.4
2.2
2.3
2.0
2.2
2.7
2.2
2.2
2.7
2.1
2.3
2.4
2.4
2
2
6
6
11
11
4
4
4
17
17
18
NA
2
4
7
5
2
0
3
1
-4
1
4
4
2
3
3
8
5
1
8
4
-2
2
8
20
4
7
8
12
8
4
14
11
1
5
13
NA
7
11
12
NA: for NY, the model-based adjustment methodology was unable to adjust Os distributions such that they would meet the lower
alternative standard level of 60 ppb.
"0" counts denote non-zero estimates that round to zero.
7-58
-------
Table 7-10. Short-Term Os-attributable Morbidity Counts, Percent of Baseline and
Reduction in Os-attributable Risk, Emergency Department Visits, 2007 and 2009 Air
Quality.
End poi nt/Study Area/Descri ptor
Air Quality Scenario
Total Ozone-
Attributable
Counts
75ppb
Change in Ozone-Attributable
Counts
75-70
75-65
75-60
Percent of
Baseline
75ppb
% Change in Ozone-Related
Risk
75-70
75-65
75-60
2007 Simulation Year
ER Visits (repiratory); Atlanta (Strickland etal., 2007)
Distributed lag 0-7 days
Average day lag 0-2
6,600
3,900
350
200
650
370
1,000
580
ER-visits (respiratory); Atlanta (Tolbert et al., 2007, Darrowetal., 2011)
Tolbert
Tolbert-CO
Tolbert-N02
Tolbert-PMlO
Tolbert-PMlO, N02
Darrow
7,000
6,300
5,700
4,400
4,300
3,800
ER-visits (asthma); NYC (Ito et al, 2007)
single pollutant model
PM2.5
N02
CO
S02
11,000
8,300
6,800
11,000
8,500
310
280
250
200
190
170
580
510
460
360
350
310
920
810
730
570
550
490
620
480
390
660
490
2,700
2,100
1,700
2,900
2,200
NA
19.6
11.6
5.8
5.1
4.6
3.6
3.5
3.1
19.9
15.5
12.8
21.0
16.1
4
5
8
8
13
13
4
4
4
4
4
4
8
8
8
8
8
8
12
12
12
12
12
12
5
5
5
5
5
22
22
23
22
22
NA
2009 Simulation Year
ER Visits (repiratory); Atlanta (Strickland etal., 2007)
Distributed lag 0-7 days
Average day lag 0-2
5,900
3,500
270
150
490
280
700
400
ER-visits (respiratory); Atlanta (Tolbert etal., 2007, Darrowetal., 2011)
Tolbert (single pollutant
Tolbert-CO
Tolbert-N02
Tolbert-PMlO
Tolbert-PMlO, N02
Darrow (single pollutant
6,400
5,700
5,200
4,100
3,900
3,500
ER-visits (asthma); NYC (Ito et al, 2007)
single pollutant model
PM2.5
N02
CO
S02
10,000
8,100
6,700
11,000
8,300
230
200
180
140
140
120
440
390
350
270
260
230
620
550
500
390
380
330
470
360
290
500
370
2,100
1,600
1,300
2,200
1,700
NA
17.2
10.1
5.1
4.5
4.1
3.2
3.1
2.8
19.3
15.1
12.4
20.4
15.5
4
4
7
7
10
10
3
3
3
3
3
3
6
6
6
6
6
6
9
9
9
9
9
9
4
4
4
4
4
17
17
18
17
17
NA
NA: for NY, the model-based adjustment methodology was unable to adjust Os distributions such that they would meet the lower
alternative standard level of 60 ppb.
7-59
-------
Table 7-11. Short-Term Os-attributable Morbidity Counts, Percent of Baseline and
Reduction in Os-attributable Risk, Asthma Exacerbations, 2007 and 2009 Air Quality.
Endpoint/Study Area/Descriptor
Air Quality Scenario
Total Ozone-
Attributable
Counts
75ppb
Change in Ozone-Attributable
Counts
75-70
75-65
75-60
Percent of
Baseline
75ppb
% Change in Ozone-Related
Risk
75-70
75-65
75-60
2007 Simulation Year
Asthma exacerbation (wheeze); Boston (Gentetal., 2003, 2004)
Chest Tightness (Ihr max)
Chest Tightness (8hr max)
Chest Tightness (Ihr max, PM2.5)a
Chest Tightness (Ihr max, PM2.5)b
Shortness of Breath (Ihr max)
Shortness of Breath (8hr max)
Wheeze (PM2.5)
40,000
30,000
41,000
38,000
29,000
35,000
76,000
1,200
680
1,200
1,100
800
780
2,200
3,300
1,900
3,300
3,000
2,200
2,100
6,000
5,100
3,000
5,100
4,700
3,400
3,400
9,300
28.9
21.2
29.1
26.9
16.3
19.6
23.3
2
2
2
2
2
2
2
5
5
5
5
6
5
6
9
8
9
9
10
8
9
2009 Simulation Year
Asthma exacerbation (wheeze); Boston (Gentetal., 2003, 2004)
Chest Tightness (Ihr max)
Chest Tightness (8hr max)
Chest Tightness (Ihr max, PM2.5)a
Chest Tightness (Ihr max, PM2.5)b
Shortness of Breath (Ihr max)
Shortness of Breath (8hr max)
Wheeze (PM2.5)
38,000
28,000
38,000
35,000
27,000
32,000
71,000
290
-110
300
270
190
-120
530
1,400
470
1,400
1,300
930
540
2,600
2,800
1,300
2,900
2,600
1,900
1,500
5,200
27.0
19.8
27.2
25.1
15.1
18.3
21.7
0.4
-0.4
0.4
0.4
1
-0.4
0.5
2
1
2
3
3
1
3
5
3
5
5
6
4
6
' previous day; b same day
7-60
-------
2007 Simulation year
Trend in ozone-related HA across standard levels
(HA per 100,000)
-Atlanta, GA
-Baltimore, MD
Boston, MA
-Cleveland, OH
-Denver, CO
Detroit, Ml
-Houston, TX
-Los Angeles, CA
-New York, NY
-Philadelphia, PA
-Sacramento, CA
St. Louis, MO
75ppb
70ppb
65ppb
60ppb
2009 Simulation year
Trend in ozone-related HA across standard levels
(HA per 100,000)
• Atlanta, GA
• Baltimore, MD
A Boston, MA
-*- Cleveland, OH
-)*- Denver, CO
—•-Detroit, Ml
^^~ Houston, TX
^^—Los Angeles, CA
— New York, NY
-•-Philadelphia, PA
^^HSacramento, CA
4 St. Louis, MO
75ppb
70ppb
65ppb
60ppb
Figure 7-5. Short-Term Os-attributable Respiratory Hospital Admissions, 2007 (top panel)
and 2009 (bottom panel) Air Quality Adjusted to Meet the Existing and Alternative
Standards. Medina-Ramon, et al. (2006) C-R functions.
7-61
-------
Table 7-12. Long-Term O3-attributable Respiratory Mortality, 2007 and 2009 Air Quality.
Jerrett et al. (2009) C-R functions, CBSA-based study area, no threshold.
Study Area
Air Qualtiy Scenario
Absolute Ozone
Atrrlbutable
Mortality
75ppb
Change In Ozone-Attributable Mortality
75-70 75-65 75-60
2007 Simulation Year
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
NewYork, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
590
(210-920)
390
(140-610)
640
(230-1000)
330
(120-510)
330
( 120 - 500)
600
(220-940)
460
( 160 - 720)
1,500
(560-2400)
2,100
(750-3300)
930
(330-1400)
300
(110-470)
480
(170-750)
35
(12-59)
17
(6-29)
20
(7-33)
16
(6-27)
13
(4-21)
28
( 10 - 46)
8.0
(3-13)
82
(28-140)
140
(47-230)
42
(14-69)
14
(5-22)
27
(9-45)
64
(22-110)
35
(12-57)
53
(18-88)
35
(12-58)
26
(9-44)
50
(17-82)
16
(5-26)
160
(54-260)
550
(190-900)
87
(30-140)
26
(9-43)
56
(19-92)
100
(34-160)
57
(19-93)
82
(28-140)
64
(22-100)
43
(15-71)
78
(27-130)
27
(9-44)
240
(83-400)
NA
130
(44-210)
44
(15-73)
84
(29-140)
2009 Simulation Year
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
NewYork, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
550
(200-860)
360
( 130 - 560)
580
(210-920)
300
(110-470)
320
(120-490)
540
(190-850)
490
(180-770)
1,600
(570-2400)
2,000
(730-3200)
850
(310-1300)
310
( 110 - 480)
440
( 160 - 690)
32
(11-53)
12
(4-20)
3.7
(1-6)
14
(5-24)
5.8
(2-10)
-6.7
(-2 --11)
11
(4-18)
77
(26-130)
120
(40-200)
31
(11-52)
14
(5-24)
19
(6-31)
59
(20-98)
27
(9-44)
23
(8-38)
32
(11-53)
18
(6-30)
14
(5-23)
24
(8-40)
160
(54-260)
420
(140-690)
66
(23-110)
28
(9-46)
41
(14-67)
82
(28-140)
41
(14-68)
47
(16-77)
50
(17-82)
45
(16-75)
38
(13-64)
40
(14-66)
250
(84-400)
NA
97
(33-160)
44
(15-73)
66
(23-110)
NA: for NY, the model-based adjustment methodology was unable to adjust Os distributions such that they would meet the lower
alternative standard level of 60 ppb.
7-62
-------
Table 7-13. Long-Term Os-attributable Respiratory Mortality Percent of Baseline
Incidence and Percent Reduction in Os-attributable Risk, 2007 and 2009 Air Quality.
Jerrett et al. (2009) C-R functions, CBSA-based study area, no threshold.
Study Area
Air Quality Scenario
% of Baseline
Incidence
75ppb
% Change in O3-Attributable
Risk
75-70
75-65
75-60
2007 Simulation Year
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
18.6
18.8
17.2
17.7
20.8
18.4
16.3
20.4
16.9
18.4
17.8
18.8
5
4
3
4
3
4
1
4
6
4
4
5
9
7
7
9
6
7
3
8
23
8
7
10
14
12
11
16
11
11
5
13
NA
12
12
15
2009 Simulation Year
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
17.0
17.4
16.0
16.8
20.0
17.0
16.9
20.7
16.7
17.2
18.0
17.7
5
3
1
4
1
-1
2
4
5
3
4
3
9
6
3
9
5
2
4
8
18
6
7
8
13
10
7
14
12
6
7
13
NA
10
12
12
NA: for NY, the model-based adjustment methodology was unable to adjust Os distributions such that they would meet the lower
alternative standard level of 60 ppb.
7-63
-------
2007 Simulation Year
Trend in ozone-related mortality across standard
levels (deaths per 100,000)
-Atlanta, GA
-Baltimore, MD
Boston, MA
-Cleveland, OH
-Denver, CO
Detroit, Ml
-Houston,TX
-Los Angeles, CA
New York, NY
-Philadelphia, PA
Sacramento, CA
St. Louis, MO
75ppb
70ppb
65ppb
60ppb
2009 Simulation Year
Trend in ozone-related mortality across standard
levels (deaths per 100,000)
5 20
I
ra 15
-Atlanta, GA
-Baltimore, MD
Boston, MA
-Cleveland, OH
-Denver, CO
Detroit, Ml
-Houston, TX
•Los Angeles, CA
New York, NY
-Philadelphia, PA
•Sacramento, CA
St. Louis, MO
75ppb
70ppb
65ppb
60ppb
Figure 7-6. Long-Term Os-attributable Respiratory Mortality for 2007 and 2009 Air
Quality Adjusted to Just Meet the Existing and Alternative Standards, Jerrett et al. (2009)
C-R functions.
7-64
-------
The presentation of key observations drawn from review of the core risk estimates is
divided into two sections: (1) the assessment of health risks associated with just meeting the
existing standard (section 7.5.1) and (2) the assessment of risk changes from meeting alternative
standards relative to meeting the existing standard (section 7.5.2). The presentation of key
observations in each of these two sections is further separated into those associated with (a)
short-term Os-attributable mortality, (b) short-term Os-attributable morbidity and (c) long-term
Os-attributable mortality. Unless otherwise noted, all risk estimates discussed in these three
sections are core risk estimates. In some cases we refer to the confidence intervals around risk
estimates. When an effect estimate is drawn from a study with low statistical power, confidence
intervals can be wide, and can include negative values because of the assumptions of normality
in the distribution of the effect estimate. Negative lower-confidence bounds do not imply that
additional exposure to Os has a beneficial effect, but rather that the estimated Os effect estimate
in the C-R function was not statistically significantly different from zero, and thus has a higher
degree of uncertainty as to the magnitude of the estimated risk. As noted earlier, presentation of
sensitivity analysis results and their use in interpreting the core risk estimates is covered in
section 7.5.3.
7.5.1 Assessment of Health Risk after Adjusting Air Quality to Just Meet the Existing
Standard
The analysis of risk after simulating just meeting the existing standard focuses on
absolute risk, since this is of greatest relevance in evaluating the adequacy of the existing
standard.
Short-term Ch-attributable mortality
• After adjusting air quality to just meet the existing standard, estimates of Os-related
all-cause mortality range across urban study areas from 58 to 3,200 deaths (for
simulation year 2007) and from 56 to 3,000 deaths (for simulation year 2009) (see
Table 7-7). This translates into from 0.8 to 4.1% of baseline all-cause mortality (for
simulation year 2007) and from 0.8 to 4.0% (for simulation year 2009) (see Table
7-8) in these study areas. Generally, Os-attributable all-cause mortality risks continue
to be lower for the 2009 simulation year as compared with the 2007 simulation year
(with the exception of Houston), reflecting the generally lower ambient Os levels
associated with 2009 for most of the study areas (see Table 7-7 and Table 7-8).
• Confidence intervals (CIs) reflecting the statistical fit of the effect estimates used in
modeling risk demonstrate substantial variability across the 12 urban study areas. In
7-65
-------
general, the upper 95th percentile CI tends to be from 2-3 times larger than the point
estimate for the 12 urban study areas (see Table 7-7). However, some study areas
have wider confidence intervals (e.g., for Denver the upper CI is about 5 times the
point estimate), while others have tighter relative CIs (e.g., New York, where the
upper CI is -1.4 times larger than the point estimate). This variation in the CIs
associated with risk estimates can reflect a number of factors including the statistical
power of the underlying epidemiological study, which is based on the population size,
and differences in the magnitude of such factors as exposure measurement error and
correlations between Os and other pollutants.
After just meeting the existing Os standard for simulation year 2007, all-cause
mortality estimates based on C-R functions from Smith et al. (2009) continue to be
driven largely by days with total Os levels within 30 to 70 ppb, with 87 to 99% of the
mortality estimate across the 12 urban study areas associated with days in this range.
A smaller fraction (9 to 24%) of the mortality risk is associated with days above 60
ppb, for most of the study areas (see Figure 7-2, "Existing standard (75)" plot).35 For
2009, this pattern continues although risk distributions are shifted downward,
reflecting the lower ambient Os levels generally seen in this simulation year
compared with 2007(see Figure 7-3, "current standard (75)" plot). For 2009, 9% to
33% of Os-attributable mortality risk is associated with days having Os measurements
55-60 ppb or higher. Approximately 0% to 2% of total mortality estimates for the
existing standard air quality scenario are associated with days having ambient Cb
levels of 20 ppb or less.36
• Estimates of Os-attributable respiratory-related HA range from tens to hundreds of
cases (after adjusting air quality to just meet the existing standard) depending on the
type of respiratory HA endpoint modeled and the specific urban study areas evaluated
35 Houston, LA, NY, and Sacramento study areas have a significantly smaller percentage (<1%) of its estimated
mortality associated with days above 60ppb.
36 In the first draft Os HREA, we included consideration for surrogate LMLs (based on the lowest composite monitor
values used in modeling short-term exposure-related mortality for each urban study area - see Table 7-5, First
Draft HREA, U.S. EPA, 2012). For the 8-hr max monitoring season LML (applicable to the Smith et al., 2009-
based core risk estimate generated for this HREA), we have values ranging from 4 to 17 ppb and from 5 to 16 ppb
(across the 12 urban study areas for 2007 and 2009, respectively). If we look at heat maps characterizing the
distribution of short-term exposure-related mortality for recent conditions (see "recent conditions" heat maps in
Figures 7B-1 and 7B-2 in Appendix 7B) we see that the majority of O3-related mortality falls above these
surrogate LML ranges. Consequently we see that had we integrated consideration for the surrogate LMLs into
modeling of short-term exposure-related mortality, there would have been small changes in estimated risk.
7-66
-------
(see Table 7-9). All 12 urban study areas were modeled for one of more respiratory-
related HA endpoints.
• Os-attributable ER (for respiratory symptoms) ranged into the thousands for both
New York and Atlanta when considering simulated air quality that just meets the
existing standard (these were the only two study areas modeled for this health
endpoint) (see Table 7-10).
• Estimates of Os-attributable asthma exacerbation (wheeze) in Boston are in the tens
of thousands to over 100,000 (see Table 7-11). The percent of baseline for this health
endpoint after adjusting air quality to just meet the existing standard ranges from
about 20 to 30%, a risk estimate higher than that estimated for the other short-term
morbidity endpoints evaluated here (see Table 7-11 and compare to values in 7-9 and
7-10).
Long-term Ch-attributable mortality
• After simulating air quality that just meet the existing standard, estimates of Os-
related respiratory mortality ranges across the urban study areas from 300 to 2,100
deaths (for 2007) and from 300 to 2,000 deaths (for 2009) (see Table 7-12). This
translates into from 16.3 to 20.8% of baseline across the 12 urban study areas (for
2007) and from 16 to 20.7% (for 2009) using the Jerrett et al. (2009) C-R national-
scale function applied to each urban area (see Table 7-13). As discussed in section
7.3.2, because of the long-term exposure metric (seasonal mean of daily maximum 8-
hr Os concentrations) employed in risk modeling, there is the potential for overlap
between short-term and long-term exposure-related mortality estimates. For that
reason, these two categories of mortality estimates cannot be considered distinct and
should not be added to estimate total mortality.
95th percentile CIs for long-term Os-attributable respiratory mortality suggest greater
power (and potential less heterogeneity) associated with modeling this health endpoint,
compared with short-term Os-attributable mortality. None of the CIs for long-term Os-
attributable mortality include negative estimates as lower bounds (see Table 7-12). 7.5.2
7.5.2 Assessment of Health Risk after Adjusting Air Quality to Just Meet Alternative
Standards
As discussed earlier, we have considered three alternative standard levels (70, 65 and 60
ppb), each evaluated using the same form and averaging time of the existing standard. In
7-67
-------
presenting risk estimates associated with air quality adjusted to just meet each of these
alternative standard levels, we focus on the change in risk associated with a comparison of Os
levels after simulation of the existing standard with levels after simulation of each of the
alternative standard levels. This is of greatest relevance in comparing the potential public health
benefit associated with each of the alternative standards relative to the level of protection
afforded by just meeting the existing standard.
In reviewing these risk estimates, it is important to keep in mind that simulation of
alternative standard levels is based on a reaching the 4th highest daily maximum 8-hr average, or
peak-based metric. Based on the simulated air quality information for the 12 urban study areas,
there is a tendency for modeled Os to increase on lower concentration days and decrease on
higher concentration days.37 Therefore, it is not immediately clear that we would expect risk
reductions with increasing stringency in peak-based standard levels when applying C-R
functions that are based on the full distribution of daily maximum 8-hr values. Specifically, risk
reductions are only expected to the extent that the composite monitor daily maximum 8-hr
concentrations decrease as lower alternative standards are simulated. As discussed in Chapter 4,
after adjustment to alternative standard levels, decreases in Os typically occur on higher Os days
which tend to occur during warmer (summer) months and are concentrated in suburban areas.
Conversely, increases in Os, typically occur lower Os days which tend to occur in the cooler
portions of the year and are focused in core urban areas. In general, variability in predicted daily
Os concentrations decreases when meeting lower standard levels.
Short-term Ch-attributable mortality
• In our analysis, the mortality risk metric is moderately responsive to adjusted air
quality that just meets the existing and alternative standard levels. This is related to
(1) how Os concentrations respond to reductions in NOx emissions used to meet the
standards, and (2) how the risk metrics are associated with temporal and spatial
patterns of Os. As discussed in section 7.1.1, mortality risk is modeled using
composite monitor values (i.e., averages of Os measurements across monitors in an
urban study area) which removes spatial variability in measured Os within an urban
study area (also removing variability in changes in Os across an urban area resulting
from NOx reductions). Furthermore, in modeling total mortality risk for the core
analysis, we add the risk changes occurring across all days within the monitored Os
' This relationship is also observed in ambient air quality measurements as discussed in Chapter 4 and appendices.
7-68
-------
season, including days with low values of Os as well as days with high values of Os.
This means that we include both decreases in risk on those days when Os is estimated
to decrease (generally occurring on days with higher values of Os) and increases in
risk when Cb is simulated to increase (generally associated with lower values of Os).
The dampened response of short-term mortality risk can be contrasted with clinical
study-based risk estimates. The clinical study-based estimates primarily reflect
changes in the upper end of the Os distribution where we tend to see more consistent,
progressive reductions in exposures of concern and health risk coincide with
progressive increases in the stringency of alternative standard levels. In addition,
clinical-based estimates of risk are based on detailed micro-environmental exposure
modeling which uses individual monitor values instead of composite monitor values,
thereby resulting is less dampening of spatial variability in Cb within a given urban
study area.
Generally, the magnitude of risk reduction increases as lower alternative standard
levels are simulated. For example, for the lowest alternative standard we evaluated,
60 ppb, across the 12 urban study areas, we predict from 5 to 96 fewer Os-attributable
deaths for simulation year 2007 (relative to estimated risk associated with air quality
that just meets the existing standard) (see Table 7-7). This range is from 5 to 98
deaths for simulation year 2009. These ranges (for the 60 ppb standard level) translate
into a 2 to 14% reduction in Os-attributable risk relative to estimated risk associated
with air quality that just meets the existing standard (see Table 7-8).
As noted in section 7.1.1, some of the urban study areas are projected to experience
increases in Os (and hence risk) when adjusting air quality that just meets the existing
standard and some of the alternative standard levels. Focusing specifically on the
alternative standard levels, we see that, for the core analysis, this potential increase in
risk only occurs for the 2009 simulation year and specifically for three of the urban
study areas (Boston, Detroit and Houston - see Table 7-7). For example, Detroit is
estimated to have an increase of 17 Os-attributable deaths after meeting the 70 ppb
standard (when compared to risk remaining after meeting the existing standard).
However, we estimate a net reduction of 12 Os-attributable deaths in this study area
after meeting the 60 ppb level (again based on comparison to risk after meeting the
existing standard). Furthermore, for all three urban study areas with initial risk
increases (based on comparing meeting the existing standard to meeting alternative
standards), we see that these increases are offset after meeting the lowest alternative
7-69
-------
standard simulated (60 ppb) (see Table 7-7). The potential for risk increases is
increased somewhat for several of the urban study areas when we simulate how the
Os distribution shifts from recent conditions to just meeting the existing standard (see
Appendix 7B, Tables 7B-1 and 7B-2). Specifically, in simulating estimated risk from
moving from recent conditions to just meeting the existing standard, we see that for
the 2007 simulation year, two of the study areas (Houston and Los Angeles) have risk
increases after meeting the existing standard compared to recent ambient conditions
while half of the twelve urban study areas have risk increases for the 2009 simulation
year in adjusting air quality to meet the existing standard relative to recent conditions.
It is also important to keep in mind that, for the urban study areas of New York and
Los Angeles, there are additional uncertainties in the simulation of existing and
alternative standards given the limitations in the application of the adjustment
methodology to very large emissions reductions and that the 95th percent confidence
interval lower bound estimate of hourly Os concentrations was used to capture a
scenario in which these urban study areas could meet lower standard levels (65 ppb
for New York and 60 ppb for Los Angeles). In five of these eight cases, the initial
risk increases (including the increase in going from recent ambient conditions to air
quality just meeting the existing standard) is fully offset after meeting the lowest
alternative standard level (60 ppb).38
• Figure 7-4 provides plots of short-term mortality risk associated with air quality that
just meets the existing and alternative standards adjusting for total exposed
population (i.e., Os-attributable deaths per 100,000 exposed). Total Os-attributable
risk, even when adjusted for population, varies substantially across the 12 urban study
areas, with New York and Philadelphia having the highest estimated risk and Atlanta
and Denver the lowest. This spread in risk (adjusted for population) reflects, to a great
extent, differences in the effect estimates used in modeling this endpoint for each
study area, which can in turn reflect a number of factors (e.g., differences in behavior
such as outdoor activity across cities and differences in exposure measurement error).
However, despite considerable variability in absolute Os-attributable risk, Figure 7-4
also suggests that most of the study areas display moderate reductions in Cb-
38 For both LA and Houston (in 2007) and Houston (2009) a modest net risk increase still persists (compared to risk
under recent conditions), even when we have simulated the lowest alternative standard considered (60 ppb) (see
Tables 7B-1 and 7B-2).
7-70
-------
attributable risk across the three alternative standards (with the exception of New
York, which has a notable decrease in risk for the 70 to 65 ppb standard level).39 This
suggests that a substantial fraction of Os-attributable risk would still remain, even
after simulating air quality that just meets the lowest alternative standard considered.
• Heat map plots of risk reductions for 2007 suggest that most of the risk reductions
associated with simulation of all three alternative standards occur on days with
composite Os level between 35 and 60 ppb (see Figure 7-4). By contrast, most of the
risk increases occur on days with composite Os levels between 20 and 35ppb (see
Figure 7-4). This is expected given that most of the increases in urban core Os are
associated with lower Os days where NOX titration is prevalent (see Appendix 4D,
section 4.6, Figures 40-54). Very little of the projected change in risk (increases, or
decreases) for any of the alternative standards considered occurred on days with Os
levels below 20 ppb Os. Similar observations hold for risk results generated for
simulation year 2009.
Short-term Ch-attributable morbidity
• Generally, because the short-term Os exposure-related morbidity endpoints use the
same air metrics as used in modeling short-term Os-attributable mortality (i.e., 8-hr
maximum and 8-hr mean) the pattern of risk reduction seen for these morbidity
endpoints are similar to those seen with short-term mortality (see Table 7-9 through
Table 7-11 and Figure 7-5). However, New York, as mentioned with regard to short-
term Os-attributable mortality, has substantially higher percent reductions (for Os-
attributable risk) compared with the other study areas. For example, with ER visits
(asthma), under the 65 ppb alternative standard in simulation year 2007, New York is
estimated to have a 22 to 23% reduction in the number of ER-visits associated with
Os exposure (see Table 7-10).
Long-term Ch-attributable mortality
Although long-term Os-attributable mortality is modeled using a different Os metric
(essentially a long-term trend in the 1-hr maximum for the hottest two seasons - see section
7.3.2) the overall magnitude and pattern of reduction in Os-related risk is similar to that seen
39 With the New York urban study area, we recognize however that there is significant uncertainty associated with
the use of the CBSA-based study area due to significant heterogeneity in short-term O3-attributable mortality
effect estimates (from Smith et al, 2009) within that larger urban study area (see discussion in section 7.6.1).
7-71
-------
with short-term exposure related mortality. Specifically, for the 2007 simulation year, for most
urban study areas risk reductions range from 11 to 16% (for the 60 ppb standard) (see Table
7-13). Risk reductions are generally slightly smaller across alternative standard levels for
simulation year 2009. For the 2009 simulation year, for Detroit, we see a relatively small risk
increase for the 70 ppb alternative standard (compared to risk under the existing standard).
However that initial increase is offset by risk reductions for the other (lower) alternative standard
levels simulated (see Table 7-13).
7.5.3 Sensitivity Analyses Designed to Enhance Understanding of the Core Risk Estimates
We have completed a number of sensitivity analyses intended to support interpretation of
the core risk estimates. These sensitivity analyses, which are described in section 7.4.3, can be
divided into two categories: (a) sensitivity analyses exploring factors impacting air quality
characterization (specifically composite monitor composition) and (b) sensitivity analyses
exploring the impact of alternative C-R function specifications. As noted in section 7.4.3, we
also completed an initial influence analysis designed to identify which of the input factors to the
risk model (for short-term exposure-related mortality) are primarily responsible for inter-city
variability in that risk metric. This section summarizes the results of these sensitivity analyses
and presents key observations related to those analyses, beginning with the influence analysis
and then proceeding to sensitivity analyses focused on air quality characterization and alternative
C-R function specification.
Influence analysis
The influence analysis considered three factors involved in modeling risk for the short-
term exposure-related mortality endpoint including: baseline incidence, composite monitor Os
levels and Bayes-adjusted city-specific effect estimates (recall that the core risk estimate is based
on effect estimates derived as part of analyses published in Smith et al., 2009). Each of these
input factors displays inter-study area variation and are responsible, collectively, for
heterogeneity in risk estimates.40 In completing the analysis, we first calculated a central
tendency estimate of risk based on the mean of each input factor across the 12 urban study areas
for the 2009 simulation year (i.e., using the average of the city-specific values for each of the
input factors). We then systematically varied each of the three heterogeneity-related factors
(effect estimate, composite monitor-based Os level and baseline incidence) to one standard
40 Note, that the demographic count input factor also varies across the study areas and is an important factor in
determining total O3-attributable mortality. However, for the influence analysis, we used deaths per 100,000 as the
risk metric which standardizes on demographic count and therefore allowed us to exclude this input parameter in
conducting the influence analysis.
7-72
-------
deviation (SD) above its mean value (reflecting variance across the 12 urban study area values)
and noted the percent increase produced by that perturbation over the initial mean risk estimate.
This influence analysis allowed us to explore the impact of both model form - specifically,
potential non-linearities in the model - as well as the relative magnitude of variability in each of
the three heterogeneity-related input factors on risk. The influence analysis generated the
following results: baseline incidence (23%), composite monitor-based Os level (10%), and effect
estimate (54%). In other words, the 54% result for effect estimate means that use of a value 1 SD
over the mean (for the effect estimate) in generating risk, resulted in a risk estimate that was 54%
larger than the risk estimate based on the mean of all input factors. These results clearly show
that, of the three input factors considered, the effect estimate is primarily responsible for inter-
city variability in short-term exposure-related risk.
Interestingly, when we look at the coefficient of variation (CV) for these three
heterogeneity-related input factors we see values almost identical to the influence analysis results
in terms of relative magnitude to each other (i.e., 0.230, 0.085, and 0.507 for baseline incidence,
composite monitor-based Os levels and effect estimate, respectively). Given that the CV values
only reflect variability in each input factor and not model form (i.e., do not reflect potential non-
linearities in the model), the fact that the CV values almost exactly match the influence analysis
results in terms of relative magnitude suggest that there is very little if any non-linearity in the
model calculations involving these three input factors. Had non-linearity existed to a significant
extent, then the influence analysis results would have differed substantially from the CV results.
The fact that both analyses suggest a primary role for the effect estimate in driving inter-city
variability in risk emphasizes the importance of the sensitivity analyses exploring alternative C-R
functions specifications that were completed for the HREA (see below).
Air quality-related sensitivity analyses
This category of sensitivity analysis covers (a) the use of a smaller study area (the Smith
et al., 2009 study areas) as contrasted with the CBSA-based study areas used in the core analysis,
and (b) the use of alternative approaches to simulate air quality that just meets the existing and
alternative standards (for a subset of the study areas) (see section 7.4.3 for additional detail). This
category of sensitivity analysis was applied to short-term Os-attributable mortality given the
importance of the endpoint in the policy-context.41
41 Observations regarding the sensitivity of core short-term O3-attributable mortality risk to these sensitivity analyses
can be applied with care to the core short-term O3-attributable morbidity endpoints, since many of these used
similar air quality metrics in modeling risk.
7-73
-------
To allow for easier visual comparisons, we have presented the results of this sensitivity
analysis category in graphical form (see Figure 7-7, numerical results are presented in Appendix
7C). This figure presents point estimates and 95th percentile confidence ranges for the core model
and for two sensitivity analyses: (a) SA1 (use of the smaller Smith et al., 2009 based study area)
and (b) SA2 (use of the alternative approach to adjusting air quality). SA2 is not presented for all
of the study areas, only for the subset included in these alternative simulations (see section
7.4.3). The sensitivity analyses results presented in Figure 7-7 are the changes in Os-related risk
that result from meeting the three alternative standards relative to meeting the existing standard.
Furthermore, these changes reflect deaths per 100,000, which standardizes the estimates on
population. This removes variation in the size of the underlying exposed population as a factor to
consider in interpreting these results.
For the sensitivity analysis examining use of the smaller Smith et al. (2009) study area,
we have also included heat maps similar to those used in conveying core estimates for short-term
exposure related mortality (see section 7.5 for a description of the heat maps used in the core
analysis). These heat maps (included in Appendix 7C - see Figure 7C-1) allow us to consider
how changes in risk, including both reductions in risk and increases in risk are distributed across
the Os air quality distributions for each study area.
Key observations related to the air quality-related sensitivity analyses include:
• Use of smaller study area reduces magnitude of risk reduction: For most of the
study areas, use of the smaller Smith et al. (2009)-based study area resulted in smaller
risk reductions (again expressed in terms of changes in deaths per 100,000). For
example, in Figure 7-7 (Baltimore plot), we see that estimated change in risk for SA1
(the smaller study area) are lower than estimated change in risk for the core scenario.
This likely reflects the mix of monitors in the smaller study areas which results in a
smaller change in the composite monitor value (for the existing standard versus
alternative standard levels) as compared with composite monitor values based on the
larger CBSA study area. However, it is important to keep the relatively small
magnitude of these risk reductions in mind when considering the results of the
sensitivity analysis. Most of these differences in risk reductions are less than 1
individual per 100,000 which reflects the fact that total risk reduction (for short-term
Os-attributable mortality) across the urban study areas is relatively small (see Table
7-7).
• Reductions in risk are focused on higher Os days while increases are focused on
lower Os days: Figure 7C-1 allows us to consider patterns in risk reductions and
increases when using the smaller Smith et al., 2009-based study areas in modeling
7-74
-------
risk. Figure 7C-1 (particularly the plots of risk decreases) suggests that decreases in
risk tend to occur on days with composite monitor Os concentrations ranging from
40-70ppb, while increases in risk tend to occur on days with composite monitor
values in the range at or below 30-40 ppb (with most risk increases falling in the
range of 15ppb to 40ppb). As noted in 7.1.1, there is less confidence in specifying the
nature of the C-R function (and therefore less confidence in specifying risk) in the
range below 20 ppb.
Application of effect estimates derived for smaller study areas to larger CBSA-
based study areas: As noted in section 7.3.2, in those instances where an
epidemiological study provides effect estimates for multiple subareas within a larger
CBSA-based study area, we are selecting the effect estimates that represent the
largest number of individuals to model that CBSA-based study areas. There is
uncertainty associated with this approach. Specifically, as illustrated in Table 7-3,
effect estimates within some of the CBSA-based study areas can display considerable
heterogeneity. For example, consider the Smith et al. (2009)-based effect estimates
that fall within the CBSA-based New York study areas (these vary from 0.0001 to
0.0009 - almost a 10 fold factor, see Table 7-3). Furthermore, with the CBSA-based
New York study area, Smith et al. (2009)-based effect estimates only cover about half
of the total population, with 8.3 million residents living within portions of the CBSA
not covered by Smith et al. (2009)-based effect estimates. As noted in section 7.3.2,
in these types of situations, we have decided to use the single effect estimates
representing the largest number of residents in modeling the larger CBSA-study area.
This reflects the observation that, in the case of the New York CBSA, one of the
available effect estimates (for the New York study area), represents ~7 times the
population of the other effect estimates (see Table 7-3). In the case of the Los
Angeles CBSA, there is significantly less difference between the available effect
estimates, making the issue of heterogeneity (and the specification of a single effect
estimate for this study area) less important. Never the less, we recognize that the issue
of heterogeneity does complicate extrapolation of effect estimates for smaller study
areas to the larger CBSA study areas modeled in this analysis and does introduce a
degree of uncertainty that is difficult to characterize.
Use of alternative approach for simulating air quality that just meets existing
and alternative standard levels: Use of an alternative approach to simulate air
quality that just meets the existing and alternative standard levels did not produce a
consistent trend in terms of changes in risk between existing and alternative standards
relative to the core analysis. For example, if we look at Figure 7-7 (plot for Houston),
7-75
-------
we see that SA2 (reflecting application of the alternative simulation approach) has a
larger risk reduction than the core estimate. By contrast, if we look at the plot for Los
Angeles, we see that the SA2 risk change is lower than the core estimate. Again, as
with the sensitivity analysis results looking at study area size, it is important to keep
in mind that the magnitude of these differences is relatively small, reflecting the small
magnitude of mortality risk associated with these analyses in general (see Table 7-7).
It is also important to note that in the alternative simulation approach, the HDDM-
adjustment approach assumed the same percent reductions of NOX and VOC and did
not examine if a different air quality distribution could have been obtained with a
different combination of NOx versus VOC reductions. For most of the urban areas,
the percent NOx and VOC reductions were very similar to the NOx-only percent
reductions. The similarity in the NOx reductions between the two approaches could be
the reason for there being little difference in the risk estimates between the core and
the alternative approach.
Sensitivity analyses related to specification of C-R functions
This category of sensitivity analysis covers a number of factors related to the
specification of C-R functions for both short-term Os-attributable and long-term Os-attributable
mortality. In the case of short-term Os-attributable mortality, we consider (a) the use of Bayes
adjusted effect estimates using regional priors (as contrasted with the Bayes adjusted values
using a national prior applied in the core analysis), (b) the use of a co-pollutants model
considering PMio (as contrasted with the single pollutant model used in the core analysis) and (c)
application of effect estimates from Zanobetti and Schwartz (2008) reflecting a summer focused
analysis (as contrasted with the Smith et al., 2009-based analysis reflecting the entire monitoring
period in each study area, which is used in the core analysis).
For long-term Os-attributable mortality, we consider the use of regionally-differentiated
single pollutant effect estimates obtained from Jerrett et al. (2009), as contrasted with the single
national co-pollutants model used in the core analysis (see section 7.1.1). In addition, we present
estimates for long-term Os attributable mortality based on application of results from a national
level single pollutant model. We have also included a simulation of the impact of potential
thresholds in the relationship between long-term Os exposure and respiratory-related mortality.
Specifically, we have modeled risk using alternative effect estimates reflecting the application of
a range of potential thresholds (40, 45, 50, 55, 56 and 60 ppb) included in Jerrett et al. (2009)
(see Sasser, 2014) and compared those estimates with estimates based on non-thresholds models
included as core estimates (see section 7.3.2).
7-76
-------
For sensitivity analyses examining alternative specification of the C-R function for short-
term Os-attributable mortality, we have used the same graphical approach as used in presenting
results of the sensitivity analyses examining air quality characterization (i.e., plots of point
estimates with 95th percentile C.I.s for the core and sensitivity analyses for each of the study
areas - see Figure 7-8). Here we also plot estimates of risk changes using deaths per 100,000 to
standardize in terms of total exposed population.
For the sensitivity analysis considering alternative C-R functions for long-term Os-
attributable mortality, we present results in both tabular and graphical formats. Results of
sensitivity analyses exploring the impact of regional-differentiated effect estimates and a national
Os-only effect estimate are presented in Table 7-14 and Table 7-15, respectively. Each table
includes both (a) the percent of baseline mortality attributable to Os (considering air quality
adjusted to just meet the existing standard) and (b) the percent reduction in Os-attributable risk
for each of the alternative standard levels for both the core and sensitivity analysis scenarios.
Results of the sensitivity analysis exploring the impact of potential thresholds in long-term Os-
attributable mortality are presented in tabular form in Figure 7-9. In that figure, we include
separate plots for each of the air quality scenarios with each plot including estimates of Os-
attributable mortality (as deaths per 100,000) for non-threshold and a range of threshold models.
Key observations related to sensitivity analyses examining alternative C-R functions
specifications for both short-term and long-term exposure-related mortality include:
• Use of regional Bayes-adjusted effect estimates in modeling short-term Os-
attributable mortality: The use of Bayes-adjusted effect estimates with regional
priors in modeling short-term Cb-attributable mortality, had a mixed impact across the
urban study areas, with some study areas having increased changes in risk and others
having smaller changes, relative to the core analysis. For example, in Figure 7-8 (plot
for Baltimore), SA1 had a larger change in risk compared with the core analysis.
However, as with the sensitivity analyses examining air quality-related factors
(discussed above), it is important to keep in mind that the overall magnitude of the
Os-attributable mortality risk is relatively small and that these differences in changes
in risk (comparing SA1 to the core analysis) are generally in the fraction of a person
per 100,000 exposed population.
• Use of a co-pollutants model (with PMio) in modeling short-term Os-attributable
mortality: The use of the PMio co-pollutant model in modeling short-term Os-
attributable mortality (as contrasted with the single pollutant model used in the core
analysis) tended to have a relatively small effect on estimates of risk changes for the
alternative standards considered. For example, in Figure 7-8 (plot for Boston), we see
that the estimates of risk changes for SA2 (reflecting application of the PMio co-
7-77
-------
pollutant model) is essentially the same as the core risk estimate. It is important to
keep in mind that the PMio co-pollutant model suffers from significantly reduced
power due to the 1/3 to 1/6 day sampling frequency used in measuring PMio (this
reduces the number of observations available to support epidemiological analysis).
This has the impact of greatly increasing the confidence intervals on the SA2 risk
estimates relative to the core estimates.
Use of Zanobetti and Schwartz (2008) effect estimates in modeling short-term
Os-attributable mortality: The use of Zanobetti and Schwartz (2008) effect
estimates (reflecting a focus on the warmer summer months) produces a mixed set of
results when compared to the core risk estimates. If we look at Figure 7-8 we see that,
for Boston, estimates of risk changes for SA3 (reflecting application of the Zanobetti
and Schwartz, 2008 effect estimates) are significantly larger than core estimates. By
contrast, SA3 estimates of risk changes for Houston are significantly smaller than the
core estimates. It is important, however to keep in mind that the Zanobetti and
Schwartz (2008) effect estimates will tend to under-estimate total risk since they only
model impacts during the summer months (while the Smith et al., 2009 effect
estimates allow us to model impacts for the entire Os monitoring season in each study
area). Note that if the Os effect were only occurring during the summer months, then
the total risk estimated using effect estimates from the two studies would be similar.
However, because the risks in many locations are smaller (using the Zanobetti and
Schwartz, 2008 based effect estimates), this suggests that a significant portion of the
Os effect occurs outside of the summer months evaluated in this study.
Use of regional-differentiated effect estimates in modeling long-term Os-
attributable mortality: Risk estimates generated using regional-specific effect
estimates for long-term Os-attributable mortality differ substantially from the core
estimates based on a single national-level effect estimate (see Table 7-14).
Furthermore, the risk estimates generated using the regional effect estimates display
considerable variability (see Table 7-14) reflecting the significant variability in the
underlying effect estimates (see Jerrett et al., 2009, Table 4). The regional effect
estimates range from 0.99 (for the Northeast) to 1.21 (for the Southwest) and include
1.00 (no Os effect for the Industrial Midwest). As noted earlier in section 7.5,
negative risk estimates should not be interpreted as suggesting that Os exposure is
beneficial. Rather, these suggest that there may be instability in the underlying
estimates or that potential confounding has not been fully addressed. Regional effect
estimates used in this analysis have considerably larger confidence intervals than the
national estimate (compare values in Jerrett et al., 2009 Table 3 with values in Table
7-78
-------
4). This suggests that the regional estimates are less stable than the national estimates
and are subject to considerably greater uncertainty. For this reason, while the results
of this sensitivity analysis point to the potential for regional heterogeneity in the long-
term Os-attributable mortality effect estimate, we do not have significant confidence
in the regionally-based risk estimates themselves given the relatively large confidence
intervals associated with those estimates.
Use of national-based single pollutant model in modeling long-term Os-
attributable mortality: Risk estimates generated using the national-level Os-only
effect estimate were significantly lower (-30%) than the core risk estimates which
utilize a co-pollutants model (which includes PIVh.s) (see Table 7-15). In this case,
control for another pollutant results in a stronger Os signal, possibly due to an
association between PIVb.s and a confounder or effect modifier associated with the
Os-related effect.
Consideration for potential thresholds in the relationship between long-term Os
exposure and respiratory-related mortality: The results of the sensitivity analysis
based on the suite of threshold-based effect estimates suggests that compared to the
estimates generating by using a linear (no-threshold) model, these models can result
in substantially lower estimates of Cb-attributable mortality across all of the standard
levels considered (see Figure 7-9 and Appendix 7C, Tables 7C-8 through 7C-12). For
example, in Figure 7-9 we see that, for where air quality was adjusted to just meet the
existing standard (i.e., the 'Current Standard 75' plot), estimates of risk across the
study areas using a model that includes a 40 ppb threshold (i.e., see 'T40' identifier
along X-axis) are 60-90% less than risks estimated using a non-threshold models (i.e.,
see 'NT86' and 'NT96' identifiers along X-axis). For most of the study areas,
estimated risk is progressively less when using models with increasing threshold
levels (i.e., 'T45' through 'T60' models), with all study areas showing no Os-
attributable risk when using a model that includes a 60 ppb threshold ('T60' in Figure
7-9). This overall pattern is also evident when estimating risks associated with air
quality adjusted to each of the lower alternative standard levels (Figure 7-9).
7-79
-------
7570
Standard levels (delta)
75*5 75-60
§ Oi •
|si
*° '
fe t(
JO-
OS .
Joo •
0,
JO •
s10
I oo .
All-mU.GA
I J
* • f
1
' i J * 3
Cleveland, OH
f f
* * T t IT
i - 3
Houston, TX
-•-1 — » *-•— L*-^
. 3 • 3 I 3
PhUddphu.PA
1
* .. * LJ i*1
; 3 5*3 ' < 3
V
1
s«J
I 10
§ ...
.
| Oi
1 ' 1
8 • "> •
J 10 .
» Oi
O 00
8 10
a 0>
| Oi
2 10
BJtmote.MD
« V
9-» It T
. . >
* 2 '3
Denver, CO
-^ 3 •• 5
Co* and StAkt'olT *im^a !*ni**!oat
Lcrt A^elti, CA
• -. li* ""I
•» * ' • 1 — \— » 1 — j— * —
: 3 -3
> i ,* i* ' " I.*. '*."
Mt MT tit
j * 3 -: 3
: O/
f
•
.
J»-
| : o
a ° '
« Oi
j!:|
f 00
JO
§ JO
8 • o
5 10
I JO
1 I0
1 i I
A 1 f •
* * • •
* . *
• 2 -2
DelroA.MI
t* * •
•
j 3 3 i 3
™
•
3 s s 3 5 s
SLLouli.MO
+ , \ i
• if
^3 -3
Cv* vid S*«»s«y«T AA*VRS VnifiTirm
Figure 7-7. Sensitivity Analysis: Effect of Air Quality Factors on Short-Term Os-attributable Mortality, 2009 Air Quality.
SAl-smaller (Smith et al., 2009-based) study area, SA2-alternative method for simulating standards.
7-80
-------
Standard levels (dclu)
75-70 7565 75*0
9 10 '
i
I )0
§ 10
§':
i io
i io
1 JO
i06!
|w
S°'
r°
gojJ
JO
I :0
"
-:
|o-,
a >°
All anu. GA
f t i+ if if If,}
1 '
* J
#w«ir*M»ii* •
Oewtand. OH
Hil
• • * . t-JL-jJ lit
' 3 3 ' ' : ' ' '
HooGten.TX
ul
.... «-*t LUJ
• • « •
'33 "^33 *
Cor* Md S«mi*wHv Aiulyvt SHltfUtont
. ~\ Ml*
. j Vt * V»
:
000' 001
-:
i 0>
1 10
««:
§"•
lot
1»
JO
g 10
5 10
e jo
1 10
S 10 •
t oo
i>»-
• LO J
Djlllmore, MD
)•
3 3 J > -• 51 S « i « 1
Drnvn. CO
• • I J
1
33 |333 \ i H 3
Los Arwetef . CA
. . . . Ml- Hit
; -j 3 j 5333 ; ; ? 5
Car*and$*mlfvH)Ani»f>«SMlbiUlani
SxrjmefHo. CA
t, t*|,
13 5333
, I0
§ 01
I :0
„ 10 •
1 °* '
I '«"
so
1 ">
§!0
JO
» :o
£ 00
i I0
0 *°
8 !0
8 jo
;
i I0
| JO
bililj'1., tlb'»
n n n
• •
< >
Cor* an tonfdnnv A«a>i»a>MiiAnaiis
Detroit, hi
i A " t * 1
I f • • • *
t Tl
- * . S S 3 J * J
New York. NY
Mn
....
? $ 3 3 ! ? 3 3 5*33
Car* and S*ytn SMlUUHt
St.Lous.MO
5 ? >
Figure 7-8. Sensitivity Analysis: Effect of C-R Function Specification on Short-Term Os-attributable Mortality, 2009 Air
Quality. SAl-regional Bayes adjustment, SA2-co-pollutant model (PMio), SA3-Zanobetti and Schwartz-based function.
7-81
-------
Table 7-14. Sensitivity Analysis for Long-Term Os-attributable Respiratory Mortality,
Alternative C-R Function Specification (regional effect estimates), 2009 Air Quality.
Percent of baseline all-cause mortality and change in Os-attributable risk, Jerrett et al.
(2009), O3 season.
Study Area
Air Quality Scenario
% of Baseline
Incidence
75ppb
% Change in O3-Attributable
Risk
75-70
75-65
75-60
Core analysis (2009)
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
17.0
17.4
16.0
16.8
20.0
17.0
16.9
20.7
16.7
17.2
18.0
17.7
5
3
1
4
1
-1
2
4
5
3
4
3
9
6
3
9
5
2
4
8
18
6
7
8
13
10
7
14
12
6
7
13
NA
10
12
12
Sensitivity analysis (regional effect estimates) (2009)
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
41.3
-6.9
-6.1
0.0
27.5
0.0
41.2
4.6
-6.5
-6.7
24.9
0.0
4
4
1
0
1
0
2
3
7
4
4
0
8
9
5
0
4
0
3
7
23
9
7
0
11
13
9
0
11
0
6
11
NA
13
11
0
NA: for NY, the model-based adjustment methodology was unable to adjust Os distributions such that they would meet the lower
alternative standard level of 60 ppb.
7-82
-------
Table 7-15. Sensitivity Analysis for Long-Term Os-attributable Respiratory Mortality,
Alternative C-R Function Specification (national Os-only effect estimates), 2009 Air
Quality. Percent of baseline all-cause mortality and change in Os-attributable risk, Jerrett
et al. (2009), O3 season.
Study Area
Air Quality Scenario
% of Baseline
Incidence
75ppb
% Change in O3-Attributable
Risk
75-70
75-65
75-60
Core analysis (2009)
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
17.0
17.4
16.0
16.8
20.0
17.0
16.9
20.7
16.7
17.2
18.0
17.7
5
3
1
4
1
-1
2
4
5
3
4
3
9
6
3
9
5
2
4
8
18
6
7
8
13
10
7
14
12
6
7
13
NA
10
12
12
Sensitivity analysis (ozone-only effect estimates) (2009)
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
11.9
12.2
11.2
11.8
14.1
11.9
11.9
14.6
11.7
12.1
12.6
12.4
5
3
1
4
1
-1
2
4
5
3
4
4
9
6
3
9
5
2
4
8
18
7
8
8
13
10
7
14
12
6
7
13
NA
10
12
13
NA: for NY, the model-based adjustment methodology was unable to adjust Os distributions such that they would meet the lower
alternative standard level of 60 ppb.
7-83
-------
LT Mortality 2009 (Baseline]
NT86 NT96 T40 T45 T50 T55 TBS TSC
Model (NT: no threshold, T#: threshold in ppb]
LT Mortality 2009 (Current Standard 75]
LT Mortality 2009 (Alternative Standard 70]
Mew York, NV
Philadelphia, P;
Los Angeles, CA
Mew Vork, NV
Philadelphia, PA
NT8S NT9S T40 T45 T50 T55 TBS
Model (NT: no threshold, T#: threshold in ppb
Vlodel (NT: no threshold, T#: threshold in ppb]
LT Mortality 2009 (Alternative Standard 65]
LT Mortality 2009 (Alternative Standard 60]
Los Angeles, CA
Vork, NV
Philadelphia, PA
Model (NT: no threshold, T#: threshold in ppb
Vlodel (NT: no threshold, T#: threshold in ppb]
Figure 7-9. Sensitivity Analysis: Long-Term Os-attributable Respiratory Mortality
(threshold models: Os-only effect estimates), 2009 Air Quality. Each figure presents results
for a particular standard level illustrating the effect of alternative threshold models on Os-
attributable risk across the 12 urban study areas - values are Os-attributable
deaths/100,000. NT86 and NT96 indicate non-thresholds models based on the 86 and 96 city
datasets, respectively. T# models are models reflecting threshold by ppb level. (Jerrett et
al., 2009 and Sasser, 2014).
7-84
-------
7.6 KEY OBSERVATIONS
This section discusses our overall confidence associated with risk estimates presented in
this HREA. We begin by presenting a set of key observations related to overall confidence in the
risk assessment. These observations are drawn largely from (a) consideration for the systematic
approach used in designing the risk assessment, (b) our assessment of the degree to which we
have captured key sources of variability in the analysis (section 7.4.1) (c) our qualitative
assessment of uncertainty in the risk assessment (section 7.4.2), and (d) the results of the
sensitivity analyses completed (section 7.5.3). Once we present these observations, we provide a
synthesis statement reflecting our overall degree of confidence in the risk estimates (at the end of
this section). Key observations addressing overall confidence in the analysis include:
• A deliberative process was used in specifying each of the analytical elements
comprising the risk model. This is in line with recommendations made by the
National Research Council in Science and Decisions, Advancing Risk Assessment
(NRC, 2009. p. 89-90) for improving risk assessment as applied in the regulatory
context. This deliberative process included first identifying specific goals for the
analysis, and then designing the analysis to meet those goals, given available
information and methods. Specific analytical elements reflected in the design include:
selection of urban study areas, characterization of ambient air Os levels, selection of
health endpoints to model and selection of epidemiological studies (and specification
of C-R functions) (see sections 7.1.1 and 7.3). In addition, the design of this HREA
reflects consideration for comments provided by the public and by CASAC in their
review of the draft HREAs (Frey, and Samet 2012; Frey, 2014).
• Review of available literature in the Os ISA(U.S. EPA. 2013a) for short-term
exposure-related mortality, resulted in a decision not to incorporate a true (no effect)
threshold into our risk modeling. However, the studies used to develop the C-R
functions specifically for short-term exposure-related mortality indicate a range of
ambient Os (area-wide daily levels, based on averaging across monitors in locations
with multiple monitors, of < 20 ppb) below which there is reduced confidence in
specifying the nature of the C-R relationship (see section 7.1.1). Only a relatively
small fraction of short-term Os-attributable mortality reflected in the risk estimates is
associated with days in this range with the vast majority of the risk estimates
reflecting days with peak Os measurements well above this level (see section 7.5.1
and 7.5.2).
• Modeling of short-term O3-attributable mortality utilized B ayes-adjusted city-specific
effect estimates (see section 7.1.1 and section 7.3.2). These effect estimates are
7-85
-------
considered to have increased overall confidence since they combine elements of the
local city-specific signal with a broader scale (national) signal.
Sensitivity analyses exploring alternative C-R functions for modeling short-term Ch-
attributable mortality (e.g., Bayes regional prior based estimates, co-pollutants
models) suggested that alternative models can have a moderate impact on risk (see
section 7.5.3). This impact reflects primarily the relatively small magnitude of short-
term Os-attributable mortality reductions associated with air quality adjusted to just
meet the alternative standard levels.
The use of alternative C-R functions for modeling long-term Ch-attributable mortality
(specifically the regional-based estimates referenced earlier) was shown to have a
significant impact on risk (see section 7.5.3). However, concerns over the power and
hence stability of the regional effect estimates used in this simulation limit our ability
to draw firm conclusions regarding the potential magnitude of that regional
heterogeneity. Estimates of mortality risk attributable to long-term Os exposure are
highly sensitive to the existence of a threshold (see section 5.3).
Use of CBSA-based study areas in modeling all health endpoints in order to address
known bias associated with using smaller study areas. As discussed in 7.1.1, we have
used larger CBSA-based study areas to avoid focusing the risk assessment only on
core urban areas (often used in the epidemiological studies providing effect estimates)
which can experiences increases in Os based on simulated air quality that just meets
the existing and alternative standard levels. There is uncertainty in using effect
estimates based on smaller study areas to represent larger CBSA-based study areas
(see section 7.4.2 and 7.5.3). A key concern is heterogeneity in the effect estimates
which may suggest increased uncertainty in applying effect estimates to larger study
areas (since larger study areas may display heterogeneity in the nature of the
relationship between Os exposure and risk). It is possible also that this heterogeneity
varies across urban areas, or regionally. For both categories of mortality endpoints
(short-term and long-term Os-attributable), potential heterogeneity in the mortality
effect even within larger urban areas remains a potentially important source of
uncertainty.
Specifically in relation to short-term exposure-related mortality and morbidity which
depend on time-series studies, there is uncertainty in applying effect estimates derived
based on evaluating the longitudinal (in terms of time) relationship between ambient
Os and a particular health effect to the modeling of a discrete shift in the entire
distribution that occurs when you simulate an alternative standard. Specifically, the
time-series studies relate unit changes in day to day Os with a degree of impact on
7-86
-------
baseline health effect rates. In the risk assessment, we use this effect estimate to
predict risk for a unit shift in daily composite monitor value. There is uncertainty in
this application of the effect estimates, although it is not possible at this time to
characterize either qualitatively, or quantitatively the magnitude of this uncertainty
and the degree of any potential bias that could be introduced into the simulation of
risk.
• Use of HDDM-adjustment approach to simulate air quality that just meets the
existing and alternative standard levels provides more refined estimates of ambient Os
distributions given its ability to characterize the physical and chemical processes of
Os formation in the atmosphere. However, in the case of both the New York and Los
Angeles study areas, given the limitations in the application of the adjustment
methodology to very large emissions perturbations and the need to use the 95th
percent confidence interval lower bound estimate to simulate air quality that just
meets of these standard levels, we have reduced overall confidence in the simulation
of the Os concentrations for these study areas and consequently all health endpoints
modeled for risk for these two study areas (see section 7.4.2 and 7.5.3).
Based on the key observations regarding confidence presented above, we draw the
following conclusions regarding overall confidence in the risk estimates generated for this
HREA. We have a reasonable degree of confidence in short-term Ch-attributable mortality and
morbidity estimates for ten of the twelve study areas. This confidence is tempered somewhat by
concerns over potential heterogeneity in effect estimates for mortality which can impact the risk
assessment given our use of larger CBSA-based study areas. Our confidence in risk estimates
generated for both New York and Los Angeles urban study areas is considerably lower than for
the remaining ten study areas due to (a) concerns over air quality modeling (specifically the use
of lower-bound fits to the DDM model) and (b) specifically in the case of New York, evidence
for significant heterogeneity in the mortality effect estimates for subareas within the CBS A. We
have somewhat lower confidence in our estimates of mortality risk attributable to long-term Os
exposures, primarily because there is only a single well designed study, and because of the large
impact of uncertainty around the existence and potential location of a threshold in the C-R
function for this endpoint. In addition, as with short-term Os-attributable mortality, overall
confidence is further tempered by concerns over regional heterogeneity in the Os effect. If we
had regionally-differentiated effect estimates for this endpoint that had sufficient power and
stability, we would consider using these as the basis for generating core risk estimates (rather
than the national-level effect estimate used in the current analysis).
7-87
-------
7.7 REFERENCES
Abt Associates Inc. 1996. A Paniculate Matter Risk Assessment for Philadelphia and Los
Angeles. Prepared for Office of Air Quality Planning and Standards. Research Triangle
Park, NC: EPA Office of Air and Radiation, OAQPS, July 1996. Available at:
.
Abt Associates, Inc. 2010. Environmental Benefits and Mapping Program (Version 4.0).
Prepared for U.S. Environmental Protection Agency, Bethesda, MD. Research Triangle
Park, NC: EPA Office of Air Quality Planning and Standards. Available on the Internet
at: .
Akinbami, L.J.; C.D. Lynch; J.D. Parker and TJ. Woodruff. 2010. The association between
childhood asthma prevalence and monitored air pollutants in metropolitan areas, United
States, 2001-2004. Environmental Research. 110:294-301.
Bell, M.L. and F. Dominici. 2008. Effect modification by community characteristics on the
short-term effects of ozone exposure and mortality in 98 U.S. communities. American
Journal of Epidemiology. 167:986-997.
Bell, M.L.; A. McDermott; S.L. Zeger; J.M. Samet; F. Dominici. 2004. Ozone and short-term
mortality in 95 U.S. urban communities, 1987-2000. JAMA. 292:2372-2378.
Darrow, L.A.; M. Klein; J.A. Sarnat; J.A. Mulholland; MJ. Strickland; S.E. Sarnat, et al. 2011.
The use of alternative pollutant metrics in time-series studies of ambient air pollution and
respiratory emergency department visits. Journal of Exposure Science and Environmental
Epidemiology. 21:10-19.
Frey, C. and J. Samet. 2012. CASAC Review of theEPAJs Health Risk and Exposure Assessment
for Ozone (First External Review Draft - Updated August 2012) and Welfare Risk and
Exposure Assessment for Ozone (First External Review Draft - Updated August 2012). U.S.
Environmental Protection Agency Science Advisory Board. EPA-CASAC-13-002.
Frey, C. 2014. CASAC Review of the EPA's Health Risk and Exposure Assessment for Ozone (Second
External Review Draft-February, 2014). U.S. Environmental Protection Agency Science
Advisory Board. EPA-CASAC-14-005.
Gent, J.F.; E.W. Triche; T.R. Holford; K. Belanger; M.B. Bracken; W.S. Beckett, et al. 2003.
Association of low-level ozone and fine particles with respiratory symptoms in children
with asthma. Journal of the American Medical Association. 290(14): 1859-1867.
Ito, K.; G.D. Thurston and R.A. Silverman. 2007. Characterization of PM2.5, gaseous pollutants,
and meteorological interactions in the context of time-series health effects models.
Journal of Exposure Science and Environmental Epidemiology. 17(S2):S45-60.
Jerrett, M.; R.T. Burnett; C.A. Pope, III; K. Ito; G. Thurston; D. Krewski; Y. Shi; E. Calle; and
M. Thun. 2009. Long-term ozone exposure and mortality." New England Journal of
Medicine. 360:1085-1095.
Katsouyanni, K.; J.M. Samet; H.R. Anderson; R. Atkinson; A.L. Tertre; S. Medina, et al. 2009.
Air Pollution and Health: A European and North American Approach (APHENA). Health
Effects Institute.
Lin, S.; E.M. Bell; W. Liu; R.J. Walker; N.K. Kim and S.A. Hwang. 2008a. Ambient ozone
concentration and hospital admissions due to childhood respiratory diseases in New York
State, 1991- 2001. Environmental Research. 108:42-47.
7-88
-------
Lin, S; X. Liu; L.H. Le; S.A. Hwang. 2008. Chronic exposure to ambient ozone and asthma
hospital admissions among children. Environmental Health Perspective. 116:1725-1730.
Linn, W.S.; Y. Szlachcic; H. Gong, Jr.; P.L. Kinney and K.T. Berhane. 2000. Air pollution and
daily hospital admissions in metropolitan Los Angeles. Environmental Health
Perspective. 108(5):427-434.
Medina-Ramon, M.; A. Zanobetti and J. Schwartz. 2006. The effect of ozone and PMio on
hospital admissions for pneumonia and chronic obstructive pulmonary disease: a national
multicity study. American Journal of Epidemiology. 163(6):579-588.
Meng, Y.Y.; R.P. Rull; M. Wilhelm; C. Lombard!; J. Balmes and B. Ritz. 2010. Outdoor air
pollution and uncontrolled asthma in the San Joaquin Valley, California. Journal of
Epidemiology Community Health. 64:142-147.
Moore, K; R. Neugebauer; F. Lurmann; J. Hall; V. Brajer; S. Alcorn; I. Tager. 2008. Ambient
ozone concentrations cause increased hospitalizations for asthma in children: an 18-year
study in southern California. Environmental Health Perspective. 116:1063-1070.
NRC. 2009. Science and Decisions, Advancing Risk Assessment. Committee on Improving Risk
Analysis Approaches. Washington, DC: The National Academies Press, National
Research Council.
Sasser, E. 2014. Response to Comments Regarding the Potential Use of a Threshold Model in
Estimating the Mortality Risks from Long-term Exposure to Ozone in the Health Risk and
Exposure Assessment for Ozone, Second External Review Draft. Memorandum to Holly
Stallworth, Designated Federal Officer, Clean Air Scientific Advisory Committee from
EPA/OAQPS Health and Environmental Impacts Division.
Silverman, R.A.; and K. Ito. 2010. Age-related association of fine particles and ozone with
severe acute asthma in New York City. Journal of Allergy Clinical Immunology.
125(2):367-373.
Smith, R.L.; B. Xu and P. Switzer. 2009. Reassessing the relationship between ozone and short-
term mortality in U.S. urban communities. Inhalation Toxicology. 21:37-61.
Strickland, M.J.; L.A. Darrow; M. Klein; W.D. Flanders; J.A. Sarnat; L.A. Waller, et al. 2010.
Short-term associations between ambient air pollutants and pediatric asthma emergency
department visits. American Journal of Respiratory Critical Care Medicine. 182:307-
316.
Tolbert, P.E.; M. Klein; J.L. Peel; S.E. Sarnat and J.A. Sarnat. 2007. Multipollutant modeling
issues in a study of ambient air quality and emergency department visits in Atlanta.
Journal of Exposure Science and Environmental Epidemiology. 17(S2):S29-35.
CDC. 2010. Table C1 Adult Self-Reported Current Asthma Prevalence Rate (Percent) and
Prevalence (Number) by State or Territory. Centers for Disease Control and Prevention,
Behavioral Risk Factor Surveillance System (BRFSS). Available at:
.
U.S. EPA. 2001. Risk Assessment Guidance for Superfund. Vol. Ill, Part A. Process for
Conducting Probabilistic Risk Assessment (RAGS 3A). Washington, DC: EPA. (EPA
document number EPA 540-R-02-002; OSWER 9285.7-45; PB2002 963302). Available
at: .
U.S. EPA. 2004. EPA 's Risk Assessment Process for Air Toxics: History and Overview. In: Air
Toxics Risk Assessment Reference Library, Technical Resource Manual, Vol. 1, pp. 3-1
7-89
-------
- 3-30. (EPA document number EPA-453-K-04-001 A). Washington, DC: EPA.
Available at: .
U.S. EPA. 2007. Ozone Health Risk Assessment for Selected Urban Areas. Research Triangle
Park, NC: EPA Office of Air Quality Planning and Standards. (EPA document number
EPA 452/R-07-009). Available at:
.
U.S. EPA. 2009. Integrated Science Assessment for Particulate Matter: Final. Research Triangle
Park, NC: U.S. Environmental Protection Agency. (EPA document number EPA/600/R-
08/139F).
U.S. EPA. 2011. Ozone National Ambient Air Quality Standards: Scope and Methods Plan for
Health Risk and Exposure Assessment. Research Triangle Park, NC: EPA. (EPA
document number EPA-452/P-11-001).
U.S. EPA. 2012. Health Risk and Exposure Assessment for Ozone. First External Review Draft.
Research Triangle Park, NC: U.S. Environmental Protection Agency, Research Triangle
Park, NC. (EPA document number EPA 452/P-12-001).
U.S. EPA. 2013 a. Integrated Science Assessment for Ozone and Related Photochemical
Oxidants: Final. Research Triangle Park, NC: U.S. Environmental Protection Agency.
(EPA document number EPA/600/R-10/076F).
U.S. EPA. 2013b. Environmental Benefits Mapping Analysis Program (BenMAP v4.0). Posted
January, 2013. < http://www.epa.gov/air/benmap/download.html />.
WHO. 2008. Part 1: Guidance Document on Characterizing and Communicating Uncertainty in
Exposure Assessment, Harmonization Project Document No. 6. Published under joint
sponsorship of the World Health Organization, the International Labour Organization and
the United Nations Environment Programme. WHO Press, World Health Organization,
20 Avenue Appia, 1211 Geneva 27, Switzerland.
Zanobetti, A; J. Schwartz. 2008. Mortality displacement in the association of ozone with
mortality: an analysis of 48 cities in the United States. American Journal of Respiratory
and Critical Care Medicine. 177:184-189.
7-90
-------
8 CHARACTERIZATION OF NATIONAL-SCALE MORTALITY RISK
BASED ON EPIDEMIOLOGICAL STUDIES AND AN URBAN-SCALE
REPRESENTATIVENESS ANALYSIS
As described in Chapter 2, the Os ISA (U.S. EPA, 2013) concluded that there is likely to
be a causal relationship between short-term Os exposure and all-cause mortality and that there is
likely to be a causal relationship between long-term Os exposure and respiratory effects,
including respiratory mortality. In Chapter 7, we estimated epidemiological-based health risks
associated with recent Os concentrations and meeting the current and alternative Os standards in
12 selected urban study areas. In this chapter we estimate nationwide premature mortality
attributable to recent short-term and long-term exposures to ambient Os (Section 8.1) based on
results from epidemiological studies; and assess the degree to which the selected urban study
areas represent the full national distribution of risk-related attributes and air quality dynamics
(Section 8.2). Compared with the urban-scale analysis in Chapter 7, this national-scale analysis
incorporates fully, ambient Os concentrations and population information across the entire
continental U.S. however uses less geographic specificity in the concentration-response (C-R)
functions that are used to calculate Os-attributable mortality. The national-scale analysis is
therefore intended as a complement to the urban-scale analysis, providing both a broader
assessment of Os-related health risks across the U.S. as well as an evaluation of how well the
urban study areas examined in Chapter 7 represent the full distribution of Os-related health risks
and air quality dynamics in the U.S.
8.1 NATIONAL-SCALE ASSESSMENT OF MORTALITY RELATED TO OZONE
EXPOSURE
This section estimates the total annual deaths for 2007 populations associated with
average 2006-2008 Os levels across the continental U.S. We first describe the methods and
inputs used to estimate Cb-attributable risk across the continental U.S., including Os exposure
estimates, population and baseline mortality rate estimates, and epidemiological-based Os
mortality effect estimates. Results for the estimation of Os-attributable risk are then discussed in
terms of the magnitude and percent of total mortality attributable to Os exposure. We provide
two analyses to give perspective on the confidence in the estimates of Os-related mortality: (1)
risk estimated only within the urban areas for which particular Os mortality effect estimates are
available; and (2) the distribution of Os-related deaths across the range of observed 2006-2008
average Os concentrations fused with modeled 2007 concentrations. These results are then
synthesized and compared with previous estimates of the burden of Os exposure on mortality in
the U.S. from the literature in a discussion section.
3-1
-------
Air Quality Inputs (Chapter 4)
National ambient ozone
12 kmx 12km
gridded spatial field
for recent year
v . /
Concentration-Response Functions
Nationwide set of city
specific C-R functions
Identify appropriate
modeling period
BenMAP
Population Information
Daily county-specific
baseline health incidence
Population allocated
to 12 km x12 km grid
Compute gridded ozone-attributable
mortality incidence for time period
matching C-R function season
County and national level
[estimates of burden of premature
mortality attributable to ozone
County and national level
estimates of % of total mortality
attributable to ozone
Figure 8-1. Conceptual Diagram for National-scale Mortality Risk Assessment.
8.1.1 Methods
This section describes the inputs and datasets used to conduct the national-scale
assessment of Os-attributable risk. As shown in the conceptual diagram in Figure 8-1, we
conduct this analysis using BenMAP 4.0, which uses projections of the size and geographic
distribution of the potentially exposed population along with estimates of the ambient Os
concentrations to estimate Os-attributable health risks. In general, this analysis uses the same
analytical structure and many of the same inputs as are used in the epidemiology-based
assessment of Os-attributable risk in the selected urban study areas in Chapter 7. We refer back
to Chapter 7 for details on these shared inputs, and describe where the urban-scale and national-
scale analyses use divergent methods.
8-2
-------
8.1.1.1 Ambient ozone concentrations
Air quality inputs to this analysis are described in detail in Chapter 4. In contrast to the
urban-scale analysis in Chapter 7, the national-scale analysis employs a data fusion approach that
takes advantage of the accuracy of monitor observations and the comprehensive spatial
information of the CMAQ modeling system to create national-scale "fused" spatial surfaces of
seasonal average Os. Measured Cb concentrations from 2006-2008 were fused with modeled
concentrations from a 2007 CMAQ model simulation, run for a 12 km domain covering the
contiguous U.S. The spatial surfaces are created using EPA's Downscaler software (Berrocal et
al., 2012).l More details on the ambient measurements, the 2007 CMAQ model simulation, the
Downscaler fusion technique, and a technical justification for changing from eVNA to
Downscaler can be found in Chapter 4.
Three "fused" spatial surfaces were created for: (1) the May-September mean of the daily
maximum 8-hour (8-hr) average concentration (consistent with the metric used by Smith et al.
2009); (2) the June-August mean of the daily 8-hr average concentration occurring between
10am to 6pm (consistent with the metric used by Zanobetti and Schwartz, 2008); and (3) the
April-September mean of the daily maximum 1-hour (1-hr) concentration (consistent with the
metric use by Jerrett et al., 2009) Os concentrations across the continental U.S. These fused
spatial surfaces each represent one seasonal average across 2006-2008, rather than three separate
years of concentrations. Section 4.3.2 presents maps, distributions, and statistical
characterizations of these Cb concentrations metrics across the U.S., including how they compare
to 2006-2008 design values.
8.1.1.2 Concentration-response functions
While Chapter 7 assessed both mortality and morbidity risks associated with Os
concentrations, due to limitations in information available for baseline morbidity incidence rates,
the national scale assessment focuses on mortality risks only. To quantify the impact of Os
concentrations on mortality, we apply effect estimates drawn from two major short-term
epidemiological studies and one long-term epidemiological study. These studies are consistent
with those used in the analysis of Os-related risk in selected urban study areas (Section 7.2) and
those mortality endpoints concluded to have a causal or suggestive causal relationship with Os
exposure by the 2013 Os ISA (U.S. EPA, 2013). As described in detail in Section 7.2, we
selected C-R functions by applying a number of criteria including: (1) the study was peer-
reviewed, evaluated in the Os ISA, and judged adequate by EPA staff for purposes of inclusion
1 In the first draft of the HREA, the spatial surfaces were created using the enhanced Voronoi Neighbor Averaging
(eVNA) technique (Timin et al., 2010) and using the EPA's Model Attainment Test Software (MATS; Abt
Associates, 2010b).
8-3
-------
in the risk assessment; (2) Preference was given for multicity studies because they typically have
greater statistical power and reflect patterns of Os related health effects over a range of urban
areas (and regions) which can display variability in key risk-related factors such as exposure
measurement error; (3) The study design is considered robust and scientifically defensible,
particularly in relation to methods for covariate adjustment, including treatment of confounders,
as well as treatment of effect modifiers; and (4) The study is not superseded by another study
(e.g., if a later study is an extension or replication of a former study, the later study would
effectively replace the former study), unless the earlier study has characteristics that are clearly
preferable (e.g., inclusion of copollutants models, or use of a peak exposure metric of interest).
For short-term mortality, we use city-specific and national average risk estimates drawn
from the Smith et al. (2009) study of Os and mortality in 98 U.S. urban communities between
1987 and 2000 as our main results, and the Zanobetti and Schwartz (2008) study of Ch and
mortality in 48 U.S. cities between 1989 and 2000 as a sensitivity analysis, consistent with the
urban study area analysis performed in Chapter 7. The city-specific effect estimates obtained
from both studies are provided in Appendix 8A.
Smith et al. (2009) found that the average non-accidental mortality increase across all 98
urban areas studied was 0.32% ± 0.08 (95% posterior interval [PI], 0.41%-0.86%) for a 10 ppb
increase in the daily maximum 8-hr average Os concentration, based on April to October Os
concentrations. Because the national-scale analysis requires a single modeling period definition
but some monitors only collect data from May to September, the corresponding city-specific
effect estimates are applied to each day from May to September in BenMAP using the May to
September mean of the daily maximum 8-hr average Os concentration based on 2006-2008
observed concentrations fused with 2007 modeled concentrations. The length of the Os season
can affect the magnitude of mortality effect estimates - a longer season may yield higher effect
estimates per unit Os concentration since Os concentrations over a longer season may be lower
than the Os concentrations that occur over the warmest months only. Conversely, if the longer
period captures periods of lower Os-related mortality incidence, the effect estimates may be
lower than effect estimates for the warmest months only. Our application of the Smith et al.
(2009) April to October effect estimates to the May through September Os concentrations likely
introduces some bias in the results, but it is unclear in which direction.
Zanobetti and Schwartz (2008) found that the average total mortality increase across all
48 cities studied was 0.53% (95% confidence interval, 0.28%-0.77%) for a 10 ppb increase in the
June to August mean of the daily 8-hr average concentration occurring between 10am to 6pm,
using a 0-3 day lag. We apply the city-specific effect estimates that correspond to this national
average effect estimate each day from June to August in BenMAP using the June to August,
means of the daily 8-hr average Os concentration occurring between 10am to 6pm based on
8-4
-------
2006-2008 observed concentrations fused with 2007 modeled concentrations. Consistent with
Chapter 7, these results are presented as a sensitivity analysis.
As in Chapter 7, we use city-specific risk estimates from the short-term epidemiology
studies, but apply them here only to the counties that were included in the epidemiology studies
rather than to the entire core-based statistical area (CBSA). Chapter 7 estimated risk across entire
CBS As to more completely capture expected Os changes across broader areas and avoid bias
resulting from including only those areas where Os is expected to increase under alternative
standards. The inclusion of the entire CBSA in that analysis required the application of a single
effect estimate to the entire CBSA. However, the national-scale assessment is a gridded analysis,
which allows greater spatial resolution in the application of effect estimates. In addition, eight
CBS As nationwide included multiple cities defined separately by Smith et al. (2009), some of
which showed considerable heterogeneity in effect estimates within the same CBSA.
Heterogeneity among effect estimates within a single CBSA implies that effect estimates from
one county may not be accurate representations of effect estimates in nearby
counties. However, since city-specific effect estimates often have low power due to small
population size, we are unable to draw a strong conclusion regarding how well one county's
effect estimates represents those in nearby counties. For this national-scale assessment, we
apply effect estimates from each city as defined in the epidemiology studies to retain the full set
of information available from those studies. In addition, for counties not included by the
epidemiology studies, we apply the average effect estimate derived from all the urban areas
included in each of the studies ("national average") as it takes advantage of a wider and more
diverse population.
Since both national average estimates from these studies are based on urban areas only,
we have higher confidence in their application to other U.S. urban areas than to rural areas. To
demonstrate the magnitude of the results for which we have the highest confidence, we present
the percentage of estimated deaths occurring within the urban areas included in the
epidemiological studies and within all urban areas across the U.S. Lower confidence in the
results for rural areas does not indicate that the mortality risk among populations living in such
areas is unaffected by Os pollution. Rather, the level of understanding for the Os-mortality
relationship in these areas is simply lower due to a lack of available epidemiological data at these
levels. We also examine the effect of varying the effect estimate applied between the cities
included by the epidemiology studies in a sensitivity analysis.
We quantify long-term Os-related respiratory mortality in this Health Risk and Exposure
Assessment (HREA) because the Os ISA concluded that the evidence supports a likely to be
causal relationship between long-term Os exposure and respiratory effects, including respiratory
morbidity and respiratory-related mortality (U.S. EPA, 2013). As detailed in Chapter 7, we
8-5
-------
quantify long-term Os-related mortality using the respiratory mortality effect estimates from the
Jerrett et al. (2009) two-pollutant model that controlled for PIVfo.s concentrations, applied to each
grid cell across the entire United States. This model found that a 10 ppb increase in the April-
September mean of the daily maximum 1-hr Os concentration was associated with a 4% (95%
confidence interval, 1.0%-6.7%) increase in respiratory mortality.
8.1.1.3 Demographic inputs
This analysis uses the same baseline mortality rates and population estimates as were
used in the urban study area analysis in Chapter 7. We derive baseline incidence rates for
mortality by age, cause, and county from the CDC Wonder database (CDC, 2004-2006). As this
database only provides baseline incidence rates in 5-year increments, we use data for the year
2005, the closest year to the analysis year 2007 used for the population and air quality modeling.
We use 2007 population because it matches both the year of the emissions inventory and
meteorology used for the air quality modeling.
The starting point for estimating the size and demographics of the potentially exposed
population is the 2010 census-block level population, which BenMAP aggregates up to the same
grid resolution as the air quality model. BenMAP back-casts this 2010 population to the analysis
year of 2007 using county-level growth factors based on economic projections (Woods and
Poole Inc., 2012).
8.1.2 Results
Table 8-1 summarizes the estimated Os-related premature mortality associated with 2006-
2008 average Os concentrations under various assumptions for the health impact function.
Applying Smith et al. (2009) effect estimates for May-September, we estimate 15,000 (95% CI,
1,400-28,000) premature Os-related non-accidental deaths annually for 2007. As a sensitivity
analysis, we apply Zanobetti and Schwartz (2008) effect estimates for June-August, finding
16,000 (95% CI, 6,000-25,000) premature Os-related all-cause deaths annually for 2007. Figure
8-2 through Figure 8-4 show that estimated Os-related mortality is most concentrated in highly
populated counties or those counties with urban areas found to have high effect estimates by
Smith et al. (2009) or Zanobetti and Schwartz (2008). For the application of Jerrett et al. (2009)
national average effect estimate for April-September, we estimate 45,000 (95% CI, 17,000-
70,000) premature Os-related respiratory deaths among adults age 30 and older.
Because the epidemiological studies included only selected urban areas, we are more
confident in the magnitude of the estimated Os-related deaths occurring within those urban areas.
As shown in Table 8-1, approximately 43% of the Os-related deaths estimated using Smith et al.
(2009) effect estimates occur in the 98 urban areas included in that study, and 30% of the Os-
-------
related deaths estimated using Zanobetti and Schwartz (2008) effect estimates occur in the 48
urban areas included in that study. The cities included in each epidemiological study are listed in
Appendix 8A. We are also more confident in extrapolating the national average effect estimates
to other urban areas than we are to rural areas, as the national average estimates are based on all
urban areas included by the study. To estimate the percentage of total Os-attributable deaths
occurring within all urban areas across the continental U.S., we sum the results for the 12 km
grid cells that have a total population greater than 12,000 (approximately equal to the 95th
percentile of grid cell populations across the continental U.S.). The percentage of Os-attributable
deaths occurring within urban areas defined in this way is 65% for results based on Smith et al.
(2009) effect estimates and 64% for results based on Zanobetti and Schwartz (2008) effect
estimates. While our confidence is lower when the national average effect estimates are
extrapolated to rural areas, less certainty in the magnitude of Os-related deaths in rural areas does
not imply that Os has no effect on health in these areas.
Table 8-1. Estimated Annual Os-related Premature Mortality in 2007 Associated with
2006-2008 Average Os Concentrations (95 percent confidence interval).
Source of Risk Estimate and Modeling
Period
Smith et al. (2009), May-Sep
(95% confidence interval)
% occurring within the 98 cities3
Zanobetti & Schwartz (2008), Jun-Aug
(95% confidence interval)
% occurring within the 48 cities3
Jerrett et al. (2009), Apr-Sep
(95% confidence interval)
Exposure
Duration
Short-term
Short-term
Long-term
Study
Subject
Ages
>0
>0
>30 years
Annual Os-related Premature
Mortality
City-specific
Effect
Estimates1
15,000
(1,400-28,000)
43%
16,000
(5,900-25,000)
30%
-
National
Average Effect
Estimate2
15,000
(7,100-22,000)
15,000
(8,300-22,000)
45,000
(16,000-70,000)
1 City-specific effect estimates are applied to the grid cells lying within the cities defined in the epidemiological
studies. Average effect estimates across all cities included in the epidemiological studies (national average) are
applied to all other grid cells. For the application of Smith et al. (2009) effect estimates, city-specific effect estimates
were applied to 2,227 grid cells and the national average to 44,064 grid cells. For the application of Zanobetti and
Schwartz (2008) effect estimates, city-specific effect estimates were applied to 925 grid cells and the national
average to 45,366 grid cells.
2 National average effect estimates are based on the average of all cities included in the epidemiological studies
applied to all 12km grid cells nationally.
3 Os-attributable deaths are summed in each city within BenMAP using area weights based on a shapefile definition
of the counties included in each city. Summed city-specific results are then divided by the total U.S. Os-attributable
deaths to generate the percentage occurring within the cities included by Smith et al. (2009) and Zanobetti and
Schwartz (2008).
3-7
-------
Table 8-1 also shows Os-related deaths estimated by applying the national average risk
estimate from the epidemiological studies to all grid cells in the U.S. As discussed in detail in
Section 7.4.1, applying city-specific effect estimates accounts for heterogeneity between cities
due to differences in population characteristics, potential confounding, potential for the presence
of averting behavior, and variation in sample sizes which can affect stability of effect estimates.
By contrast, the national average effect estimate does not distinguish differential effects between
cities but has the advantage of a substantially larger sample size that includes a wider range of
populations, confounding factors, and behaviors. Compared with applying city-specific effect
estimates to the grid cells corresponding to each urban area, using the national average effect
estimate for all grid cells yields equivalent central estimates. However, applying the national
average also results in tighter confidence intervals since the national average effect estimates had
higher statistical power and thus tighter confidence bounds compared with the effect estimates
for individual cities.
Table 8-2 shows the mean, median, 2.5th percentile and 97.5th percentile of the estimated
percentage of mortality attributable to ambient Cb across all counties in the U.S. Using Smith et
al. (2009) effect estimates, Os-attributable mortality contributes an average of 1.5% (95%
confidence interval, 1.1%-1.8%) to county-level May-September non-accidental mortality (all
ages) and 0.6% (0.4%-0.7%) to all year all-cause mortality (all ages). For results using Zanobetti
and Schwartz (2008) effect estimates, Os-attributable mortality contributes an average of 2.5%
(95% confidence interval, 1.7%-3.0%) to county-level June-August all-cause mortality (all ages)
and 0.6% (0.4%-0.8%) to all year all-cause mortality (all ages). For the results using Jerrett et al.
(2009) effect estimates, Os-attributable mortality contributes an average of 18.5% (95%
confidence interval, 15.2%-21.5%) to county-level April-September adult (age 30+) respiratory
mortality and 1.9% (1.3%-2.6%) to all year all-cause mortality (all ages). Figure 8-5 through
Figure 8-7 show that the counties with the highest percentage of mortality attributable to Os are
typically those with the highest Os levels.
-------
Table 8-2. Mean, Median, 2.5th Percentile, and 97.5th Percentile of the Estimated
Percentage of Mortality Attributable to Ambient Os for all 3,109 Counties in the
Continental U.S.
Source of Risk Estimate, Modeling
Period, and Mortality Endpoints1
Smith et al. (2009), May-Sep
Non-accidental mortality, all ages
All-cause mortality, all ages
All-cause mortality, all ages, all year
Zanobetti & Schwartz (2008), Jun-Aug
All-cause mortality, all ages
All-cause mortality, all ages, all year
Jerrett et al. (2009), Apr-Sep
Respiratory mortality, ages 30+
All-cause mortality, all ages
All-cause mortality, all ages, all year
Total
Mortality
(2005)
960,000
1,000,000
2,500,000
620,000
2,500,000
240,000
1,200,000
2,500,000
Percentage of Total Mortality Attributable to Os
Mean
(%)
1.5
1.4
0.6
2.5
0.6
18.5
3.8
1.9
Median
(%)
1.5
1.4
0.6
2.5
0.6
18.7
3.7
1.9
2.5th
Percentile
(%)
1.1
1.0
0.4
1.7
0.4
15.2
2.6
1.3
97.5th
Percentile
(%)
1.8
1.7
0.7
3.0
0.8
21.5
5.2
2.6
For the mortality endpoints matching the epidemiology studies as a percentage of incidence of the same endpoint
for the same seasonal definition, and as a percentage of all-cause mortality for all age groups (both seasonal and all
year).
8-9
-------
Jo
Figure 8-2. Estimated Annual non-Accidental Premature Deaths (individuals) in 2007
Associated with Average 2006-2008 May-September Means of the Daily Maximum 8-hr
Average Os Levels by U.S. County using Smith et al. (2009) Effect Estimates.
Figure 8-3. Estimated Annual all-Cause Premature Deaths (individuals) in 2007 Associated
with Average 2006-2008 June-August Means of the Daily 8-hr Average (10am-6pm) Os
Levels by U.S. County using Zanobetti and Schwartz (2008) Effect Estimates.
8-10
-------
n,
K
Figure 8-4. Estimated Annual Adult (age 30+) Respiratory Premature Deaths (individuals)
in 2007 Associated with Average 2006-2008 April-September Means of the Daily Maximum
1-hr O3 Levels by U.S. County using Jerrett et al. (2009) Effect Estimates
<
'
Figure 8-5. Estimated Percentage of May-September Total non- Accidental Mortality (all
ages) Attributable to 2006-2008 average O3 Levels by U.S. County using Smith et al. (2009)
Effect Estimates.
8-11
-------
Figure 8-6. Estimated Percentage of June-August Total all-Cause Mortality (all ages)
Attributable to 2006-2008 Average Os Levels by U.S. County using Zanobetti and Schwartz
(2008) Effect Estimates.
Figure 8-7. Estimated Percentage of April-September Respiratory Mortality among Adults
Age 30+ Attributable to 2006-2008 Average O3 Levels by U.S. County using Jerrett et al.
(2009) Effect Estimates.
8-12
-------
M
re
A-f
c
01
u
OJ
Q.
01
'+-«
re
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
0 0.5 1 1.5 2 2.5 3
Percentage of total mortality attributable to ozone
—Smith et al. (2009) —Zanobetti and Schwartz (2008) —Jerrett et al. (2009)
Figure 8-8. Cumulative Distribution of County-level Percentage of all-cause, all-year, and
all-age Mortality Attributable to 2006-2008 Average O3 for the U.S.
Figure 8-8 displays the cumulative distribution of the percent of county-level all-cause,
all-age, and all-year mortality attributable to ambient Os using effect estimates from all three
epidemiological studies.2 For the results based on Smith et al. (2009) and Zanobetti and Schwartz
(2008) effect estimates, 0.8% of all-cause, all-age, and all-year mortality is attributable to Os for
approximately 99% of U.S. counties. For the results based on Jerrett et al. (2009) effect
estimates, 2.8% of all-cause, all-age, and all-year mortality is attributable to Os for
approximately 99% of U.S. counties.
Figure 8-9 shows the cumulative distribution of the county-level percent of total Os-
related deaths by Os concentration. The mortality results based on Smith et al. (2009) C-R
functions are compared with the May-September mean of the daily maximum 8-hr average Os
concentration, those based on Zanobetti and Schwartz (2008) C-R functions are compared with
the June-August mean of the 8-hr average Os concentration from 10am to 6pm, and those based
2 Estimated Os-attributable deaths are based on the mortality cause, age group, and season inherent to the
epidemiological study upon which it is based (May-September non-accidental mortality for all ages for results
based on Smith et al. (2009) effect estimates, June-August all-cause mortality for all ages for results based on
Zanobetti and Schwartz (2008) effect estimates, and April-September respiratory mortality for ages 30+ for results
based on Jerrett et al. (2009) effect estimates).
8-13
-------
on Jerrett et al. (2009) C-R functions are compared with the April-September mean of the daily
maximum 1-hr Os concentration, consistent with the Os concentration metrics used in each
study. The mortality results based on Zanobetti and Schwartz (2008) effect estimates are shifted
to the right of the mortality results based on the Smith et al. (2009) C-R functions because the
seasonal averaging time for the results based on Zanobetti and Schwartz (2008) is limited to the
summer months when Os tends to be highest. Similarly, the mortality results based on Jerrett et
al. (2009) effect estimates are shifted to the right of the mortality results based on Zanobetti and
Schwartz (2008) and Smith et al. (2009) because Jerrett et al. (2009) results use the seasonal
mean of the daily maximum 1-hr, which tends to be higher than the seasonal mean of the daily
maximum 8-hr and seasonal mean of the daily 8-hr average metrics (see Figure 4-17). For all
three epidemiology studies, we find that 90-95% of Os-related deaths occur in locations where
the May to September mean of the daily maximum 8-hr average concentration, June to August
mean of the daily 8-hr average (10am-6pm), or April to September mean of the daily maximum
1-hr Os concentrations are greater than 40 ppb. A seasonal mean concentration of 40 ppb
corresponds to 2006-2008 design values ranging from approximately 50 to 90 ppb, depending on
the seasonal mean concentration metric (see Figure 4-18).
100
20
80
30 40 50 60
— Results based on Smith et al. (2009) effect estimates
—Results based on Zanobetti and Schwartz (2008) effect estimates
—Results based on Jerrett et al. (2009) effect estimates
Figure 8-9. Cumulative Percentage of Total Os deaths by Baseline Os Concentration.
Ozone Concentrations are Reported as May-September Mean of the Daily Maximum 8-hr
Average Concentration for Results based on Smith et al. (2009) Effect Estimates, June-
August Mean of the Daily 8-hr Average (10am to 6pm) for Results based on Zanobetti and
Schwartz (2008) Effect Estimates, and April-September Mean of the Daily Maximum 1-hr
Concentration for Results based on Jerrett et al. (2009) Effect Estimates.
8-14
-------
8.1.3 Sensitivity Analysis
For the results presented above, the national average effect estimate for results based on
Smith et al. (2009) and Zanobetti and Schwartz (2008) was applied to all grid cells between the
cities included in the studies. However, Os-mortality effect estimates have been shown to exhibit
significant regional variability across the U.S. (e.g. Smith et al. 2009). Smith et al. (2009) found
that using the national average effect estimate may overestimate risk in cities that have low effect
estimates, including Los Angeles and Denver, but may underestimate risk in cities that have high
effect estimates, including New York City and Chicago. We conduct two sensitivity analyses
aimed at characterizing the sensitivity of estimated Os-attributable premature deaths to the use of
national average effect estimates between the cities that were included in these two studies.
First, we examine the sensitivity of estimated Cb-attributable premature deaths to the
application of the 5th highest and 5th lowest effect estimates of all the cities included in the Smith
et al. (2009) and Zanobetti and Schwartz (2008) studies to the grid cells between the cities
included in these studies (Table 8-3). As in the main results, city-specific effect estimates are
applied to the grid cells in which the cities lie. Applying the 5th highest effect estimate from Las
Vegas to the grid cells between the cities included by Smith et al. (2009) yields a 36% lower
estimate of Os-attributable deaths as compared with the main results. Applying the 5th lowest
effect estimate from Dallas/Ft. Worth yields a 42% higher estimate of Os-attributable deaths as
compared with the main results. Applying the 5th lowest effect estimate from Los Angeles to the
grid cells between the cities included by Zanobetti and Schwartz (2008) yields a 37% lower
estimate of Os-attributable deaths as compared with the main results. Applying the 5th highest
effect estimate from Columbus, OH, yields a 30% higher estimate of Os-attributable deaths as
compared with the main results.
Second, we examine the sensitivity of estimated Os-attributable premature deaths to the
application of Smith et al. (2009) Bayesian-shrunken city-specific estimates using regional
average priors rather than the national average prior (Table 8-4). For grid cells between the cities
included by Smith et al. (2009), we apply the regional average effect estimate, rather than the
national average effect estimate as in the main results. Regional definitions are shown in Figure
8-10. Estimated Os-attributable deaths using the regional prior city-specific effect estimates and
the regional average effect estimates between the 98 cities included by Smith et al. (2009) are
approximately 20% larger than the main results, with 38% of estimated deaths occurring in the
98 cities rather than 43%. The 95% confidence interval for the results using the regional prior
spans zero, whereas the 95% confidence interval for the results using the national prior does not.
Since the regional average effect estimates are all based on fewer data points (in some regions,
the regional average is based on only seven cities; see Appendix 8A) than is the national average,
the confidence interval for each regional average effect estimate is large and sometimes spans
8-15
-------
zero. The large confidence intervals for the regional average effect estimates drive the
confidence interval that spans zero for Os-attributable mortality estimated using regional prior
effect estimates. Confidence intervals that span zero do not imply that higher Os is associated
with decreased mortality, as there is no biologically plausible mechanism for such an effect, and
in no case do we see a significant negative central estimate. Rather, confidence intervals
spanning zero indicate a lack of statistical power to precisely determine the magnitude of an
effect.
Table 8-3. Sensitivity of Estimated Os-attributable Premature Deaths to the Application of
the 5th lowest and 5th highest City-specific Risk estimates found by Smith et al. (2009) and
Zanobetti and Schwartz (2008) to the Grid Cells that do not correspond to the Cities
Included in Those Studies.
Source of Risk Estimate and
Sensitivity Investigated
Smith et al. (2009), May-Sep
5th lowest beta
5th highest city beta
Zanobetti & Schwartz (2008),
Jun-Aug
5th lowest city beta
5th highest city beta
Beta Used and City
0.00014
Las Vegas, NV
0.000538
Dallas/Ft. Worth, TX
0.000274
Los Angeles, CA
0.000739
Columbus, OH
Os-attributable
Mortality
(95% confidence
interval)
9,600
(-20,000 - 38,000)
21,000
(-1000-43,000)
9,800
(-3,700 - 23,000)
20,000
(100-40,000)
Percent Change
from Main
Results
-36%
+42%
-37%
+30%
Table 8-4. Sensitivity of Estimated Os-attributable Premature Deaths to the Application of
Smith et al. (2009) Regional-prior Bayes-shrunken City-specific and Regional-average
Effect Estimates, as Compared with the National-prior Bayes-shrunken City-specific and
National-average Effect Estimates as Done for the Main Results.
Risk Estimate
City-specific, national-prior
with national-average
City-specific, regional-prior
with regional-average
Os-attributable Premature Deaths
15,000
(1,400-28,000)
18,000
(-2,000-24,000)
Percent Os-attributable
Deaths in 98 cities
43%
38%
8-16
-------
Northwest
Southern
California
Upper
Midwest
Industrial
Midwest
Northeast
Southwest
Figure 8-10. Seven Regions Used in the Sensitivity Analysis based on the Smith et al.
(2009) Regional-prior Bayes-shrunken City-specific and Regional Average Effect Estimates
(Source: Samet et al., 2000).
Figure 8-11 shows estimated Os-attributable deaths by region using the national average
prior compared with using the regional average priors from Smith et al. (2009). Results generally
follow conclusions made by Smith et al. (2009) based on the magnitude of the regional effect
estimates. For example, using the national average effect estimate may substantially
underestimate Os-attributable deaths in the North East and Industrial Midwest where regional
effect estimates are large. Using the national average effect estimate may also overestimate Os-
attributable deaths in the Upper Midwest, Southern California, and South West, which were
found to have small effect estimates. However, these three regions have very large confidence
intervals which all span zero, since these regional averages are based on few cities (7, 7, and 9,
respectively, compared with 26 in the South East, 19 in Industrial Midwest, 16 in North East, and
12 in North West; see Appendix 8A).
8-17
-------
Upper Midwest
Southern California
South West
South East
North West
North East
Industrial Midwest
-5000 0 5000 10000 15000
Ozone-attributable premature deaths
• national prior • regional prior
Figure 8-11. Ozone-attributable Premature Deaths by Region as Calculated by Applying
Smith et al. (2009) Regional-prior Bayes-shrunken and Regional-average Effect Estimates,
as Compared with the National-prior Bayes-shrunken and National-average Effect
Estimates as Done for the Main Results.
As in Chapter 7, we also examine the sensitivity of respiratory mortality associated with
the six month mean of the daily maximum 1-hr Os concentrations to possible low-concentration
thresholds. The Jerrett et al. (2009) mortality model that included a low-concentration threshold
at 56 ppb had the lowest log likelihood value for all models examined. However, it is unclear
whether the 56 ppb threshold model better predicts respiratory mortality compared with a linear
model for the Jerrett et al. (2009) data. Using different but valid statistical tests produced
different conclusions about the strength of the threshold model compared with the linear model.
A less stringent statistical test suggested the 56 ppb threshold model to have an improved model
fit compared to that of the linear model, but as indicated by the confidence interval, valid
threshold levels could range anywhere from 0 to 60 ppb. None of the threshold models produce
better predictions than the linear model when a more stringent statistical test was used.
We examine as a sensitivity analysis the influence of various low-concentration
thresholds on estimated Os-related premature respiratory deaths calculated using the Jerrett et al.
(2009) effect estimates. We examine a range of 40 ppb to 60 ppb thresholds in 5 ppb increments,
including also a threshold equal to 56 ppb, which had the lowest log likelihood value for all
models examined. Because the threshold model effect estimates given by Jerrett et al. (2009) did
not control for PM2.5 but our main results use the 2-pollutant model which did control for PM2.5,
we also include as a sensitivity a single pollutant no threshold model to be comparable with the
8-18
-------
single pollutant threshold models. In addition, the two-pollutant model used for the main results
was based on 86 cities, while the single pollutant models were based on 96 cities. Therefore, for
comparability to the main results, we also include as a sensitivity a single pollutant no threshold
model based on 86 cities.
When no low-concentration threshold is applied (as in the main results), we estimate
31,000 (95% confidence interval, 10,000 - 51,000) Os-attributable premature deaths using the 86
city Os-only model and 34,000 (13,000-53,000) using the 96 city Os-only model (Table 8-5).
These results are 30% and 25% fewer than the 45,000 (17,000 - 70,000) Os-attributable
premature deaths calculated using the two-pollutant model for 86 cities (used for the main
results). For the results based on the Os-only models with thresholds ranging from 40-60 ppb, we
estimate 71-98% fewer Os-attributable deaths compared with results based on the Os-only no
threshold model of 96 cities and 78-99% fewer compared with the main results. For a 56 ppb
threshold, which had the lowest log likelihood value for all models examined, we estimate 1,600
(710-2,400) Os-attributable premature deaths, 95% fewer compared with results based on the Os-
only no threshold model of 96 cities and 97% fewer than the main results.
Table 8-5. Sensitivity of Estimated Os-attributable Premature Deaths to the Application of
Jerrett et al. (2009) Effect Estimates with and without a Low Concentration Threshold.
Model
Two-pollutant
model, 86 cities
Os-only model,
86 cities
Os-only model,
96 cities
Threshold
0 ppb
0 ppb
0 ppb
40 ppb
45 ppb
50 ppb
55 ppb
56 ppb
60 ppb
Beta
0.0039221
0.0026642
0.00286
0.00312
0.00336
0.00356
0.00417
0.00432
0.00402
Os-attributable
Premature Deaths
(95% Confidence
Interval)
45,000
(16,000-70,000)
31,000
(9,500-51,000)
34 000
(13,000-53,000)
9,800
(4,000-16,000)
7,000
(2,900-11,000)
3,900
(1 ,700 - 6,200)
1,900
(840 - 2,900)
1,600
(710-2,400)
620
(210-1,000)
Percent
Change
from Main
Results
n/a
-30%
-25%
-78%
-84%
-91%
-96%
-97%
-99%
Percent
Change from
Os only, 96
cities, No
Threshold
Model
+33%
-6%
n/a
-71%
-79%
-88%
-94%
-95%
-98%
8-19
-------
8.1.4 Discussion
We estimated the total all-cause deaths associated with short-term exposure to recent Os
levels across the continental U.S., using average 2006-2008 observations from the Os monitoring
network fused with a 2007 CMAQ simulation and city-specific Os-mortality effect estimates
from two short-term epidemiology studies. Applying Smith et al. (2009) effect estimates for
May-September, we estimate 15,000 (95% CI, 1,400-28,000) premature Os-related non-
accidental deaths (all ages) annually for 2007. Using Smith et al. (2009) effect estimates, Os-
attributable mortality contributes an average of 1.5% (95% confidence interval, 1.1%-1.8%) to
county-level May-September non-accidental mortality (all ages) and 0.6% (0.4%-0.7%) to all
year all-cause mortality (all ages). As a sensitivity, we apply Zanobetti and Schwartz (2008)
effect estimates for June-August, finding 16,000 (95% CI, 6,000-25,000) premature Os-related
all-cause deaths (all ages) annually for 2007. For results using Zanobetti and Schwartz (2008)
effect estimates, Os-attributable mortality contributes an average of 2.5% (95% confidence
interval, 1.7%-3.0%) to county-level June-August all-cause mortality (all ages) and 0.6% (0.4%-
0.8%) to all year all-cause mortality (all ages). For the application of Jerrett et al. (2009) effect
estimates for April-September, we estimate 45,000 (95% CI, 17,000-70,000) premature Os-
related adult (age 30 and older) respiratory deaths. For the results using Jerrett et al. (2009) effect
estimates, Os-attributable mortality contributes an average of 18.5% (95% confidence interval,
15.2%-21.5%) to county-level April-September adult (age 30+) respiratory mortality and 1.9%
(1.3%-2.6%) to all year all-cause mortality (all ages). For all three epidemiology studies, we find
that 90-95% of Os-related deaths occur in locations where the May to September means of the
daily maximum 8-hr average concentration, June to August means of the daily 8-hr average
(10am-6pm), or April to September means of the daily maximum 1-hr Os concentrations are
greater than 40 ppb. A seasonal mean concentration of 40 ppb corresponds to 2006-2008 design
values ranging from approximately 50 to 90 ppb, depending on the seasonal mean concentration
metric.
A previous analysis estimated that short-term Os exposure was associated with 4,700
(95% CI, 1,800-7,500) premature deaths nationwide annually, based on 2005 Os concentrations
and Bell et al. (2004) national average effect estimates (Fann et al., 2012). The results estimated
here are higher, resulting mainly from two important differences. First, Fann et al. (2012)
estimated risk only above North American background, simulated Os concentrations in the
absence of North American anthropogenic emissions, which was set to 22 ppb in the east and 30
ppb in the west. Fann et al. (2012) also used a national average mortality effect estimate for daily
maximum 8-hr average Os during the warm season only, calculated using ratios of 24-hr mean
concentrations to daily maximum 8-hr average concentrations (see Abt Associates 2010). The
Smith et al. (2009) national average beta used here, 0.000322, is based on April-October Os data
8-20
-------
and is approximately 23% larger than that used by Fann et al. (2012), 0.000261. Because the risk
modeling period (and the seasonal definition for the seasonal mean of the daily maximum 8-hr
average concentration) was May to September for both studies, the higher beta used here yields a
larger Os mortality estimate. These two differences in methods explain the larger Os mortality
estimates of this analysis compared with the previous estimate by Fann et al. (2012).
Estimated Os-attributable premature deaths based on Jerrett et al. (2009) effect estimates
are approximately three times larger than results based on Smith et al. (2009) and Zanobetti and
Schwartz (2008) effect estimates. The mean estimated county-level percent of all-cause, all-year,
and all-age mortality is also three times larger for results based on Jerrett et al. (2009) effect
estimates, indicating that the larger estimate does not simply result from a longer modeling
period or different population subset (e.g. adult respiratory disease for Jerrett et al. (2009) effect
estimates versus all-age non-accidental or all-cause mortality for Smith et al. (2009) and
Zanobetti and Schwartz (2008) effect estimates). Recent studies using long-term Os-mortality
relationships found by Jerrett et al. (2009) to quantify the burden of mortality due to
anthropogenic Os globally (Anenberg et al., 2010, 2011) and for the U.S. specifically (Fann et
al., 2012) have also found that using Jerrett et al. (2009) long-term effect estimates yields Os-
related mortality burden estimates that are approximately two to four times larger than estimates
based on short-term effect estimates. Since long-term mortality relationships include both acute
and chronic exposure effects, the significantly larger mortality estimates calculated using long-
term concentration-mortality relationships suggest that considering only short-term mortality
may exclude a substantial portion of Os-related risk. However, since the short-term mortality
relationships include a larger population (all ages versus adults ages 30 and older only) and all
mortality causes, the short-term mortality relationships may capture some Os effects that are not
captured by Jerrett et al. (2009). It is likely that some portion of the estimated premature deaths
attributable to short-term Os exposure is captured by estimated premature deaths attributable to
long-term Os exposure, but the extent of the overlap between these estimates is unknown.
8.2 EVALUATING THE REPRESENTATIVENESS OF THE URBAN STUDY
AREAS IN A NATIONAL CONTEXT
To further support interpretation of risk estimates generated in Section 7.2, this section
presents three analyses that assess the representativeness of the 12 urban study areas in the
national context. First, we assess the degree to which the 12 urban study areas represent the
range of air quality levels and key Os risk-related attributes that vary spatially across the nation.
We have partially addressed this issue by selecting urban study areas in different geographical
regions of the country (see Section 7.2). In this section, we evaluate how well the selected urban
areas represent the overall U.S. for a set of spatially-distributed Os risk related variables (e.g.
8-21
-------
weather, demographics including socioeconomic status, baseline health incidence rates; Section
8.2.1). Section 8.2.2 identifies where our 12 urban study areas fall along the distribution of Os-
attributable mortality risk across the U.S. This analysis allows us to assess the degree to which
the 12 urban study areas capture locations within the U.S. likely to experience elevated levels of
risk related to ambient Os. Finally, we give a national context to the estimated Os responses to
emission changes in the urban study areas by assessing how well these 12 areas and the 3
additional exposure assessment study areas represent air quality trends and responses to
emissions across the entire U.S. (Section 8.2.3).
We do not attempt to assess the representativeness of the 15 urban study areas considered
in the exposure assessment for Os related risk because data limitations preclude us from being
able to characterize individual-level exposure across the U.S. However, the urban study areas
considered in both the exposure and risk assessments shared common selection criteria,
including consideration of Os concentrations, availability of adequate monitoring data,
demographics, and exposure factors. Therefore, conclusions from this analysis of the
representativeness of the 12 urban study areas for risk would also apply to those areas for
exposure.
8.2.1 Analysis Based on Consideration of National Distributions of Risk-Related
Attributes
This section evaluates how well the urban study areas reflect national-level variability in
a series of Os risk-related variables. For this analysis, we first generate distributions for risk-
related variables across the U.S. and for the specific urban study areas considered in Section 7.2
from generally available data (e.g. from the 2000 Census, Centers for Disease Control (CDC), or
other sources). We then plot the specific values of these variables for the selected urban study
areas on these distributions, and evaluate how representative the selected study areas are of the
national distributions for these individual variables.
Estimates of risk (either relative or absolute, e.g., number of cases) within our risk
assessment framework are based on four elements: population, baseline incidence rates, air
quality, and the coefficient relating air quality and the health outcome (i.e. the Os effect
estimates). Each of these elements can contribute to heterogeneity in risk within and across urban
study areas. In addition, there may be other identifiable factors that contribute to the variability
of the four elements across urban study areas. In this assessment, we examine the
representativeness of the selected urban study areas for the four main elements, as well as factors
that have been identified as influential in determining the magnitude of the C-R function across
urban study areas.
8-22
-------
While personal exposure is not incorporated directly into Os epidemiology studies, city-
specific Os effect estimates are affected by differing levels of exposure which in turn are related
to variability in exposure determinants. The correlation between monitored Os and personal Os
exposure also varies between cities. The Os ISA has comprehensively reviewed epidemiological
and toxicological studies to identify variables which may affect the Os effect estimates used in
the city-specific risk analysis in Section 7.2 and the national-scale risk analysis in Section 8.1
(U.S. EPA 2013). Determinants of the Os effect estimates used in risk assessment can be grouped
into four broad areas:
• Demographics: education, income, age, unemployment rates, race, body mass index and
physical conditioning, public transportation use, and time spent outdoors.
• Baseline health conditions: asthma, chronic obstructive pulmonary disease,
cardiovascular disease (atherosclerosis, congestive heart disease, atrial fibrillation,
stroke), diabetes, inflammatory diseases, and smoking prevalence.
• Climate and air quality: Os levels, co-pollutant levels (annual mean PIVb.s), temperatures
(days above 90 degrees, mean summer temp, 98th percentile temp).
• Exposure determinants: air conditioning prevalence.
Although data limitations preclude our ability to conduct a national-scale exposure
assessment as we have done for Os-attributable risk in Section 8.1, we assess the
representativeness of the urban study areas across the national distribution of climate, air quality,
and air conditioning prevalence, factors which influence individual exposure. As discussed in
detail in Chapter 5, no available data base is sufficient to assess the national representativeness of
time spent outdoors, another important personal exposure determinant, among persons residing
in each of the urban study areas. However, previous analyses suggest that children's time spent
outdoors varies little across U.S. regions (section 8.10.2 of U.S. EPA, 2009). In addition, as
discussed in Section 5.4.1, time spent outdoors and the percent of person-days having at least one
minute outdoors (participation rate) does not appear to vary much over the past few decades
based on analyses using the CHAD database, nor does there appear to be a temporal trend over
the past decade based on analyses using the American Time Use Survey (ATUS). In considering
that many of the activity pattern studies in CHAD were from national surveys conducted in
metropolitan areas and that the evaluation results indicate little difference in time expenditure
over broad geographic areas and survey collection years, it is likely that the distribution of time
spent outdoors generated for the simulated persons in the 15 urban study areas (Chapter 5)
reasonably reflects the most important elements of a national distribution of time spent outdoors.
In addition, we were unable to assess the representativeness of the urban study areas in
terms of fine-scale spatial variability in exposure, including near-roadway gradients. Cities with
large populations living along highly trafficked roadways are expected to have higher Os
-------
exposure rates than cities with greater geographical separation between busy roadways and
residential areas. Differences in near-roadway and other highly local exposure patterns contribute
to heterogeneity among city-specific C-R functions, which are directly compared for this
representativeness analysis. However, we are unable to make specific conclusions as to the
differences in near-roadway and other highly local exposures among the urban study areas and
between the urban study areas and the national distribution.
Based on these identified potential risk determinants, we identify datasets that could be
used to generate nationally representative distributions for each parameter. We are not able to
identify readily available national datasets for all variables. In these cases, if we are able to
identify a broad enough dataset covering a large enough portion of the U.S., we use that dataset
to generate the parameter distribution. In addition, we are not able to find exact matches for all of
the variables identified through our review of the literature. In cases where an exact match is not
available, we identify proxy variables to serve as surrogates. For each parameter, we report the
source of the dataset, its degree of coverage, and whether it is a direct measure of the parameter
or a proxy measure (Table 8-6). Summary statistics for the most relevant variables are provided
in Table 8-7.
8-24
-------
Table 8-6. Data Sources for Os Risk-related Attributes.
Potential Risk-
related Attribute
Attribute Metric
Year
Source
Degree of
National
Coverage
Demographics
Age
Age
Age
Education
Unemployment
Income
Race
Population
Population density
Urbanicity
Percent age 85 years
and older
Percent age 65 years
and older
Percent age 14 years
and younger
Population with less
than high school
diploma
Percent unemployed
Per capita personal
income
Percent nonwhite
Total population
Population/square mile
ERS Classification
Code
2005
2005
2005
2000
2005
2005
2006
2008
2008
2003
County Characteristics, 2000-2007 Inter-university
Consortium for Political and Social Research
County Characteristics, 2000-2007 Inter-university
Consortium for Political and Social Research
County Characteristics, 2000-2007 Inter-university
Consortium for Political and Social Research
USDA/ERS, http://www.ers.usda.gov/Data/Education/
County Characteristics, 2000-2007 Inter-university
Consortium for Political and Social Research
County Characteristics, 2000-2007 Inter-university
Consortium for Political and Social Research
County Characteristics, 2000-2007 Inter-university
Consortium for Political and Social Research
Cumulative Estimates of Resident Population Change for the
United States, States, Counties, Puerto Rico, and Puerto
Rico Municipios: April 1 , 2000 to July 1 , 2008, Source:
Population Division, U.S. Census Bureau
Cumulative Estimates of Resident Population Change for the
United States, States, Counties, Puerto Rico, and Puerto
Rico Municipios: April 1 , 2000 to July 1 , 2008, Source:
Population Division, U.S. Census Bureau (calculated "as the
crow flies")
County Characteristics, 2000-2007 Inter-university
Consortium for Political and Social Research
All counties
All counties
All counties
All counties
All counties
All counties
All counties
All counties
All counties
All counties
8-25
-------
Potential Risk-
related Attribute
Attribute Metric
Year
Source
Degree of
National
Coverage
Climate and Air Quality
Os levels
Os levels
Os levels
Os levels
Plvh .5 levels
Temperature
Relative Humidity
Monitored 4th high daily
maximum 8-hr average
Seasonal mean of the
daily maximum 8-hr
average
Seasonal mean of the
daily maximum 1-hr
Seasonal mean
Monitored annual mean
Mean July temp
Mean July RH
2007
Avg. 2006-
2008
Avg. 2006-
2008
Avg. 2006-
2008
2007
1941-1970
1941-1970
EPA Air Quality System (AQS)
AQS
AQS
AQS
AQS
County Characteristics, 2000-2007 Inter-university
Consortium for Political and Social Research
County Characteristics, 2000-2007 Inter-university
Consortium for Political and Social Research
725 Monitored
counties
671 Monitored
counties
671 Monitored
counties
671 Monitored
counties
617 Monitored
counties
All counties
All counties
Exposure Determinants
Ventilation
Percent residences with
no air conditioning
2004
American Housing Survey
76 cities
Baseline Health Conditions
Baseline mortality
Baseline mortality
Baseline mortality
Baseline mortality
Baseline morbidity
Baseline morbidity
Baseline morbidity
All Cause
Non Accidental
Cardiovascular
Respiratory
Acute myocardial
infarction prevalence
Diabetes prevalence
Stroke prevalence
2007
2007
2007
CDC Wonder 1999-2005
CDC Wonder 1999-2006
CDC Wonder 1999-2007
CDC Wonder 1999-2008
Behavioral Risk Factor Surveillance System (BRFSS)
BRFSS
BRFSS
All counties
All counties
All counties
All counties
184
metropolitan
statistical areas
(MSA)
184 MSA
184 MSA
8-26
-------
Potential Risk-
related Attribute
Baseline morbidity
Obesity
Level of exercise
Level of exercise
Respiratory risk
factors
Smoking
Attribute Metric
Congestive heart
disease prevalence
Body Mass Index
Vigorous activity 20
minutes
Moderate activity 30
minutes or vigorous
activity 20 minutes
Current asthma
Ever smoked
Year
2007
2007
2007
2007
2007
2007
Source
BRFSS
BRFSS
BRFSS
BRFSS
BRFSS
BRFSS
Degree of
National
Coverage
184 MSA
184 MSA
184 MSA
184 MSA
184 MSA
184 MSA
C-R Estimates
Mortality risk
Mortality risk
Mortality risk
Mortality risk
Non Accidental
All Cause
Cardiovascular
Respiratory
2009
2008
2008
2008
Smith et al. (2009)
Zanobetti and Schwartz (2008)
Zanobetti and Schwartz (2008)
Zanobetti and Schwartz (2008)
98 cities
48 cities
48 cities
48 cities
8-27
-------
Table 8-7. Summary Statistics for Selected Oa Risk-related Attributes.
Risk-related Attribute
Demographics
Population per county
Population density (Pop/sq mile)
Median age (Years)
% Age 0 to 14 years
% Age 65+ years
% Age 85+ years
Unemployment rate (%)
% with less than high school
diploma
Income per capita ($)
% Non-white
% Commute by public
transportation*
Health Conditions
Prevalence of CHD (%) *
Prevalence of asthma (%) *
Prevalence of diabetes (%) *
Prevalence of AM I (%) *
Prevalence of obesity (%) *
Prevalence of stroke (%) *
Prevalence of ever smoked (%)*
Prevalence of exercise (20
minutes, %)*
Prevalence of exercise (30
minutes, %)*
Non-accidental mortality (deaths
per 100, 000 people)
Average
Urban
Study U.S.
Areas Dataset
1,642,198 97,020
10,378 258
35.7 38.6
20.7 19.0
11.3 14.9
1.7 2.1
5.7 5.4
20.9 22.6
40305 27367
36.4 13.0
7.1 1.6
3.6 4.3
8.5 8.1
8.1 8.5
3.6 4.1
24.7 26.0
2.6 2.7
18.3 19.6
29.5 28.0
50.2 49.7
756.2 950.6
Standard Deviation
Urban
Study U.S.
Areas Dataset
1,972,403 312,348
16,550 1,757
2.3 4.4
2.4 2.9
2.5 4.1
0.6 0.9
1.2 1.8
7.9 8.8
14238 6604
15.3 16.2
8.1 2.5
0.8 1.3
1.3 1.9
1.2 2.1
0.6 1.3
4.0 4.1
0.7 1.0
3.1 4.0
2.7 4.8
2.3 5.4
204.1 249.6
Maximum
Urban
Study U.S.
Areas Dataset
9,862,049 9,862,049
71,758 71,758
40.0 55.3
24.6 36.8
15.2 34.7
2.5 7.7
8.6 20.9
37.7 65.3
93377 93377
86.7 95.3
30.7 30.7
4.6 8.7
11.2 13.2
10.6 16.5
4.8 10.2
32.7 35.7
3.7 6.5
23.1 34.4
33.8 44.1
55.3 67.1
1139.5 1958.4
Minimum
Urban
Study U.S.
Areas Dataset
354,361 42
1,313 0
32.1 20.1
14.7 0.0
5.8 2.3
0.5 0.1
4.1 1.9
8.7 3.0
23513 5148
31.7 0.0
1.5 0.0
2.6 1.8
6.0 3.6
5.4 2.2
2.8 1.7
18.7 14.0
1.5 0.7
14.2 6.5
23.7 15.4
47.4 37.3
361.6 117.7
Sample Size
(# of counties
or cities)
Urban
Study U.S.
Areas Dataset
23 3143
23 3143
23 3141
23 3141
23 3141
23 3141
23 3133
23 3141
23 3086
23 3141
12 366
11 184
11 184
11 184
11 184
11 182
11 184
11 184
11 183
11 182
23 3142
8-28
-------
Risk-related Attribute
All cause mortality (deaths per
100,000 people)
Cardiovascular mortality (deaths
per 100, 000 people)
Respiratory mortality (deaths per
100,000 people)
Air Quality and Climate
03 4th high daily maximum 8-hr
average (ppm)
Os seasonal mean (ppb)
Os seasonal mean of daily
maximum 8-hr average (ppb)
Os seasonal mean of daily
maximum1-hr (ppb)
PM25 annual mean (ug/m3)
Plvh .5 98th %ile daily average
(ug/m3)
Average temperature (°F)
July temperature long term
average (°F)
July Relative Humidity long term
average (%)
Exposure Determinants
% No air conditioning*
C-R Estimates
Non-accidental mortality Os risk*
All Cause mortality Os risk*
Respiratory mortality Os risk*
Cardiovascular mortality Os risk*
Average
Urban
Study U.S.
Areas Dataset
810.1 1022.3
310.5 392.1
66.2 97.3
0.087 0.077
33.9 34.5
50.7 48.6
58.8 54.7
14.1 11.7
35.8 30.7
57.2 57.2
76.0 75.9
61.5 56.2
15.5 16.6
0.000388 0.000322
0.000627 0.000527
0.000877 0.000800
0.000898 0.000825
Standard Deviation
Urban
Study U.S.
Areas Dataset
217.4 258.6
93.9 121.0
17.0 32.3
0.009 0.010
5.4 6.6
7.5 7.2
7.5 8.0
2.6 3.1
8.1 9.3
5.0 7.9
3.4 5.4
10.2 14.6
85.7 79.1
0.000217 0.000131
0.000314 0.000205
0.000282 0.000186
0.000173 0.000124
Maximum
Urban
Study U.S.
Areas Dataset
1257.8 2064.2
459.6 970.4
90.1 351.0
0.105 0.126
51.0 64.8
70.2 79.7
85.1 92.4
16.9 22.5
59.0 81.1
70.3 76.2
83.3 93.7
70.0 80.0
42.9 86.7
0.000917 0.000917
0.001092 0.001092
0.001424 0.001424
0.001064 0.001064
Minimum
Urban
Study U.S.
Areas Dataset
402.5 176.8
122.4 37.5
34.8 13.3
0.072 0.033
25.8 8.6
40.8 13.3
46.5 17.6
8.4 3.4
21.2 9.1
50.1 39.0
68.5 55.5
28.0 14.0
0.4 0.0
0.000148 -0.000033
0.000163 0.000096
0.000307 0.000307
0.000418 0.000418
Sample Size
(# of counties
or cities)
Urban
Study U.S.
Areas Dataset
23 3142
23 3142
23 3136
23 725
22 671
22 671
22 671
23 617
23 617
23 202
23 3104
23 3104
12 76
12 98
12 48
12 48
12 48
*Attribute for which only city-specific data were available.
8-29
-------
Figure 8-12 through Figure 8-18 show national-level cumulative distribution functions
(CDF) for the four critical risk function elements (population, air quality, baseline incidence, and
the Os effect estimate), as well as where the urban study areas fall on the distribution. While the
urban-scale analysis in Chapter 7 includes the full core-based statistical area for the selected
cities, we consider here only the counties included in each city as defined by the epidemiological
studies, since we only have information on Os effect estimates for these counties. This approach
is consistent with the national-scale assessment of Os-attributable risk in Section 8.1, from which
we draw county-level Os-attributable risk estimates for the representativeness analysis in Section
8.2.3. These figures focus on critical variables representing each type of risk determinant, e.g. we
focus on all-cause and non-accidental mortality rates, but we also have conducted analyses for
cardiovascular and respiratory mortality separately. The vertical black lines in each graph show
the values of the variables for the individual urban study areas. The city-specific values that
comprise the national CDF for mortality risks found by Zanobetti and Schwartz (2008) are also
displayed on the graphs of those attributes, as the number of cities included in that study is
smaller (48 cities). The complete analysis is provided in Appendix 4A.
These figures show that the selected urban study areas represent the upper percentiles of
the distributions of population and do not represent the locations with lower populations (urban
study areas are all above the 90th percentile of U.S. county populations). This is consistent with
the objectives of our case study selection process, e.g. we are characterizing risk in areas that are
likely to be experiencing excess risk due to Os levels above alternative standards. The urban
study areas span the full range of seasonal means of the daily maximum 8-hr average Os
concentrations in monitored U.S. counties and the full distribution of Os risk coefficients across
the cities included by Smith et al. (2009) and Zanobetti and Schwartz (2008). The urban study
area analysis includes the two cities with the highest risk coefficients found by Smith et al.
(2009) - New York City and Philadelphia - as well as the two highest found by Zanobetti and
Schwartz (2008) - New York City and Detroit. In Table 8-7, respiratory and cardiovascular
mortality have higher C-R relationships than non-accidental and all-cause mortality because they
are based on a smaller baseline population and are the diseases most affected by Os exposure.
The urban study areas do not capture the upper end of the distribution of baseline mortality,
including all-cause (Figure 8-15) and non-accidental mortality (Figure 8-16), as well as
cardiovascular and respiratory mortality (see Appendix 8B). The interpretation of this is that the
case study risk estimates may not capture the additional risk that may exist in locations that have
the highest baseline mortality rates.
8-30
-------
100% n
90%
3
|
I
D
*O
Urban case study areas
are all above the 90th
percentile of county
populations
100
1000 10000 100000 1000000
Population, 2008
10000000
• All Counties CDF
• Case Studv Counties
Figure 8-12. Comparison of County-level Populations of Urban Study Area Counties to the
Frequency Distribution of Population in 3,143 U.S. Counties.
30 40 50 60 70
Seasonal Mean 8-hr Daily Max Ozone Concentration, Average 2006-2008
(ppb)
-All Counties CDF
Case Study Counties
Figure 8-13. Comparison of County-level Seasonal Means of the Daily Maximum 8-hr
Average Os Concentrations in Urban Study Area Counties to the Frequency Distribution of
Seasonal Mean of the Daily Maximum 8-hr Os Concentrations in 671 U.S. Counties with Os
Monitors.
8-31
-------
50
60 70 80 90 100
4th High 8-hr Daily Maximum Ozone, 2007 (ppb)
110
• AII Counties CDF
Case Study Counties
Figure 8-14. Comparison of 2007 County-level 4th highest Daily Maximum 8-hr Average
Os Concentrations in Urban Study Area Counties to the Frequency Distribution of 2007 4th
highest Daily Maximum 8-hr Average Os Concentrations in 725 U.S. Counties with Os
Monitors.
100% •
90%
80%
V)
.2 70%
P
g 60%
O
/
1/1
.X
Urban case
study areas are
all below the
85th percentile
of county all
cause mortality
500
600 700 800 900 1000 1100 1200 1300 1400
All Cause Mortality per 100,000 Population, 1S99-2005
1500
• AII Counties CDF
Case Study Counties
Figure 8-15. Comparison of County-level all-cause Mortality in Urban Study Area
Counties to the Frequency Distribution of all-cause Mortality in 3,137 U.S. Counties.
8-32
-------
100%
90%
80%
| 70% -
c
g 60% -
u
>« 50%
3
*S 40%
3?
30%
20%
10% -
n%
——\
i -i
— -*
l
J
*x^
1
1 1
1 1
-
l
.X
l
>
/^
1 1
,
.J
X
; . i
^x — "
Urban case study
counties are all
below the 80th
percent! le of non
accidental
mortality
300 500 700 900 1100 1300 1500
Non Accidental Mortality per 100,000 Population, 1999-2005
All Counties CDF • Case Study Counties
Figure 8-16. Comparison of County-level non-accidental Mortality in Urban Study Area
Counties to the Frequency Distribution of non-accidental Mortality in 3,135 U.S. Counties
0.0004
0.0006 0.0008 0.001
All Cause Mortality Risk Coefficient ((3)
0.0012
All cities
-All Cities CDF • Case Study Cities
Figure 8-17. Comparison of City-level all-cause Mortality Risk Coefficients from Zanobetti
and Schwartz (2008) in Urban Study Areas to the Frequency Distribution of all-cause
Mortality Risk Coefficients from Zanobetti and Schwartz (2008) in 48 U.S. Cities.
8-33
-------
AUV/0
90%
80% -
3
5 70% -
£ 60%
| 50% -
* 40% -
0
* 30% -
20% -
10% -
0%
/
•""
/
/
/
/
/
/
^
0.0002 0.0004 0.0006 0.0008
Non Accidental Mortality Risk Coefficient (p)
All Cities CDF •Case Study Cities
0.001
Figure 8-18. Comparison of City-level National-prior Bayes-shrunken non-accidental
Mortality Risk Coefficients from Smith et al. (2009) in Urban Study Areas to the
Frequency Distribution of National-prior Bayes-shrunken non-accidental Mortality Risk
Coefficients from Smith et al. (2009) in 98 U.S. Cities
Figure 8-19 through Figure 8-24 show national CDFs and the urban study area values for
several selected potential risk attributes. These potential risk attributes do not directly enter the
risk equations, but have been identified in the literature as potentially affecting the magnitude of
the Os C-R functions reported in the epidemiological literature. Comparison graphs for other risk
attributes are provided in Appendix 4A. The selected urban study areas do not capture the higher
end percentiles of several risk characteristics, including populations 65 years and older, baseline
cardiovascular disease prevalence, baseline respiratory disease prevalence, and smoking
prevalence. Summarizing the analyses of the other risk attributes, we conclude that the urban
study areas provide adequate coverage across population, population density, Os levels (seasonal
mean, seasonal means of the daily maximum 8-hr, and seasonal means of the daily maximum 1-
hr), PM2.5 co-pollutant levels, temperature and relative humidity, unemployment rates, percent
non-white population, asthma prevalence obesity prevalence, income, and less than high school
education. We also conclude that while the urban study areas cover a wide portion of the
distributions, they do not provide coverage for the upper end of the distributions of percent of
population 65 and older (below 60th percentile), percent of population 85 years and older (below
75th percentile), prevalence of angina/coronary heart disease (below 70th percentile), prevalence
of diabetes (below 85th percentile), stroke prevalence (below 90th percentile), prevalence of heart
attack (below 80th percentile), prevalence of smoking (below 85th percentile), all-cause mortality
8-34
-------
rates (below 85th percentile), non-accidental mortality rates (below 80th percentile),
cardiovascular mortality rates (below 75th percentile) and respiratory mortality rates (below 50th
percentile), and percent of residences without air conditioning (below 90th percentile). In
addition, the urban study areas do not capture the highest or lowest ends of the distribution of
exercise prevalence and do not capture the low end of the distribution of public transportation
use (above the 65th percentile).
100%
90%
80%
| 70%
E
o 60%
"! 50%
"o 40%
30%
20%
10%
14
16
IS 20 22 24 26
% Younger than 15 Years Old, 2005
28
30
• AII Counties CDF
Case Study Counties
Figure 8-19. Comparison of County-level Percent of Population 0 to 14 years old in Urban
Study Area Counties to the Frequency Distribution of Percent of Population 0 to 14 years
old in 3,141 U.S. Counties.
8-35
-------
100% -
90% -
80% -
.2 70%
B
§ 60%
u
«/i 50%
^j
«S 40% -
30%
20% -
10% -
n%
- -1
.— r-!
1
1
'
1
1
i I
/
1
1
Urban case study
counties are all
belowthetttth
percentile of county
% of population 65
years and older
11 13 15 17 19 21
% 65 Years and Older, 2005
23
25
27
• All Counties CDF
Case Study Counties
Figure 8-20. Comparison of County-level Percent of Population age 65 years old and older
in Urban Study Area Counties to the Frequency Distribution of Percent of Population age
65 and older in 3,141 U.S. Counties.
o%
10000 20000 30000 40000 50000 60000 70000 80000 90000 100000
Income per capita, 2005 ($)
• All Counties CDF
Case Study Counties
Figure 8-21. Comparison of County-level Income per capita in Urban Study Areas to the
Frequency Distribution of Income per capita in 3,141 U.S. Counties.
8-36
-------
100% -
90%
80% -
3 70% -
§ 60% -
rf 50% -
* 40%
30% -
20% -
10%
0%
66 68 70 72 74 76 78 80 82 84 86
Mean Temperature for July, 1941-1970 (F)
AII Counties CDF • Case Study Counties
Figure 8-22. Comparison of County-level July Temperature in Urban Study Area Counties
to the Frequency Distribution of July Temperature in all U.S. Counties.
Comparison of Urban Case Study Area with U.S. Distribution (184 BFRSS
Cities) - Asthma Prevalence
100%
90% -
80%
70% -
„ 60%
'& 50% -
•s
vO "rUTo
3?
30%
20% -
10% -
MA
^^^
*****^
8 10 12
Asthma Prevalence, 2007 (%)
14
-All Cities CDF
Case Study Cities
Figure 8-23. Comparison of City-level Asthma Prevalence in Urban Study Areas to the
Frequency Distribution of Asthma Prevalence in 184 U.S. Cities.
8-37
-------
100% -
90% -
80%
70% -
« 60% -
™
" 50%
o
* 40%
30%
20% -
10% V
n% -
y
I
i
i
i
^-^
^^
i i
.^
.
|Xx«— -"
Urban study areas
areallbelowthe
90th percentileof
percent of
residences with no
air conditioning
10
20 30 40 50 60 70
No air conditioning, 2004 (%)
80
90
100
-All Cities CDF
CaseStudvCrties
Figure 8-24. Comparison of City-level Air Conditioning Prevalence in Urban Study Areas
to the Frequency Distribution of Air Conditioning Prevalence in 76 U.S. Cities.
Based on the above analyses, we can draw several inferences regarding the
representativeness of the urban study areas. First, the selected urban study areas are among the
most populated in the U.S. Second, they represent urban areas with relatively high levels of Os
(4th highest daily maximum 8-hr average, seasonal mean of the daily maximum 8-hr average,
seasonal mean of the daily maximum 1-hr, and seasonal mean). Third, they capture well the range
of city-specific effect estimates found by Smith et al. (2009) and Zanobetti and Schwartz (2008)
studies. These three factors would suggest that the urban study areas should capture well overall
risk for the nation, with a potential for better characterization of the high end of the risk
distribution.
However, there are several other factors that suggest that the urban study areas may not
be representing areas that may have a high risk per ppb of Os. Several of the factors with
underrepresented tails, including age and baseline mortality are spatially correlated (R=0.81), so
that certain counties which have high proportions of older adults also have high baseline
mortality and high prevalence of underlying chronic health conditions. Because of this, omission
of certain urban areas with higher percentages of older populations, for example, cities in
Florida, may lead to underrepresentation of high risk populations. However, with the exception
of areas in Florida, most locations with high percentages of older populations have low overall
populations, less than 50,000 people in a county. And even in Florida, the counties with the
highest Os levels do not have a high percent of older populations. This suggests that while the
8-38
-------
risk per exposed person per ppb of Os may be higher in these locations, the overall risk to the
population is likely to be within the range of risks represented by the urban study locations.
Due to data limitations, we were only able to assess the representativeness of the urban study
areas in terms of one exposure-related attribute, air conditioning prevalence. Assessing the
representativeness of the urban study areas in terms of air conditioning prevalence, we found that
the urban study areas do not capture the highest end of percent of residences without air
conditioning. If the cities with the lowest air conditioning prevalence also have high Os levels,
we could be missing a high risk portion of the population that is exposed to Os indoors as air
infiltrates indoors from outdoors. However, 4th highest daily maximum 8-hr averageOs levels in
the cities in the top 10th percentile of percentage of residences without air conditioning (mainly
in northern California and Washington) are approximately average (0.08 ppm) or lower than
average. Since these concentrations are not the highest found across the U.S., we are likely not
excluding a high risk population that has both low air conditioning prevalence and high Os
concentrations, and the overall risk to the population is likely to be within the range of risks
represented by the urban study areas.
8.2.2 Analysis Based on Consideration of National Distribution of Ozone-Related
Mortality Risk
In this section we discuss the second representativeness analysis which identifies where
the 12 urban study areas examined in Chapter 7 fall along the distribution of estimated national-
scale mortality risk. This assessment reveals whether the baseline Os mortality risks in the 12
urban study areas represent more typical or higher end risk relative to the national risk
distribution presented in Section 8.1. For consistency, we compare the national Os mortality risk
distribution to the Os mortality risk results for the urban study areas that were generated from the
national-scale assessment in Section 8.1, rather than the results from the urban study area
analysis in Chapter 7 which uses different methods. To be consistent with the national-scale
assessment, we define the urban study areas here as they were defined in the epidemiology
studies, rather than including full core-based statistical areas as in Chapter 7. The results of this
representativeness analysis are presented graphically in Figure 8-25 through Figure 8-27, which
display the cumulative distribution of total mortality attributable to ambient Os at the county
level developed as part of the national-scale analysis. Values for the 23 counties included in the
urban study areas as defined in the epidemiology studies are then superimposed on top of the
cumulative distribution to assess the representativeness of the urban study areas.
For the results based on Smith et al. (2009) effect estimates, New York City and
Philadelphia have the highest percentage of May-September non-accidental mortality attributable
to ambient Os of the 12 urban study areas and are located at the highest end of the distribution of
8-39
-------
U.S. Os-related mortality risk (Figure 8-25). Of the 12 urban study areas, these two cities had the
highest effect estimates found by Smith et al. (2009; See Appendix 4A). Boston and Los Angeles
had the lowest Os-related mortality risk of the 12 urban study areas and are located at the lowest
end of the U.S. distribution. Overall, Os mortality risk in the 12 urban study areas are
representative of the full distribution of U.S. Os-related mortality risk, with the mean percentage
of May-September non-accidental mortality for all ages of 1.5% (95% confidence interval, 1.1-
1.8%).
For the results based on Zanobetti and Schwartz (2008) effect estimates, Detroit and New
York City are at the very highest end of the U.S. distribution of county-level risk of June-August
all-cause mortality due to ambient Os (Figure 8-26). These two cities had the highest effect
estimates of the 48 cities included in the study (see Appendix 4A). The high effect estimates in
Detroit and New York City could be due to high rates of public transportation use (for New York
City), low air conditioning prevalence, high smoking prevalence (in Detroit), high incidence of
mortality and other adverse health outcomes (e.g. diabetes, stroke, acute myocardial infarction,
etc.), and high unemployment rates. Houston and Los Angeles had the lowest risk and were
located at the very lowest end of the U.S. distribution of county-level risk of mortality due to
ambient Os. These two cities had the lowest effect estimates found by Zanobetti and Schwartz
(2008), possibly because they cover a large spatial extent and have high rates of time spent
driving, which could lead to exposure misclassification in the underlying epidemiologic study.
Houston also has a very high rate of air conditioning use (nearly 100% of residences) and Los
Angeles has been shown to have high rates of adaptive behavior on high ambient Os days (i.e.
more time spent indoors as a result of high ambient Os concentrations; Neidell 2009, 2010), both
of which would lead to lower personal Os exposure relative to other cities. Overall, Os mortality
risk in the 12 urban study areas are representative of the full distribution of U.S. Os-related
mortality risk, with the mean percentage of June-August all-cause mortality for all ages of 2.5%
(95% confidence interval, 1.7-3.0%).
For the results based on Jerrett et al. (2009) effect estimates, the 12 urban study areas are
centered more in the middle of the distribution of U.S. county-level risk of adult (ages 30 and
older) respiratory mortality due to ambient Os exposure. These results are based on the
application of a single national average effect estimate to all grid cells across the U.S., rather
than city-specific effect estimates as were applied for the results based on Smith et al. (2009) and
Zanobetti and Schwartz (2008) effect estimates. Therefore, the location of the urban study areas
on the distribution of county-level risk is driven mainly by Os concentration and not by the effect
estimate. While Denver, Atlanta, Sacramento, and Los Angeles are at the highest end of the U.S.
distribution, Figure 8-27 shows that some counties have a higher percentage of mortality
attributable to Os than these four cities. Overall, Os mortality risk in the 12 urban study areas are
8-40
-------
representative of the full distribution of U.S. Os-related mortality risk, with the mean percentage
of April-September respiratory mortality for adults ages 30 and older of 18.7% (95% confidence
interval, 15.2-21.5%). However, we are not capturing the very highest end of Os-related risk
based on Jerrett et al. (2009) effect estimates in the 12 urban study areas.
100%
0%
1 1.5 2 2.5 3 3.5 4
Percentage of baseline mortality attributable to ozone
Results based on Smith et al. (2009) effect estimates
• Selected urban studyarea
Figure 8-25. Cumulative Distribution of County-level Percentage of May-September non-
accidental Mortality for all ages Attributable to 2006-2008 Average Os for the U.S. and the
Locations of the Selected Urban Study Areas along the Distribution, using Smith et al.
(2009) Effect Estimates.
8-41
-------
1 1.5 2 2.5 3 3.5 4 4.5
Percentage of baseline mortality attributable to ozone
Results based on Zanobetti and Schwartz (2008) effect estimates
• Selected urban study areas
Figure 8-26. Cumulative Distribution of County-level Percentage of June-August all-cause
Mortality for all ages Attributable to 2006-2008 Average Os for the U.S. and the Locations
of the Selected Urban Study Areas along the Distribution, using Zanobetti and Schwartz
(2008) Effect Estimates.
100%
90%
»
|
1
*S
I
a 40%
m
>
H
re
J
20%
/
14 16 18 20 22 24 26
Percentage of baseline mortality attributable to ozone
28
^—Results based on Jerrett et al. (2009) effect estimate
• Selected urban study areas
Figure 8-27. Cumulative Distribution of County-level Percentage of April-September
Adult (age 30+) Respiratory Mortality Attributable to 2006-2008 Average O3 for the U.S.
and the Locations of the Selected Urban Study Areas along the Distribution, using Jerrett
et al. (2009) Effect Estimates and City Definitions from Zanobetti and Schwartz (2008).
8-42
-------
8.2.3 Analysis Based on Consideration of National Responsiveness of Ozone
Concentrations to Emissions Changes
Estimates of Os response to precursor emissions reductions (NOx and VOC) are
important inputs to estimation of risk for scenarios of just meeting existing and alternative Os
standards. To evaluate the national representativeness of Os responses to decreases in precursor
emissions in the 15 urban study areas, we examine two different types of air quality data. In
section 8.2.3.1 we examine ambient Os trends that have been measured at monitor locations
across the country over a recent period of decreasing NOx emissions. This analysis provides real-
world observations but does not isolate the effects of emissions changes alone and can only
characterize past phenomena. In section 8.2.3.2, we look at air quality model predictions of
temporal and spatial patterns of Os changes in response to further NOx reductions from 2007
levels. This analysis is subject to typical model limitations but has the advantage of isolating the
effects of precursor emissions changes and has the ability to simulate how Os would change in
response to NOx (and VOC) emissions reductions (relative to recent 2007 levels) similar to those
used in the HDDM adjustment scenarios for just meeting existing and alternative standards.
These two complimentary analyses give qualitatively similar results, building confidence that the
overarching conclusions are robust across the U.S. as a whole.
8.2.3.1 Ambient patterns in trends of measured ozone concentrations
This section describes how annual distributions of Os measurements collected by EPA's
national monitoring network have changed between 1998 and 2011. These years were chosen
because large reductions in anthropogenic NOx emissions have occurred over this time period
especially in the Eastern half of the United States. From 2000 to 2011 nationwide NOx emissions
were cut almost in half (from 22.6 to 12.9 million tons per year).3 However, it should be noted
that these reductions did not occur uniformly across the country. Improvements in vehicle
emissions standards helped reduce NOx emissions in many locations throughout the country. In
contrast, EPA rules like the NOx SIP call were focused on controlling emissions from power
plants in the Eastern US and consequently there have been relatively larger reductions in NOx
emissions in the East. In addition, some urban areas which have traditionally had high Os levels,
like Los Angeles and Houston, have substantially cut local NOx and VOC emissions to improve
their air quality. Also, there may be some localized areas in which NOx emissions have increased
due to population growth, new sources such as oil and gas development, or increased wildfire
activity. Appendix 8C provides plots of emissions trends by region of the U.S. These plots show
that each of nine regions of the U.S. have experienced decreasing NOx emissions ranging from
3 Dataware accessed from EPA's emission trend website on August 15, 2013:
http://www.epa.gov/ttn/chief/trends/trends06/national_tierl_caps.xlsx
8-43
-------
approximately a 20% decrease to a 45% decrease from 2002 to 2011 depending on the region.
Conversely, VOC emissions have increased in some regions since 2002 (the South, the
Southwest, and the West-North-Central) and decreased in others. Due to non-linear Os formation
chemistry and the potential for changes in local chemical regimes resulting from these emissions
reductions, past trends may not reflect the ambient changes which will occur from future
emissions reductions. Nonetheless, these ambient data provide information on actual Os changes
in response to emissions reductions and can give insight into the types of changes in Ch that have
occurred both within and outside the urban study areas.
First, we look at national maps which show changes in 50th percentile and 95th percentile
summer season (April-October)4 daily maximum 8-hr Os values (Figure 8-28, Figure 8-29).
These maps reflect the absolute (ppb) difference between Os percentiles from two three-year
periods (2001-2003 and 2008-2010)5. Figure 8-28 shows that increases in median Os
concentrations occurred in many large urban areas including both study area locations in Chapter
76 and non-study area locations7. Only a few monitors with increasing median Os appear outside
of cities, most notably in southwestern Colorado and central Kansas. The increases in urban
areas are likely explained by Os disbenefits in response to NOX reductions which were described
in Chapter 4, Appendix 4C and in the following section of this chapter. Widespread decreases of
median Os in suburban and rural locations suggest the efficacy of large NOX emissions
reductions on reducing Os over large regions of the country. Finally, the less frequently observed
cases of median Os increases in rural areas are likely caused by different phenomena. Cooper et
al. (2012) suggested that increasing rural Os in the Western US may be due to increasing oil and
gas development, wildfires and Os transport from Asia. Conversely, Figure 8-29 shows that 95th
percentile Os values for these two sets of years decrease in almost all urban as well as rural areas
of the country. Only a few sites in Colorado, Nevada, and California show any increases in 95th
percentile Os between 2001-2003 and 2008-2010. The consistent decreases across most of the
United States indicate that the large NOX reductions from power plants and mobile sources have
4 The April-October time period corresponds to the required monitoring season for most of the 12 urban areas.
Therefore, in selecting a consistent time period that could be analyzed for the urban case study areas, we chose to
use the April-October time period in Chapter 4 for composite monitor distributions to summarize Os
concentrations relevant to the epidemiological-based risk assessment.
5 These two three-year periods were chosen to represent years before and after most NOx emission reductions were
in place. In addition the 2001-2003 period was used to designate areas for the 1997 8-hr Os standard and the 2008-
2010 period was used to designate areas for the 2008 8-hr Os standard. Data from these two time periods have
undergone extensive quality checks.
6 Los Angeles, Denver, Houston, Atlanta, Chicago, Detroit, Cleveland, New York, Philadelphia, and Washington
D.C.
7 San Francisco, Reno, Phoenix, New Orleans, Birmingham, Miami, and Cincinnati.
8-44
-------
been quite successful in reducing Cb on the highest Os days. These results suggest that many of
the urban study areas may show Os responses that are typical of other large urban areas in the
U.S. However, decreasing Os in large non-urban portions of the country may not be fully
captured in the urban study areas.
To examine these trends further, we evaluate the 1998-2011 data from the 15 case-study
areas. Only monitors within the 15 study areas8 were analyzed, and within each study area,
monitors were put into three groups based on the degree of urbanization. The degrees of
urbanization were determined by the population density of the census tract containing the
monitor (plotted in Figure 8-30). Population data were obtained from the U.S. Census Bureau9,
and the classes were determined by breaks in the population density calculated from those data:
"high population density" (> 1000 people/km2), "medium population density" (between 400 and
1000 people/km2), and "low population density" (< 400 people/km2). Data were additionally split
out into three different time periods (all months, warm months: May through September, and
cool months: October through April). These warm and cool season categorizations were chosen
to isolate effects that are observed at different times of year. The April-October time period
which was examined in Figure 8-28 and Figure 8-29 include all warm season and two cold
season months and thus show behavior that has influences from both. Summaries were thus
calculated for groups of monitors specific to (1) Study Area, (2) Month subset, and (3) Urban
class.
8 These 15 areas are the 12 urban case study areas in the epidemiological-based risk assessment and the 3 additional
exposure urban case study areas.
9 Obtained from: http://www2.census.gov/geo/tiger/TIGER2010DPl/Tract_2010Census_DPl.zip
8-45
-------
Chang* In April - October MedUn Dally Maximum 8-tiour Ozone Concentration from 2001 -1003 to 2008 - 2010
*• Increase 10* ppb
* Increase 5 ppb
• No Change
* Decrease 5 ppb
v Decrease 104- ppb
Figure 8-28. Change in 50th Percentile Summer Season (April-October) Daily Maximum 8-
hr Average O3 Concentrations between 2001-2003 and 2008-2010.
Chang* In April - October 9Sth Percontllp Daily Maximum 8-hour Ozon* Concentration from 2001 - 2003 to 2008 • 2010
*• Increase 10+ ppb
• Increase S ppb
• No Change
* Decrease 5 ppb
v Decrease 10+ ppb
Figure 8-29. Change in 95th Percentile Summer Season (April-October) Daily Maximum 8-
hr Average O3 Concentrations between 2001-2003 and 2008-2010.
8-46
-------
Figure 8-30. Population Density within Census Tract where each Os Monitor is Located.
Figure 8-31, Figure 8-32, and Figure 8-33 display the data described above in ribbon
plots for high, medium, and low population density monitor locations in each case study area.
The lines bordering the dark and light red ribbons in this plot are (from top to bottom) the 95th,
75th, 25th, and 5th percentiles of the annual data indicated by each panel, and the median (i.e. 50th
percentile) is shown by the line in the middle of the central lighter ribbon. The colors of the lines
separating the ribbons depict significant trends (dark blue for decreasing and light blue for
increasing) or no significant trend (white). Statistical significance for multi-year Os trends was
determined using the Spearman rank order correlation coefficient (p-value < 0.05). Plots showing
a characterization of the entire Os distribution (not just discrete cut points of 5th, 25th, 50th, 75th,
and 95th percentiles) are provided in Appendix 8C.
These plots show consistent trends over the past 13 years for Os, with high Os values
decreasing fairly uniformly across different regions and areas of different degrees of
urbanization. Conversely, mean and median trends appear quite different in high, medium, and
low population density areas. Mid-range Os concentrations at low population density locations
within the case study areas (so still relatively close to a major city) have generally decreased over
a period of substantial NOX emissions reductions.10 This decrease is most pronounced in the
10 Denver is unique among the case study areas with consistently increasing mid-range Os trends across seasons and
urban classifications. Denver may be subject to increasing emissions from large wildfires and oil and gas
8-47
-------
summer months and in the Eastern half of the U.S. (low population density monitors in 3 out of
the 5 Western11 case-study areas and in only 4 out of the 10 Eastern12 case study areas do not
have significant decreases in summertime median Os concentrations). Mid-range Os
concentrations in many, but not all, high population density areas have significantly increased in
winter months. Wintertime increases were significant in 11 of the 15 areas (only Atlanta, Boston,
Houston and Sacramento did not increase significantly). Thirteen out of 15 summertime high
population density area trends in median Os were not significant,13 but combining winter and
summer measurements to determine annual trends showed that Denver, Los Angeles, New York
and Philadelphia high population density sites had significantly increasing annual median Os
while Boston, Chicago, Dallas and St. Louis had significantly increasing 25th percentile Os but
no significant median trend. These results reflect increasing mid-range Cb concentrations mainly
confined to urban centers during periods of NOx reductions. One important point to note is that
the design value monitor (the monitor with the highest average (over three years) of 4th highest
daily maximum value) in most of the case-study locations is located outside of the high
population density area (as defined here). Downward trends in medium and low population
density areas are therefore generally representative of the behavior at the highest Cb monitor in
an area, whereas trends in urban centers may be important from an exposure perspective.
In summary, any increasing Os trends occur more in highly populated areas, during cool months,
and at the lower end of the Os distribution. Conversely, any decreasing Os trends occur more
during warm months, in lower population areas, and at the upper end of the Os distribution. One
result of these two phenomena is a narrowing of the range of Os concentrations over this period
of decreasing NOx emissions. For instance, there are many cases where the top and bottom of a
single distribution exhibit different trends. For example, the low population density monitors of
Dallas, Los Angeles, Philadelphia and Saint Louis and the high population density monitors for
Baltimore, Dallas, and Philadelphia for all months show a significant increase in the 5th
percentile and a simultaneous significant decrease in the 95th percentile. More common is a
significant change in one end of the distribution, but no significant change in the other (e.g., the
summer months at high population density monitors in all case study areas except Baltimore,
Chicago, and Detroit). It is important to note that there are also cases where both ends of the
distribution change in the same manner and there is therefore no narrowing of the range of Os
concentrations in these areas.
development which are not typical of other urban case study areas. In addition, Denver is particularly susceptible
to influences from stratospheric intrusions and international transport due to its high altitude.
11 Western case study areas for this purpose include: Dallas, Denver, Houston, Los Angeles, and Sacramento.
12 Eastern case study areas for this purpose include: Atlanta, Baltimore, Boston, Chicago, Cleveland, Detroit, New
York, Philadelphia, St. Louis, and Washington D.C.
13 Only Houston and Dallas had statistically significant trends in median summertime urban Os.
8-48
-------
f
Trend
Chin
— StgnrfNsg
SigmPos
ln*gnit.Po»
ii! fi! lit Hi fit III HI It! HI it! tit tit itt tit ttt
Year
Figure 8-31. Distributions of Os Concentrations for High Population Density Monitors by
Different Subsets of Months over a 13-year Period. From top to bottom in each ribbon plot,
the blue and white lines indicate the spatial mean of the 95th, 75th, 50th, 25th, and 5th
percentiles for each monitor for every year from 1998-2011.
til It! til HI tit tit ttt tit tit ttt tit ttt itt
Trend
Chan
—
SgmtPos
Figure 8-32. Distributions of Os Concentrations for Medium Population Density Monitors
by Different Subsets of Months over a 13-year Period. From top to bottom in each ribbon
plot, the blue and white lines indicate the spatial mean of the 95th, 75th, 50th, 25th, and 5th
percentiles for each monitor for every year from 1998-2011.
8-49
-------
Trend
Chin
— Sign,tNeg
SiJ«f:Pc»
HI H! Hi Hi HI HI HI HI HI HI Hi lit Hi HI HI
Year
Figure 8-33. Distributions of Os Concentrations for Low Population Density Monitors by
Different Subsets of Months over a 13-year Period. From top to bottom in each ribbon plot,
the blue and white lines indicate the spatial mean of the 95th, 75th, 50th, 25th, and 5th
percentiles for each monitor for every year from 1998-2011.
Maps of ambient trends in both New York City and Chicago most clearly show these
trends and further illustrate this behavior. Figure 8-34 and Figure 8-35 show trends in daily
maximum 8-hr Os values these two cities for May-September. Plots for other case-study areas
are provided in Appendix 8C. For both cities, the 4th highest daily maximum 8-hr average Os
value either has a downward trend or no trend at all monitors. In New York (Figure 8-34), mean
and median Os values significantly decrease at downwind locations in New York and
Connecticut. Conversely, median Os values significantly increase from 1998 to 2011 at two core
urban sites (one at City College of NY in upper Manhattan and one near Queen's college) and at
a nearby site on Long Island. Similarly, in Chicago (Figure 8-35), mean and median trends in Os
are downward or insignificant in Indiana and in suburban Illinois locations and show increases
near the highly populated urban core.
8-50
-------
Use in
Epi Studies
O Not In Epi Study
O Smith
O Smith + Zano
Trend
Direction
Insignificant
V Negative
A Positive
Figure 8-34. Map of Os Trends at Specific Monitors in the New York Area. All upward and downward facing triangles
represent statistically significant trends from 1998-2011 (p < 0.05), circles represent locations with no significant trends. Sites
used in Smith et al (2009) and the Zanobetti and Schwartz (2008) epidemiology studies are represented by colored dots. Only
monitors with at least seven years of data are displayed. The pink star indicates the site with the higher design value in 2011.
The MSA border as defined by the U.S. census bureau is delineated by the light blue line. Left panel shows trends in annual 4th
highest daily maximum 8-hr average Os values, center panel shows trends in annual means of the daily maximum 8-hr average
Os values, and right panel shows trends in annual medians of the daily maximum 8-hr average Os values.
8-51
-------
Max4
Mean
Median
* Elgin" "f~ ^ I I • Evanston
Nagerville
/ \ v
f \ Haj
>-'-JofietT
J \ Han
f \ Hammond;
/
Trend
Direction
Insignificant
V Negative
A Positive
Use in
Epi Studies
O Not In Epi Study
Figure 8-35. Map of Os Trends at Specific Monitors in the Chicago Area. All upward and downward facing triangles
represent statistically significant trends from 1998-2011 (p < 0.05), circles represent locations with no significant trends. Sites
used in Smith et al (2009) and the Zanobetti and Schwartz (2008) epidemiology studies are represented by colored dots. Only
monitors with at least seven years of data are displayed. The pink star indicates the site with the higher design value in 2011.
The MSA border as defined by the U.S. census bureau is delineated by the light blue line. Left panel shows trends in annual 4th
highest daily maximum 8-hr average Os values, center panel shows trends in annual means of the daily maximum 8-hr average
Os values, and right panel shows trends in annual medians of the daily maximum 8-hr average Os values.
8-52
-------
To demonstrate how changes in emissions of NOx and anthropogenic VOCs might be
driving these trends, Table 8-8 shows trends of Os in high and low population areas and annual
National Emissions Inventories (NEI) for 2002, 2005, 2008 and 201114 aggregated to the level of
the NOAA Climate Regions.15 There is moderate correspondence between the decreases in NOx
emissions across the regions with the observed decreases in warm season Os concentrations in
low population areas. VOCs show little correspondence to any of the Os trends, which is likely
due to complications from (1) the mix of chemicals with a large range of reactivities; (2)
complex non-linear chemistry; and (3) the potential impact of the much larger magnitude of
biogenic vs. anthropogenic emissions on a regional scale. Details of these calculations can be
found in Appendix 8C.
Table 8-8. Broad Regional Annual Trends of Concurrent
Emissions of NOX and VOCs over years 2000-2011.
Concentrations and
Trend
High Population
Density, May-Sept
03
High Population
Density, Oct-Apr Os
Low Population
Density, May-Sept
03
Low Population
Density, Oct-Apr Os
NOx Emission
VOC Emission
Central
None
Up
Down
None
Down
None
East
North
Central
None
Up
Down
None
Down
None
North
East
None
Up
Down
Top %'s
up
Down
Down
South
Down
Low %'s
up
Down
None
Down
Up
South
East
None
None
Down
None
Down
Down
South
West
None
Up
Low %'s
up
Up
Down
Up
West
None
Up
Top %'s
down
None
Down
Down
8.2.3.2 Modeled ozone response to emissions reductions across the United States
This section presents an analysis of the CMAQ modeling of Os responses to "across-the-
board" U.S. anthropogenic precursor emissions reductions described in Appendix 4B. In this
analysis, we compare the modeled responses of Os concentrations in the case study areas to the
modeled Os responses in the rest of the U.S. For this purpose, we used five CMAQ model
simulations: (1) a base simulation which included 2007 emissions of all Os precursors, (2) a 50%
NOx cut simulation in which U.S. anthropogenic NOx emissions were reduced by 50% from
14 http://www.epa.gov/ttnchiel/trends/
15 http://www.ncdc.noaa.gov/monitoring-references/maps/us-climate-regions.php
8-53
-------
2007 levels, (3) a 90% NOx cut simulation in which U.S. anthropogenic NOx emission were
reduced by 90% from 2007 levels, (4) a 50% NOx/VOC cut simulation in which both U.S.
anthropogenic NOx and VOC emission were reduced by 50% from 2007 levels, and (5) a 90%
NOx/VOC cut simulation. These simulations are analyzed to characterize responses in Os to
"across the board" emissions cuts at four distinct levels and do not represent the exact adjustment
cases that were used to estimate Os concentrations consistent with individual case study areas
just meeting various potential levels of the NAAQS standard. However, these four cases
generally span the range of emissions perturbations that were applied in the HDDM adjustment
methodology described in Chapter 4 and in Appendix 4D.
In this analysis we focus on seasonal mean Os and population as proxies for
epidemiology based risk estimates in Chapter 7. Since the epidemiology studies used in Chapter
7 show relatively linear response of health outcomes to Os concentrations throughout the entire
range of measured Os values, examining seasonal mean values should provide some
understanding of locations where Os health effects are expected to increase and decrease as a
result of precursor emission reductions. By combining population information with these spatial
distributions of seasonal Os responses, we can better understand expected Os behavior in
locations where people live. This is not a detailed risk assessment but can provide information on
the representativeness of the case-study areas to the nation as a whole in terms of expected Os-
related health outcomes.
To begin, we examine maps displaying ratios of mean Os concentrations in the emissions
cut simulations to mean Os concentrations in the 2007 base simulation. Figure 8-36 and Figure
8-37 show the ratio of seasonal (April-October) mean Os in the two NOx emissions reduction
simulations to that in the base simulation for the entire model domain. Figure 8-38 and Figure
8-39 depict the ratio of January mean Os for the two NOx cut simulations. Figures showing the
ratios based on the May-September seasonal average and figures for the NOx/VOC emissions
reductions scenarios are provided in Appendix 8C. The maps show widespread decreases (i.e.,
ratios less than 1) in seasonal mean Os across the country. These decreases are especially
pronounced in the Eastern U.S. and in California. Ozone increases (i.e., ratios greater than 1) are
confined to urban core areas except in January. The spatial extent of these Os increases are
generally less for the 90% NOx cut simulation than for the 50% NOx cut simulation although the
magnitude is greater over very limited areas in Chicago, Seattle, and San Francisco. The Os
increases are most widespread in the cooler months (January, April, and October). For the April-
October seasonal average Os concentrations, VOC in addition to NOx cuts did not substantially
change the ratios of Os in the emissions reduction scenarios to Os in the base scenario. In the
Northeast and Midwest, increases in seasonal mean Os concentrations were mainly confined to
urban study areas of New York, Detroit, Chicago, and St. Louis. In the Southeastern U.S., the
8-54
-------
urban areas which show up as having increased seasonal mean Ch in the 50% NOX cut
simulations include Miami, Orlando, Tampa, and New Orleans (only Miami has Os increase in
the 90% NOx cut simulation). The only case-study area in the southeast, Atlanta, does not
experience increases in seasonal mean Os in the model simulation (this is consistent with the
changes in risk estimated for Atlanta, which show no increases in total risk as alternative
standards are simulated). In the central U.S., seasonal mean Os in the case study areas of Denver,
Houston, and Dallas and non-case-study areas of San Antonio, Duluth, and Minneapolis
increased with 50% NOx reductions. Seasonal mean Os increases were seen only in Houston,
Minneapolis, and Duluth with 90% reductions in simulated NOx emissions. The Northwestern
U.S. showed some of the most widespread increases in seasonal mean Os in the 50% and 90%
NOx cut simulations covering the Seattle and Portland metro areas as well the San Francisco Bay
area and in a single model grid cell for Sacramento (50% NOx reduction case only). Sacramento
is the only city in the Northwest that was included as a case study area. Finally, Los Angeles (a
case study area), San Diego, Phoenix, and Bakersfield were the areas for which CMAQ predicted
seasonal mean Os increases with the 50% NOx cut simulation. These Os increases disappeared
(or were largely diminished in the case of LA) in the 90% NOx cut case. Based on these maps, it
appears that in the Northeast and the Central U.S., the case-study area selection likely
oversampled these Os increases on a geographic basis since all locations outside of city centers
experienced decreasing seasonal mean Os with the NOx reduction model simulations. However
in two regions, the Southeast and the Northwest, the urban study area did not experience
increases in seasonal mean Os concentrations while other urban areas in the region did. In these
two regions, the urban study area selection likely under-sampled the locations which experienced
increases in seasonal mean Os.
8-55
-------
0.4 0.6 0.8 1.0 1.2 1.4
ratio of seaonal mean ozone
Figure 8-36. Ratio of April-October Seasonal Average Os Concentrations in the Brute-
force 50% NOX Emissions Reduction CMAQ Simulations to April-October Seasonal
Average Os Concentrations in the 2007 Base CMAQ Simulation.
0.4 0.6 0.8 1.0 1.2
ratio of seaonal mean ozone
1,4
Figure 8-37. Ratio of April-October Seasonal Average Os Concentrations in the Brute-
force 90% NOX Emissions Reduction CMAQ Simulations to April-October Seasonal
Average Os Concentrations in the 2007 Base CMAQ Simulation.
8-56
-------
0.4 0.6 0.8 1.0 1.2 1.4
ratio of seaonal mean ozone
Figure 8-38. Ratio of January Monthly Average Os Concentrations in Brute-force 50%
NOX Emissions Reduction CMAQ Simulations to January Monthly Average Os
Concentrations in the 2007 Base CMAQ Simulation.
0.4 0.6 0.8 1.0 1.2 1.4
ratio of seaonal mean ozone
Figure 8-39. Ratio of January Monthly Average Os Concentrations in Brute-force 90%
NOX Emissions Reduction CMAQ Simulations to January Monthly Average Os
Concentrations in the 2007 Base CMAQ Simulation.
8-57
-------
To better characterize the representativeness of case study areas, paired Os concentrations and
population data16 were extracted from each model grid cell and categorized. Figure 8-40 and
Figure 8-41 depict the percent of U.S. population living in areas with increases or decreases in
monthly or seasonal mean Os under the emissions reductions scenarios (50% NOx cut and 90%
NOx cut respectively) compared to Os in the base modeling scenario. The top panels show data
for January monthly mean Os, the center panels show data for seasonal mean Os (June-August),
and the bottom panels show data for seasonal mean Os (April-October). Tabulated results and
equivalent plots for the combined NOx/VOC cut simulations are provided in Appendix 8C.
Month by month break-outs for each case study area are also available in Appendix 8C.
The vast majority of the U.S. population lives in areas where the CMAQ simulations
predict mean Os decreases for the June-August and April-October time periods. The majority of
population living in case-study areas also lives in locations with decreasing seasonal mean Os
concentration under NOx reduction scenarios. As discussed previously, more locations have
increasing mean Os in the cooler months as demonstrated by the fact that almost all of the U.S.
population lives in locations where the model predicts increases in mean Os in January. The case
study areas represent 29% of the total U.S. population. These areas account for 20-30% of the
U.S. population that experience decreasing seasonal mean Os for April-October in the NOx cut
simulations and 50-60% of the U.S. population that experience increasing seasonal Os for April-
October. Consequently, the urban study areas over-sample populations living in locations with
increasing seasonal mean Os in response to NOx cuts compared to populations living in locations
with decreases in seasonal mean Os. In all panels displayed in Figure 8-40 and Figure 8-41, most
of the population lives in locations where increases or decreases in mean Os were > 1 ppb.
The proportion of the population living in locations of increasing seasonal mean Os in
response to NOx emissions reductions varies considerably between case study areas. Figure 8-42
and Figure 8-43 show these proportions by city for the 50% and 90% NOx reduction scenarios.
For the 50% NOx reduction scenario, the CMAQ results predict that four out of fifteen study ares
(Chicago, Detroit, Los Angeles, and New York) have more than 50% of their populations living
in locations with increasing mean Os for April-October. Most other urban study areas have
between 5% and 30% of their populations living in these areas with increasing mean Os levels.
For the 90% NOx reduction scenario, the percent of population living in such locations decreases
substantially for all cities, leaving four out of fifteen study areas (Detroit, Houston, Los Angeles,
and New York) with more than 5% of their populations living in areas with increasing mean Os
levels.
16 Block level population data from the 2010 Census was aggregated to the 12km CMAQ grid cell level. The 2007
population was then calculated using population growth factors developed by Woods and Poole Economics, Inc.
8-58
-------
Non-Study Area
i Area
o
60
40
ppb change
Figure 8-40. Histograms of U.S. Population Living in Locations with Increasing and
Decreasing Mean Os Concentrations. Values on the x-axis represent change in mean Os
(ppb) from the 2007 base CMAQ simulation to the 50% NOX cut CMAQ simulation. The
percentages of the U.S. population living in areas that have changes less than -1 ppb, from
1 to +1 ppb, and greater than 1 ppb are shown on the y-axis. Left plots show population
numbers in locations not included in one of the cases study areas while right plots show
population numbers in locations included in one of the case study areas. Top plots show
changes in January monthly mean Os, middle plots show changes in seasonal mean June-
August Os, and bottom plots show changes in seasonal mean April-October Os.
8-59
-------
Non-Study Area
Study Area
40-
20-
60
.2
73
340-
0.
o
Q.
60
40 •
20-
ppb change
Figure 8-41. Histograms of U.S. Population Living in Locations with Increasing and
Decreasing Mean Os Concentrations. Values on the x-axis represent change in mean Os
(ppb) from the 2007 base CMAQ simulation to the 90% NOX cut CMAQ simulation. The
percentages of the U.S. population living in areas that have changes less than -1 ppb, from
1 to +1 ppb, and greater than 1 ppb are shown on the y-axis. Left plots show population
numbers in locations not included in one of the cases study areas while right plots show
population numbers in locations included in one of the case study areas. Top plots show
changes in January monthly mean Os, middle plots show changes in seasonal mean June-
August Os, and bottom plots show changes in seasonal mean April-October Os.
8-60
-------
Figure 8-42. Population (as % of total case-study area population) Living in Locations of
Increasing April-October Seasonal Mean Os in the 50% NOX Reduction CMAQ
Simulation.
35.0%
30.0%
25.0%
10.0%
0.0%
1
t
Figure 8-43. Population (as % of total case-study area population) Living in Locations of
Increasing April-October Seasonal Mean Os in the 90% NOX Reduction CMAQ
Simulation.
8-61
-------
We can further understand these results by looking at them in terms of population density
in the case-study areas versus across the U.S. as a whole. As in Section 8.2.3.1, we define census
tracts with population density greater than 1000 people/km2 as high population density, but the
low-mid population density classification used here is a combination of the low and medium
classifications in that section. Figure 8-44 and Figure 8-45 split out the April-October results
from Figure 8-40 and Figure 8-41 into high and low-mid sub-categories. Appendix 8C provides
similar breakouts for the other panels in Figure 8-40 and Figure 8-41. First, based on these
definitions, we see that 57% of the population in case-study areas lives in high population
density locations while only 27% of the U.S. population does. As discussed above, the high
population areas are more likely to experience increases in mean Os as a result of NOX emission
reductions compared to lower population areas. Therefore, the fact that the case-study areas used
in the risk and exposure assessments are more densely populated than the country as a whole
means that these analyses may estimate higher risks under emissions reduction scenarios than
would be experienced, on average, across the country. Figure 8-44 and Figure 8-45 show
generally similar shapes for the high population density histograms in the study-area and non-
study area locations. In the 50% NOx cut simulation, 69% of the population living in high density
case-study areas would experience increases in mean seasonal Os compared to 63% of the
population the population living in high density areas of the country as a whole. Similarly in the
90% NOx cut simulation, 28% of the population in high density locations both within the study
areas and across the U.S. as a whole lives in locations of increasing seasonal mean Os. This
suggests that the selected study areas adequately represent population-weighted changes in mean
Os for people living in high density areas. Similarly, less densely populated locations within the
case-study areas show Os increases equivalent to those seen in less densely populated areas in
the U.S. as a whole. In the 50% NOx cut simulation, 7% of people in low-mid density study area
locations live where mean seasonal Os is increasing, while 5% of people in all low-mid density
U.S. locations live where mean seasonal Os is increasing. Similarly, in the 90% NOx cut
simulation, the numbers are 2% for both low-mid density study area populations and for low-mid
density populations in the U.S. as a whole. Thus the oversampling of populations living in
locations of increasing mean seasonal Os in response to NOx cuts, as shown in Figure 8-44 and
Figure 8-45, appears to be entirely due to the fact that the study areas oversample populations
living in high density areas compared to the U.S. population as a whole.
8-62
-------
Non-SfttJy
Study
40
I
o
n>
3
20-
,2
5
3 0-
o.
o
Q.
CO
40-
ppb change
Figure 8-44. Histograms of U.S. Population Living in Locations with Increasing and
Decreasing Mean Os Concentrations. Values on the x-axis represent the change in seasonal
mean (April-October) O3 from the 2007 base CMAQ simulation to the 50% NOX cut
CMAQ simulation. The percentages of the U.S. population living in areas that have
changes less than -1 ppb, from -1 to +1 ppb, and greater than 1 ppb are shown on the y-
axis. Left plots show population numbers in locations not included in one of the cases study
areas while right plots show population numbers in locations included in one of the urban
study areas. Bottom plots show histograms for low-mid population density areas while top
plots show histograms for high population density areas.
8-63
-------
Non-Study Area
StucJyArea
40-
I
I
O
C
•2
CO
%J>
40
a
I
ppb change
Figure 8-45. Histograms of U.S. Population Living in Locations with Increasing and
Decreasing Mean Os Concentrations. Values on the x-axis represent the change in seasonal
mean (April-October) O3 from the 2007 base CMAQ simulation to the 9Q% NOX cut
CMAQ simulation. The percentages of the U.S. population living in areas that have
changes less than -1 ppb, from -1 to +1 ppb, and greater than 1 ppb are shown on the y-
axis. Left plots show population numbers in locations not included in one of the cases study
areas while right plots show population numbers in locations included in one of the urban
study areas. Bottom plots show histograms for low-mid population density areas while top
plots show histograms for high population density areas.
8-64
-------
8.2.4 Discussion
We evaluated two different questions, (1) to what degree are the 15 cities evaluated in the
exposure and risk analyses representative of the overall U.S. population with regards to total Os
risk, and (2) to what degree are they representative of the overall U.S. population with regards to
the degree of risk reduction that might be observed in response to just meeting the existing and
alternative standards.
Regarding the first question, we observe that the 12 urban study areas considered in the
urban-scale risk assessment presented in Section 7.2 capture urban areas that are among the most
populated in the U.S., have relatively high Os levels, and represent the range of city-specific
effect estimates found by Smith et al. (2009) and Zanobetti and Schwartz (2008). These three
factors suggest that the urban study areas capture overall risk for the nation well, with a potential
for better characterization of the high end of the risk distribution. We find that the urban study
areas are not capturing areas with the highest baseline mortality rates, those with the oldest
populations, and those with the lowest air conditioning prevalence. These areas tend to have
relatively low Os concentrations and low total population, suggesting that the urban study areas
are not missing high risk populations that have high Cb concentrations in addition to greater
susceptibility per unit Os. We also find that the 12 urban study areas represent the full range of
county-level Cb-related risk across the entire U.S. We conclude from these analyses that the 12
urban study areas adequately represent Os-related risk across the U.S.
Concerning the second question, we observe that the 15 urban areas considered in the
exposure and risk assessment case study areas over-sample populations living in locations with
increasing seasonal mean Os in response to NOX cuts. This suggests that the selected study areas
adequately represent population-weighted changes in mean Os concentrations for urban
populations, but may be under-representing decreasing median Os concentrations in suburban
and rural areas. As a result, the risk estimates for populations in the selected urban study areas
may understate the risk reductions that might be achieved across the broader U.S. population.
8-65
-------
8.3 REFERENCES
Abt Associates, Inc. 2010. Model Attainment Test Software (Version 2), prepared for EPA.
Research Triangle Park, NC: EPA Office of Air and Radiation, Office of Air Quality
Planning and Standards. Research Triangle Park, NC.
.
Abt Associates, Inc. 2010. Environmental Benefits and Mapping Program (Version 4.0),
prepared for EPA. Research Triangle Park, NC: EPA Office of Air Quality Planning and
Standards, .Anenberg, S.C.; JJ. West; L.W. Horowitz
and D.Q. Tong. 2010. An estimate of the global burden of anthropogenic ozone and fine
particulate matter on premature human mortality using atmospheric modeling.
Environmental Health Perspective. 118:1189-1195.
Anenberg, S.C.; JJ. West; L.W. Horowitz and D.Q. Tong. 2010. "An Estimate of the Global
Burden of Anthropogenic Ozone and Fine Particulate Matter on Premature Human
Mortality Using Atmospheric Modeling." Environmental Health Perspective, 118:1189-
1195.
Anenberg, S.C.; JJ. West; L.W. Horowitz and D.Q. Tong. 2011. The global burden of air
pollution mortality: Anenberg et al. respond. Environmental Health Perspective.
119:A158-A425.
Bell, M.L.; A. McDermott; S.L. Zeger; J.M. Samet and F. Dominici. 2004. Ozone and short-term
mortality in 95 U.S. urban communities, 1987-2000. Journal of the American Medical
Association. 292:2372-2378.
Byun, D. and K.L. Schere. 2006. Review of governing equations, computational algorithms, and
other components of the Models-3 Community Multi-scale Air Quality (CMAQ)
modeling system. Applied Mechanics Reviews. 59:51-77.
Centers for Disease Control: Wide-ranging Online Data for Epidemiological Research (CDC-
Wonder) (data from years 2004-2006), Centers for Disease Control and Prevention
(CDC), U.S. Department of Health and Human Services. .
Cooper, O.R.; R. Gao; D. Tarasick; T. Leblanc and C. Sweeney. 2012. Long-term ozone trends
at rural ozone monitoring sites across the United States, 1990-2010. Journal of
Geophysical Research. 117(D22), doi: 10.1029/2012JDO18261.
FannN.; A.D. Lamson; S.C. Anenberg; K. Wesson; D. Risley; BJ. Hubbell. 2012. Estimating
the national public health burden associated with exposure to ambient PM2.5 and ozone.
Risk Analysis. 32:81-95.
George, BJ. and T. McCurdy. 2009. Investigating the American Time Use Survey from an
exposure modeling perspective. Journal of Exposure Science and Environmental
Epidemiology. 21:92-105.
Graham, S. and T. McCurdy. 2004. Developing meaningful cohorts for human exposure models.
Journal of Exposure Analysis and Environmental Epidemiology. 14:23-43.
Hollman, F.W.; TJ. Mulder and J.E. Kalian. 2000. Methodology and assumptions for the
population projections of the United States: 1999 to 2100. Population Division Working
Paper No. 38, Population Projections Branch, Population Division, U.S. Census Bureau,
Department of Commerce.
8-66
-------
Jerrett, M.; R.T. Burnett; C.A. Pope, III; K. Ito; G. Thurston; D. Krewski; Y. Shi; E. Calle; M.
Thun. 2009. Long-term OB exposure and mortality. New England Journal of Medicine.
360:1085-1095.
Neidell, M. 2009. Information, avoidance behavior and health. JHuman Res. 44:450-478.
Neidell, M. 2010. Air quality warnings and outdoor activities: evidence from Southern California
using a regression discontinuity approach design. J Epidemiol Community Health.
64:921-926.
Samet, J.M.; S.L. Zeger; F. Dominici; F. Curriero; I. Coursac; D.W. Dockery; J. Schwartz and A.
Zanobetti. 2000. The national morbidity, mortality, and air pollution study part II:
morbidity and mortality from air pollution in the United States. Health Effects Institute,
Boston, MA, Number 94, Part II.
Smith R.L.; B. Xu and P. Switzer. 2009. Reassessing the relationship between ozone and short-
term mortality in U.S. urban communities. Inhalation Toxicology. 21:37-61.
Timin, B.; K. Wesson and J. Thurman. 2010. Application of model and ambient data fusion
techniques to predict current and future year PM2.5 concentrations in unmonitored Areas.
(2010). In Steyn DG, Rao St (eds). Air Pollution Modeling and Its Application XX.
Netherlands: Springer, pp. 175-179.
U.S. EPA. 2007. Ozone Population Exposure Analysis for Selected Urban Areas. Research
Triangle Park, NC: EPA. (EPA document number EPA-452/R-07-010).
.
U. S. EPA. 2009. Risk and Exposure Assessment to Support the Review of the SO2 Primary
National Ambient Air Quality Standards: Final Report. Research Triangle Park, NC:
EPA. (EPA document number EPA-452/R-09-007).
U.S. EPA. 2010. Quantitative Health Risk Assessment for Particulate Matter. Research Triangle
Park, NC: EPA. (EPA document number EPA-452/R-10-005).
U.S. EPA. 2013. Integrated Science Assessment for Ozone and Related Photochemical Oxidants.
Research Triangle Park, NC: EPA. (EPA document number EPA 600/R-10/076F).
.
Woods and Poole Economics Inc. 2012. Complete Demographic Database. Woods and Poole
Economics, Inc. Washington, DC. .
Zanobetti, A. and J. Schwartz. 2008. Mortality displacement in the association of ozone with
mortality: an analysis of 48 cities in the United States. American Journal of Respiratory
and Critical Care Medicine. 177:184-189.
Zanobetti, A. and J. Schwartz. 2011. Ozone and survival in four cohorts with potentially
predisposing diseases. American Journal of Respiratory and Critical Care Medicine.
194:836-841.
Zhang, L.; DJ. Jacob; N.V. Smith-Downey; D.A. Wood; D. Blewitt; C.C. Carouge; A. van
Donkelaar; D.B.A. Jones; L.T. Murray and Y. Wang. 2011. Improved estimate of the
policy-relevant background ozone in the United States using the GEOS-Chem Global
Model with l/2°x2/3° horizontal resolution over North America. Atmospheric
Environment. 45:6769-6776.
8-67
-------
This page left intentionally blank
8-68
-------
9 SUMMARY AND SYNTHESIS
9.1 INTRODUCTION
This assessment estimates exposures to Cb and resulting mortality and morbidity health
risks based on the findings of the Os ISA (U.S. EPA, 2013) that short-term Os exposures are
causally related to respiratory effects, and likely causally related to cardiovascular effects, and
that long term Os exposures are likely causally related to respiratory effects. The assessment
evaluates total exposures and risks associated with the full range of observed Os concentrations,
as well as the incremental changes in exposures and risks between just meeting the existing
standard of 75 ppb and just meeting alternative standard levels of 70, 65, and 60 ppb using the
form and averaging time of the existing standard: the annual 4th highest daily maximum 8-hr Cb
average concentration, averaged over three consecutive years. We evaluated alternative standard
levels of 70, 65, and 60 consistent with recommendations from CASAC to consider alternative
standard levels between 60 and 70 ppb (Frey and Samet, 2012).
Following the conceptual framework described in Chapter 2, the assessment evaluates
exposures and lung function risk in 15 urban study areas, and mortality and morbidity risks based
on concentration-response (C-R) functions derived from epidemiology studies in 12 of these
urban study areas1. The results from these assessments will help inform consideration of the
adequacy of the existing primary Os standards, and potential risk reductions associated with
several alternative levels of the standard (for the current form and averaging time). In addition, to
place the urban study area analyses in a broader context, Chapter 8 of this assessment estimates
the national burden of mortality associated with recent Os levels, and evaluates the
representativeness of the 15 urban study areas in characterizing Os exposures and risks across the
U.S. This synthesis focuses on the urban study area assessments of exposure and risk for the
scenarios of just meeting the existing and alternative standards. For this synthesis, we discuss the
results of the national-scale assessment as they relate to understanding the breadth of Os risks
across the U.S. and to the national representativeness of the urban study area risk results.
To facilitate interpretation of the results of the exposure and risk assessment, this chapter
provides a synthesis of the various results, focusing on comparing and contrasting those results
to identify common patterns, or important differences. These comparisons will focus on patterns
1 Three additional urban study areas were evaluated for the human exposure assessment and lung function risk
assessment to provide greater geographic representation. There was insufficient information available to conduct
the epidemiology-based risk assessment in these 3 additional areas. Also, we originally planned to include Seattle,
WA as a 16th urban study area, but due to limitations in the available air quality monitoring data, we determined
that it would not be appropriate to model exposure and risks for Seattle (see appendix 4-E).
9-1
-------
across urban study areas, across years of analysis, and across alternative standards. In addition,
factors related to each specific type of analysis that may influence comparisons between the
analyses are identified and discussed. The degree to which the integrated results are
representative of national patterns of exposure and risk is evaluated. Overall confidence in the
results, as well as relative confidence between the different analyses is also assessed. The chapter
concludes with an overall integrated characterization of exposure and risk in the context of key
policy-relevant questions raised in Chapter 2.
9.2 SUMMARY OF KEY RESULTS
This section provides summaries of key methodological considerations and results from
each of the major analytical chapters of the Health Risk and Exposure Assessment (HREA). An
overall summary of the exposure and risk results across urban study areas and alternative
standard levels is provided in Table 9-1. For human exposures, the results for percent of all
school-age children experiencing exposures above a particular level of concern are almost
identical for the percent of asthmatic school-age children experiencing those exposures. In Table
9-1 and throughout the presentation of summary results, ranges of results across urban study
areas should be compared between alternative standard levels with the caveat that the ranges for
the 60 ppb alternative standard level exclude the New York study area, for which we were
unable to adjust air quality to just meet the alternative standard level. As a result, the ranges may
not be fully comparable between the alternative standards.
9-2
-------
Table 9-1. Summary of Urban Scale Os-Exposure Risk Across Alternative Os Standard Levels.
Analysis
Human Exposure
Controlled
Human
Exposure-based
Lung Function
Risk
Metric
% of all school-age
children with at least one
exposure above 60, 70,
or 80 ppb
% of all school-age
children with at least two
exposures above 60, 70,
or 80 ppb
% of all school-age
children with at least one
occurrence of lung
function decrement
>10%, 15%, or 20%
% of all school-age
children with at least two
occurrences of lung
function decrement
>10%, 15%, or 20%
Estimated Risks at Alternative 4th Highest Daily Maximum 8-hr Average Os Level
70 ppb
65 ppb
60 ppb*
Average year 2006-2010
60 ppb: 3 to 10%
70 ppb: 1 % or less
80 ppb: -0%
60 ppb: 4% or less
70 ppb: 0.2% or less
80 ppb: 0%
60 ppb: 1 % or less
70 ppb: 0%
80 ppb: 0%
Worst case year 2006-201 0
60 ppb: 5 to 19%
70 ppb: 3% or less
80 ppb: 0.2% or less
60ppb:<10%
70ppb:<1%
80 ppb: 0%
60 ppb: 2% or less
70 ppb: -0%
80 ppb: 0%
Average year 2006-2010
60 ppb: 0.5 to 3.5%
70 ppb: -0%
80 ppb: 0%
60 ppb: <1%
70 ppb: 0%
80 ppb: 0%
60 ppb: -0%
70 ppb: 0%
80 ppb: 0%
Worst case year 2006-201 0
60 ppb: 1 to 9%
70 ppb: 0.4% or less
80 ppb: 0%
60 ppb: 3% or less
70 ppb: 0%
80 ppb: 0%
60 ppb: 0.3% or less
70 ppb: 0%
80 ppb: 0%
Average year 2006-2010
1 0% decrement: 1 1 to 1 7%
1 5% decrement: 2 to 4%
20% decrement: 2% or less
1 0% decrement: 3 to 1 5%
15% decrement: 3% or less
20% decrement: <1 .5%
1 0% decrement: 5 to 1 1 %
1 5% decrement: <3%
20% decrement: <1%
Worst case year 2006-201 0
10% decrement: 14 to 20%
1 5% decrement: 3 to 5%
20% decrement: 2% or less
1 0% decrement: 4 to 1 8%
15% decrement: 4% or less
20% decrement: <1 .5%
1 0% decrement: 5 to 1 3%
1 5% decrement: <3%
20% decrement: <1%
Average year 2006-2010
1 0% decrement: 6 to 1 1 %
15% decrement: 1 to 2.5%
20% decrement: 1% or less
1 0% decrement: 1 to 9%
15% decrement: 2% or less
20% decrement: <1%
1 0% decrement: 2 to 6%
15% decrement: <1.5%
20% decrement: <0.5%
Worst case year 2006-201 0
1 0% decrement: 7 to 1 3%
1 5% decrement: 1 to 3%
20% decrement: 1% or less
1 0% decrement: 2 to 1 1 %
15% decrement: 2.5% or less
20% decrement: <1%
1 0% decrement: 2 to 7%
15% decrement: <1.5%
20% decrement: <0.5%
9-3
-------
Analysis
Epidemiology-
based Risk
Metric
%/number per 100,000
population of premature
deaths from short-term
exposures to Os
%/number of per
100,000 population
premature respiratory-
related deaths from
long-term exposures to
03
number of respiratory
hospital admissions from
short-term exposures to
03
Estimated Risks at Alternative 4th Highest Daily Maximum 8-hr Average Os Level
70 ppb
2007: 0.8 to 3.9% of deaths
2.4 to 17 per 100,000
2009: 0.8 to 3. 8% of deaths
2.2 to 16 per 100,000
2007: 16 to 20% of deaths
15 to 28 per 100,000
2009: 16 to 20% of deaths
15 to 26 per 100,000
2007: 14 to 190
2009: 15 to 190
65 ppb
2007: 0.8 to 3.2% of deaths
2.3 to 15 per 100,000
2009: 0.8 to 3.3% of deaths
2.2 to 14 per 100,000
2007: 13 to 19% of deaths
15 to 26 per 100,000
2009: 14 to 19% of deaths
15 to 25 per 100,000
2007: 14 to 150
2009: 15 to 160
60 ppb*
2007: 0.8 to 3.0% of deaths
2.2 to 14 per 100,000
2009: 0.7 to 2.8% of deaths
2.1 to 13 per 100,000
2007: 15 to 19% of deaths
15 to 25 per 100,000
2009: 15 to 18% of deaths
14 to 23 per 100,000
2007: 13 to 96
2009: 14 to 100
*Note that the ranges across the urban areas for the 60 ppb alternative standard level excludes the New York study area, for which we were unable to adjust air
quality to just meet the alternative standard level. As a result, the ranges may not be fully comparable between the alternative standards.
9-4
-------
9.2.1 Air Quality Considerations
Table 9-2 below gives information on the monitoring network, population, and observed
peak Os concentrations for the 15 urban study areas, for the years included in the exposure, lung
function risk, and epidemiology based risk assessments. The number of counties, number of Os
monitors, population, and design values (DV) are based on the area definitions used in the
exposure modeling and clinical-based lung function risk assessments, while the 2007 and 2009
annual 4th highest values are based on the Core Based Statistical Areas (CBSAs) used in the
epidemiology-based risk assessment. The "N/A" values in the 2007 and 2009 4th high columns
are for the three urban areas not included in the epidemiology-based risk assessment. The data
show a trend of lower peak Cb concentrations (i.e., the 2008-2010 design values and 2009 4th
high values are generally lower than the 2006-2008 design values and 2007 4th high values,
respectively).
Table 9-2. Area and Monitoring Information for the 15 Urban Study Areas.
Area Name
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
Number
of
Counties
33
7
10
16
8
11
13
9
10
5
27
15
7
17
26
Number
OfC-3
Monitors
13
7
14
26
13
20
26
12
22
54
31
19
26
17
22
Population
(2010)
5,618,431
2,710,489
5,723,468
9,686,021
2,881,937
6,366,542
3,390,504
5,218,852
5,946,800
17,877,006
21,056,173
7,070,622
2,755,972
2,837,592
5,838,518
2006-2008
DV (ppb)
95
91
83
78
82
89
86
81
91
119
90
92
102
85
87
2007 4th
Highest
(ppb)
102
92
89
N/A
83
N/A
97
93
90
105
94
102
93
94
N/A
2008-2010
DV (ppb)
80
89
77
74
77
86
77
75
85
112
84
83
102
77
81
2009 4th
Highest
(ppb)
77
83
75
N/A
72
N/A
79
73
91
108
81
74
96
74
N/A
In this analysis, we employed a photochemical model-based adjustment methodology
Simon et al. (2013) using the Higher-Order Decoupled Direct Method (HDDM) capabilities in
the Community Multi-scale Air Quality Model (CMAQ) (hereafter referred to as HDDM air
9-5
-------
quality adjustment). The HDDM air quality adjustment methodology replaced the quadratic
rollback technique used in the first draft HREA to estimate Os concentrations consistent with just
meeting existing and alternative Os standards. The HDDM air quality adjustment procedure
estimates the change in observed hourly Os concentrations at a given set of monitoring sites
resulting from national across-the-board reductions in U.S. anthropogenic NOx and/or VOC
emissions. In this analysis, we adjusted Os concentrations to just meet the existing standard of 75
ppb2 and potential alternative standards of 70, 65, and 60 ppb at ambient monitoring sites in the
15 urban study areas for the 2006-2008 and 2008-2010 periods. In most locations, only NOx
reductions were used to adjust the distribution of Cb concentrations, because of the
ineffectiveness of VOC reductions in reducing peak Os concentrations needed to meet the
existing and alternative standard levels. Sensitivity analyses were also conducted in some
locations to evaluate the impact of decreasing both NOx and VOC emissions.
The HDDM air quality adjustment methodology represents a substantial improvement
over the quadratic rollback method used to adjust Os concentrations in previous reviews. First,
quadratic rollback was a purely mathematical technique which attempted to reproduce the
distribution of observed Os concentrations just meeting various standards, while the new
methodology uses photochemical modeling to simulate the response in Os concentrations due to
changes in precursor emissions based on current understanding of atmospheric chemistry and
transport. Second, quadratic rollback used the same mathematical formula to adjust
concentrations at all monitors within each urban study area for all hours, while HDDM allows
the adjustments to vary both spatially across each urban study area and temporally across hours
of the day and across seasons. Finally, quadratic rollback was designed to only allow decreases
in Os concentrations, while the HDDM air quality adjustment allows both increases and
decreases in Os concentrations in response to reductions in NOx or VOC emissions. For example,
in response to reductions in NOx emissions, the HDDM methodology is able to capture increases
in Os concentrations that can occur in urban cores characterized by titration of Os by fresh NO
emissions and decreases in Os concentrations downwind. In addition, the HDDM methodology
removes the need to assume a particular "floor" of background ozone concentrations, as the fully
characterized chemistry in the model accounts for both anthropogenic emissions (domestic and
international) and natural emission sources, and we are able to focus emissions reductions on
U.S. anthropogenic emissions alone in adjusting Os concentrations to just meet existing and
alternative Os standard levels.
: Just meeting the existing standard level of 75ppb is determined by the 4th highest daily maximum 8-hr average O3
concentration, averaged over 3 years (hereafter referred to as the existing standard).
9-6
-------
Following HDDM adjustment of Os concentrations, several general trends are evident in
the changes in Os patterns across the urban study areas and across the alternative standard levels.
In all 15 urban study areas, peak Os concentrations tended to decrease while the Os
concentrations in the lower part of the distribution of Os tended to increase as the concentrations
were adjusted to meet the existing and alternative standards. In addition, Os concentrations in the
high and mid-range portions of the Os distribution generally decreased in the outer, more rural
and suburban portions of the urban study areas, while the Cb response to NOX reductions was
more varied within the urban cores. In particular, while the peak (annual 4th highest daily
maximum 8-hr average) concentrations upon which the existing and alternative standards are
defined generally decreased in the urban core of the urban study areas in response to modeled
reductions in primarily NOx emissions, the Os responses near the center of the Os distribution at
these locations followed one of three patterns when focusing on the mean of the daily maximum
8-hr Os concentrations from May to September, as shown in Table 9-3.
Table 9-3. General Patterns in Seasonal (May-Sept) Mean of Daily Maximum 8-hr
Concentrations after Adjusting to Meet Existing and Alternative Standards.*
After Adjusting Air Quality to
Just Meet Existing Standard
Decreased
Increased
Increased
After Further Adjusting to Just Meet
Lower Alternative Standards
Continued to decrease
Decreased
Continued to increase or remained
constant
Urban Study Areas
Showing Pattern
Atlanta, Sacramento
Washington, D.C
Baltimore, Cleveland
Dallas, Detroit, Los
Angeles. New York
Philadelphia, St. Louis
Boston, Chicago,
Denver, Houston
* These patterns refer to O3 responses in the urban core of each urban study area based on analysis of the
interpolated monitor values used as inputs to the exposure and lung function risk analyses.
The air quality inputs to the exposure modeling and clinical-based lung function risk
assessments were estimated hourly Os concentrations at each census tract in the 15 urban study
areas. These values were interpolated from the observed and HDDM-adjusted monitoring data
using the Voronoi Neighbor Averaging (VNA) technique. This technique was shown to be an
improvement upon the nearest neighbor technique used in the first draft HREA and previous Os
NAAQS reviews (see Appendix 4-A for details). Consequently, the spatial variability of
observed and HDDM-adjusted Os is better accounted for in these analyses compared to those in
the first draft HREA.
The air quality inputs to the epidemiology based risk assessment were "composite
monitor" values, a time-series of the spatially averaged monitoring data, in 12 of the 15 urban
study areas. Consequently, in cases of urban study areas within which Os was predicted to
9-7
-------
increase in some locations and decrease in others, the air quality inputs to this analysis represent
a "net" effect for each urban study area. The spatial extent of the urban study areas used in the
composite monitor averages were CBSAs. These CBSA areas are larger than the Zanobetti and
Schwartz, 2008 (Z & S) study areas (used in the first draft HREA) which include only a subset of
the CBSA focused on urban cores. Figure 9-1 (reproduced from Figure 4-7) shows box plots of
the composite monitor values for 2006-2008 based on the observed data (black), data adjusted to
meet the existing standard using quadratic rollback (blue), and the HDDM adjustment procedure
(red) for the Z & S areas. The two adjustment methods were generally comparable in terms of
the changes in the upper quartile of the distribution. However, by design, quadratic rollback
always estimated decreases in the 75th percentile, median, and 25th percentile of the composite
monitor values, while HDDM estimated decreases in these values in some urban study areas, and
increases in other areas consistent with atmospheric chemistry. HDDM-based adjustments
always produced increases in the lower tail of the distribution, while the lower tail values
generally remained unchanged with quadratic rollback. The differences between the two
adjustment procedures were the most pronounced in Los Angeles and New York, where the
largest reductions in NOX were required in order to meet the existing and alternative standards.
These large reductions in NOX caused a relatively large increase in lower Os concentrations
because of the reduction in NOX titration of Os. As was noted in Chapter 4, the HDDM-based Os
estimates become more uncertain for larger changes in NOx and VOC emissions, and thus there
was less overall confidence in those results. Even in these cases, the HDDM approach is still
preferable because it captures better the overall shift in the distribution of Os concentrations.
9-8
-------
Atlanta: Z & S. April-October. Z006-Z008
Baltimore: Z & S. April-October. Z006-Z008
Boston: Z & S. April-October, Z006-Z008
Cleveland: Z & S. April-October. Z006-Z008
Denver: Z & S. April-October. Z006-ZOOB
Detroit: Z & S, April-October, Z006-ZOOB
• quadratic rollback
• ddm adjustment
I
75
Houston: Z & S. April-October. 2006-2008
LosAngeles: Z & S. April-October. Z006-Z008
NewVork: Z & S. April-October. Z006-Z008
Philadelphia: Z & S, April-October, 2006-2008
Sacramento: Z & S. April-October. Z006-Z008
SaintLouis: Z & S, April-October, Z006-Z008
• quadratic rollback
• ddm adjustment
i
75
Figure 9-1. Distributions of Composite Monitor Daily Maximum 8-hr Average Os
Concentrations from Ambient Measurements (black), Quadratic Rollback (blue), and the
HDDM Adjustment Methodology (red) Used for Just Meeting the Existing Os Standard.
9-9
-------
9.2.2 Human Exposure Modeling
The population exposure assessment evaluates exposures to Os using the Air Pollution
Exposure (APEX) model for the general population, all school-age children (ages 5-18),
asthmatic school-age children (ages 5-18), asthmatic adults (ages > 18), and older adults (ages 65
and older), with a focus on populations engaged in moderate or greater exertion (e.g. children
engaged in outdoor recreational activities). The strong emphasis on children, asthmatics, and
older adults reflected the findings of the last Cb NAAQS review (U.S. EPA, 2007) and the ISA
(U.S. EPA, 2013, Chapter 8) that these are important at-risk groups.
We assessed exposure in 15 urban study areas - Atlanta, Baltimore, Boston, Chicago,
Cleveland, Dallas, Denver, Detroit, Houston, Los Angeles, New York, Philadelphia, Sacramento,
St. Louis, and Washington, D.C. - for recent Os concentrations (2006-2010) and for Os
concentrations adjusted to just meet the existing 8-hr Os standard (75 ppb) and alternative
standard levels of 70, 65, 60 ppb3 for two time periods (2006-2008 and 2008-2010)4. The
analysis provided estimates of the percent of several study groups of interest exposed to
concentrations at or above three controlled human exposure-based 8-hr average Os exposure
benchmarks: 60, 70, and 80 ppb. The ISA includes studies showing statistically significant
effects at each of these benchmark levels (U.S. EPA, 2013). These benchmarks were selected to
provide some perspective on the public health impacts from exposures to various concentrations
that have been associated with Os-related health effects (e.g., lung inflammation and increased
airway responsiveness) in controlled human exposure and toxicological studies, but cannot
currently be evaluated in quantitative risk assessments. In addition, the exposure assessment also
identified the specific microenvironments and activities that contribute most to exposure and
evaluated at what times and how long individuals were in key microenvironments and were
engaged in key activities. This assessment focused on persons experiencing the highest daily
maximum 8-hr exposure within each study area. The assessment found that:
Childhood is an important lifestage where higher exposures and risks can occur, due
to the higher time spent outdoors by children, the higher exposure concentration
3 We were not able to adjust air quality to just meet the 60 ppb alternative standard in the New York study area by
reducing U.S. NOX and VOC emissions (see chapters and appendix 4-D for details). Detroit was already meeting
the existing standard for 2008-2010.
4 Just meeting the Os standard is based on the 4th highest daily maximum 8-hr average Os concentration, averaged
over 3 years. We evaluated two different 3-year periods in determining how air quality in each of the analytical
years would respond to just meeting the existing and alterative levels of the standard. This was done to evaluate
the effect of variability in meteorology and emissions on exposures and risks associated with just meeting the
existing and alternative standards. For the exposure and lung function risk analyses, which provide estimates for
each of the five analytical years, this results in two estimates for 2008, because 2008 is included in each of the 3-
year averaging periods and there are separate analytical results for 2008 for the adjusted air quality resulting from
simulating just meeting the standard in each of the two 3-year periods.
9-10
-------
experienced by children while outdoors (i.e. when they are dismissed from school in
the afternoon and during the summer, when they may be at an outdoor camp all day)
and engagement in moderate or high exertion level activities.
• Persons spending a large portion of their time outdoors during afternoon hours
experienced the highest 8-hr average Os exposure concentrations given that Os
concentrations in other microenvironments were simulated to be lower than ambient
concentrations.
• Highly exposed children spend half of their outdoor time (on average) engaged in
moderate or greater exertion levels, such as in sporting activities. Highly exposed
adults also spent their outdoor time engaged in moderate or greater exertion levels
though on average, not as frequently as children.
Across the 15 urban study areas, we find that school-age children are of greatest concern
for Os exposures compared to other lifestages due to the greater amount of time they spend
outdoors engaged in moderate or higher exertion activities. The exposure analysis estimates that
school-age children have the highest percent of exposures of concern of any of the at-risk
populations or lifestages. As a result, we focus on the results for school-age children (ages 5-18)
in the remainder of this discussion. Figure 9-2 (reproduced from Figure 5-5) shows the results of
the exposure assessment for all 15 urban study areas, showing trends across the analytical years
for the percent of all school-age children with at least one daily maximum 8-hr average Os
exposure at or above the 60, 70, and 80 ppb benchmarks.
The limited availability of longitudinal activity diary data and the general population
modeling approach used may underestimate the correlation in activity patterns for certain
susceptible populations (e.g., outdoor workers), and underestimate how often there are repeated
exposures to Os concentrations above the exposure benchmarks. As a result, although we are
able to report the percent of the population with at least one exposure greater than the alternative
exposure benchmarks, we are less confident in the estimated percent of the population
experiencing more than one exposure. Individuals with repeated exposures may be at greater risk
of significant health effects (U.S. EPA, 2013, Section 6.2.1.1). In addition, the limited data on
responses to air quality alerts (e.g., averting behavior) indicates that a small percentage of the
population may engage in averting behavior in response to air pollution, which may overstate
actual exposures if individuals reduce their exposure during periods of high Os.
The benchmark exposures of concern are not equivalent to ambient standard levels, as
exposures reflect the full pattern of Cb concentrations throughout a season, coupled with time
spent outdoors and indoors engaged in different activities. Thus, just meeting the existing
standard will result in shifts in the entire distribution of Os over a three year period, and will
change the percent of populations experiencing each of the exposure benchmarks of concern.
9-11
-------
Figure 9-2 shows that the percent of all school-age children at or above the 60 ppb exposure
benchmark declines consistently across the 15 urban study areas when just meeting potential
alternative standards of 70, 65, and 60. For most urban study areas and years, the percent of all
school-age children at or above the 60 ppb exposure benchmark is reduced by over half when Cb
is adjusted to meet the 65 ppb alternative standard relative to the 75 ppb standard. In many urban
study areas and years, just meeting the 65 ppb alternative standard results in close to zero percent
of all school-age children above the 60 ppb benchmark. For the 70 and 80 ppb benchmarks,
meeting an alternative standard of 70 ppb results in a small percentage of all school-age children
exceeding the exposure benchmarks.
Year-to-year variability is relatively pronounced for exceedances of the 60 ppb
benchmark. In addition, we observe a geographic pattern to the years with the maximum percent
of exceedances of the exposure benchmarks reflecting the regional Os patterns across years. In
general, northeastern urban study areas saw the highest percentage of benchmark exceedances
during 2007, while southern and western urban study areas saw a higher percentage of
benchmark exceedances during 2006. However, these patterns are somewhat dependent on the 3-
year averaging period used to determine whether the standards are met. In general variability in
the percent of all school-age children exceeding the 60 ppb exposure benchmark across urban
study areas is similar to the variability across years.
The percent of all school-age children with multiple exposures above the exposure
benchmarks is generally much lower compared to the percent of all school-age children with
single exposures above the benchmarks. However, as noted above, we have lower confidence in
these estimates. Even for the lowest benchmark level of 60 ppb, most locations and years have
less than 10 percent of all school-age children experiencing 2 or more exposures when just
meeting the existing standard of 75 ppb, less than 5 percent when just meeting an alternative
standard of 70 ppb, and less than 1 percent when just meeting an alternative standard of 65 ppb.
For most urban study areas and years, less than 1 percent of all school-age children experience 2
or more exposures at or above the 70 ppb exposure benchmark when just meeting the existing
standard.
9-12
-------
0)
0_
<^ <£ <^ <^ <£ <£" £ <£ <^ <£ <^ <£ <^ .I* <£ / & J/ J? ^
Figure 9-2. Effect of Just Meeting Existing (column 1) and Alternative (columns 2 through
4) Standards on the Percent of All School-age Children (ages 5-18) with at Least One Daily
Maximum 8-hr Os Exposure at or above 60, 70, and 80 ppb while at Moderate or Greater
Exertion, 2006-2010 Air Quality.
9-13
-------
Table 5-11 summarized the percent of the all school-age children (ages 5-18) with at least
one daily 8-hr exposure above the 60, 70, and 80 ppb benchmarks, providing both the mean and
maximum percentage across the five analytical years for each urban study area. For Os adjusted
to just meet the existing standard of 75 ppb, the highest maximum percentage of all school-age
children exceeding the 60 ppb benchmark across years, 26 percent, occurs in Denver and St.
Louis, both study areas also having the highest mean percentage averaged across years (about
16-17%). After just meeting the existing standard, Los Angeles has the lowest maximum (10
percent) and mean (9.5 percent) percentage of all school-age children exceeding the 60 ppb
benchmark across years, likely reflecting the highly skewed nature of Os concentrations in that
urban study area. For example, just meeting the existing standards in Los Angeles moves the
majority of Os concentrations (sites and days) well below 60 ppb (See Appendix 4-D). Patterns
across urban study areas are generally similar after just meeting alternative standards of 70, 65,
and 60 ppb, with the exception that the lowest maximum and mean percentage of all school-age
children for the alternative standard level of 65 ppb occurs in the New York urban study area,
which had very large (greater than 90 percent) reductions in NOx emissions that were used to
adjust air quality to just meet the 65 ppb standard level in that urban study area. This resulted in
the distribution of Os concentrations covering most days of the year and most monitoring sites
shifting dramatically downward, with most concentrations well below 60 ppb across the New
York urban study area. The level of confidence in the results for the New York and Los Angeles
urban study areas for just meeting the alternative standards is lower than that for some of the
other urban study areas due to the FtDDM-based Os estimates becoming more uncertain for very
large changes in precursor emissions.
Figure 9-3 (reproduced from Figure 5-25) presents the results of the exposure assessment
for all 15 urban study areas, showing the effect on the percent of all school-age children with one
or more exposures above the 60 ppb benchmark of just meeting the existing and alternative
standards. For each alternative standard, Figure 9-3 shows the maximum percent of all school-
age children exceeding the benchmark across the modeled years 2006-2010. Patterns of results
are similar for the 70 ppb and 80 ppb benchmarks, however, the maximum percents of all school-
age children exceeding those higher benchmarks are much smaller for all alternative standards.
The percent of all school-age children exceeding the 80 ppb benchmark is close to zero once the
existing standard is met. The percent of all school-age children with two or more exposures
exceeding the 60 ppb benchmark level is substantially lower when just meeting the existing
standard, and is close to zero for the 70 ppb and 80 ppb benchmarks. This percentage drops
substantially when meeting the 70 ppb standard, and is close to zero in most urban study areas
when meeting the 65 ppb and 60 ppb alternative standards. Patterns for asthmatic school-age
children are very similar to patterns for all school-age children.
9-14
-------
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20% 22% 24% 26%
Percent of All School-Age Children with at Least One 8-hr Daily Max Exposure >= 60 ppb
standard level (ppb) I 1 60 I 1 65 I 1 70 I 1 75
Figure 9-3. Effect of Just Meeting the Existing (75 ppb) and Alternative Standards on the
Percent of All School-age Children (ages 5-18) at or above the 60 ppb Exposure
Benchmark, Maximum Value Across All Years in Each Study Area, 2006-2010 Air Quality.
9.2.3 Health Risks Based on Controlled Human Exposure Studies
Using the estimates of exposure from APEX combined with results from controlled
human exposure studies, we estimated the number and percent of at-risk populations or lifestages
(all school-age children ages 5-18, children with asthma ages 5-18, adults ages 18-35, adults ages
36-55, and outdoor workers) experiencing selected decrements in lung function. The analysis
focuses on estimates of the percent of each at-risk population or lifestage experiencing a
reduction in lung function (mostly for durations of one to five hours) for three different levels of
impact, 10, 15, and 20 percent decrements in forced expiratory volume in one second (FEVi).
These levels of impact were selected based on the literature discussing the adversity associated
with increasing lung function decrements (US EPA, 2013, Section 6.2.1.1). Consistent with the
exposure assessment, we focus this summary on lung function decrements in children as they are
the lifestage likely to have the greatest percentage at-risk due to higher levels of exposure and
exertion. Within the overall population of children, asthmatic children may have less reserve
lung capacity to draw upon when faced with decrements, and therefore a >10 percent decrement
in lung function may be a more adverse event in an asthmatic child than a healthy child.
Lung function risks (based on experiencing a 10, 15, or 20 percent decrement in lung
function) were estimated in each of the 15 urban study areas. Two methods were used to estimate
9-15
-------
lung function risks: one based on application of a population level exposure-response (E-R)
function consistent with the approach used in the 2008 Os NAAQS review, and one based on
application of an individual level E-R function introduced in this review (the McDonnell-
Stewart-Smith (MSS) model) that incorporates individual differences in physiology, age, and
activity patterns (McDonnell et al., 2012) as well as accounting for the full distribution of
exposures, not simply the maximum. Because the individual level E-R function approach allows
for a more complete estimate of risk (incorporating risk responses at varying activity levels, not
just moderate or greater exertion), we focus on the results of that approach here.
The MSS model as implemented in APEX has a variable that adjusts the lung function
response according to an individual's age. The MSS model was fit using data from subjects who
ranged in age from 18 to 35. Thus, the MSS model is not able to account for differences in lung
function at different age groups between the ages of 5 and 18. However, age does have a
pronounced effect on lung function response in the APEX model. APEX models differences in
physiological parameters due to age, and these result in age-dependent predictions of ventilation
rates, which are used in the MSS model. Ventilation rates also depend on the activities being
performed, which are also age-dependent. As a result of differences in physiology and activities,
the lung function responses vary by age (see Appendix 6-E).
Figure 9-4 (reproduced from Figure 6-7) shows the results of the lung function risk
assessment for all 15 urban study areas, showing trends across the analytical years for the percent
of children with predicted lung function decrements greater than or equal to 10 percent5.
Specifically, Figure 9-4 shows that the percent of children (age 5-18) with greater than or equal
to 10 percent lung function decrement declines consistently across the 15 urban study areas when
just meeting the existing 75 ppb standard, as well as the alternative standards of 70, 65, and 60.
The percent of children at-risk at the 10 percent decrement level remains at or above 10 percent
in many locations after just meeting the 60 ppb alternative standard. The percentage of children
with greater than or equal to a 15 or 20 percent lung function decrement is much lower for all
alternative standards, with close to zero percent of children at-risk when just meeting the
alternative standard of 60 ppb. In general variability in percent of children at-risk across urban
study areas is similar to variability across years.
5 We introduced this new method (relative to the 2008 O3 NAAQS review) for calculating the percent of the at-risk
study groups experiencing lung function decrements, based on modeling individual level responses to O3
exposures. This new method yields higher estimates of the percent of children experiencing lung function
decrements greater than 10, 15, and 20 percent. This may be partly due to the specific data inputs from clinical
studies used to derive the function, but is also to be expected because the new MSS model can reflect greater
sensitivity of children to O3 exposures because it allows for age variability in the relationship between O3 and
FEVi decrements, and younger people are more responsive to O3 exposures than older people.
9-16
-------
o
Al
"c
0)
0)
O
0)
(0
re
0)
re
c
0)
O
0)
O)
re
"o
o
C
0)
e
0)
Q.
^S:
Fev
* FEV10
-e- FEV15
-e- FEV20
Figure 9-4. Effect of Just Meeting the Existing (column 1) and Alternative (columns 2-4)
Standards on the Percent of All School-age Children with FEVi decrements > 10,15, and
20%, 2006-2010 Air Quality.
9-17
-------
Figure 9-5 (reproduced from Figure 6-13) shows the results of the lung function risk
assessment for all 15 urban study areas, showing the effect on the risk of a 10 percent or greater
lung function decrement in children (ages 5-18) of just meeting the existing and alternative Os
standards. For each alternative standard, Figure 9-5 shows the maximum percent risk over all of
the modeled years 2006-2010.
There is no consistent pattern in the percent of children with 10 percent or greater lung
function decrement across urban study areas just meeting the existing standard of 75 ppb. The 5-
year maximum estimated percent of children at-risk ranges from 17 to 22 percent across urban
study areas. The percent reduction in 5-year maximum risk when just meeting the 70 ppb
alternative standard is more consistent across urban study areas, ranging from 8 to 23 percent
(excluding the New York study area, which had a reduction of 29 percent). Reductions in risk
when just meeting the 65 ppb alternative standard are also generally consistent across urban
study areas, with the exception of the New York study area. Incremental reductions in risk when
just meeting the alternative 65 ppb standard compared with just meeting the 70 ppb alternative
standard range from 17 to 31 percent excluding the New York study area, which has a reduction
in risk of more than twice as much as the next largest reduction. Incremental reductions in risk
from just meeting the alternative 60 ppb standard compared with just meeting the 65 ppb
standard are generally consistent, ranging from 16 to 46 percent, with somewhat larger
reductions in risk occurring in Cleveland and Denver. Overall, the 5-year maximum percent of
children at-risk for lung function decrements of 10 percent or more exceeds 13 percent, 10
percent, and 5 percent in all urban study areas except New York after just meeting alternative
standards of 70, 65, and 60, respectively. Patterns of risk reductions are also similar for the
alternative lung function decrement levels of 15 percent and 20 percent. However, the initial
percent of the population experiencing these decrements when just meeting the existing standard
are substantially lower.
Patterns of risk responses using the population level E-R model are similar to the MSS
individual risk model. However, the starting values for the percent of the population at risk are
lower, reflecting the limits of the model in reflecting individual level responses, and the limited
coverage of the model for exposures at lower exertion levels. For children, the MSS model gives
results typically a factor of three higher than the population level E-R model used in the previous
Os NAAQS review.
9-18
-------
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St Louis
Washington
0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20%
percent of school-aged children with FEV1 decrement > 10%
standard level (ppb) I 1 60 I 165 I 1 70 I 1 75
Figure 9-5. Effect of Just Meeting Existing (75 ppb) and Alternative Standards on the
Percent of All School-age Children (ages 5-18) with FEVi Decrements > 10%, Maximum
Value for Each Urban Study Area, 2006-2010 Air Quality.
9.2.4 Health Risks Based on Epidemiological Studies (Chapters 7 and 8)
The epidemiology-based risk assessment evaluated mortality and morbidity risks from
short-term Os exposures and mortality risks from long-term exposures to Os by applying C-R
functions derived from selected epidemiology studies. The analysis included both a set of urban
scale analyses and a national-scale assessment. The urban study analyses evaluated mortality and
emergency department (ED) visits, hospitalizations, and respiratory symptoms associated with
recent Os concentrations (2006-2010) and with Os concentrations adjusted to just meet the
existing and alternative Os standards (see section 9.2.1 and Chapter 4). Mortality and hospital
admissions (HA) were evaluated in 12 urban study areas, while ED visits and respiratory
symptoms were evaluated in a subset of areas with supporting epidemiology studies. The 12
urban study areas were: Atlanta, GA; Baltimore, MD; Boston, MA; Cleveland, OH; Denver, CO;
Detroit, MI; Houston, TX; Los Angeles, CA; New York, NY; Philadelphia, PA; Sacramento,
CA; and St. Louis, MO. The urban study analyses focus on risk estimates for the middle year of
each three-year design value period (2006-2008 and 2008-2010) in order to provide estimates of
risk for a year with generally higher Os concentrations (2007) and a year with generally lower Os
concentrations (2009).
9-19
-------
Most of the endpoints evaluated in epidemiology studies cover the entire study
population including children and adults. Because most mortality and hospitalizations occur in
older persons, these epidemiology-based risk estimates are better indicators of effects in adults
than in children. This is an important distinction from the human exposure and lung function risk
assessments, which focus on children. The only endpoints specific to children are asthma and all
respiratory hospital admissions using the New York specific epidemiology study, respiratory ER
visits in Atlanta, and respiratory symptoms in asthmatic children in Boston.
Both the urban study area and national-scale assessments provide the absolute incidence
and percent of incidence attributable to Os. In addition, risks are presented in terms of incidence
per 100,000 population to control for the differences in the sizes of the populations across urban
study areas, and to allow for comparison of risks using different definitions of urban extent. In
previous reviews, Os risks have only been estimated for the portion of total Os attributable to
North American anthropogenic sources (above what was referred to in previous reviews as
"policy-relevant background Os") In contrast, this assessment estimates risk for Os
concentrations down to zero, reflecting the lack of evidence for a detectable threshold in the C-R
functions (U.S. EPA, 2013, Chapter 2), and the understanding that U.S. populations may
experience health risks associated with Os resulting from emissions from all sources, both natural
and anthropogenic, within and outside the U.S. In order to better reflect how Os distributions are
likely to respond to just meeting existing and potential alternative standard levels, we adjusted
Os concentrations to just meet existing and potential alternative standard levels using reductions
in only U.S. anthropogenic emissions of Os precursors. Thus, the estimated changes in risk
between just meeting the existing standards and just meeting potential alternative standard levels
only reflect reductions in U.S. anthropogenic emissions.
However, consistent with the conclusions in the Os ISA (U.S. EPA, 2013), we have
relatively lower certainty about the shape of the C-R function towards the lower end of the
distribution of Os concentrations used in fitting the function due to the reduction in the number
of Os measurements in this portion of the distribution. We discuss this source of uncertainty
below. In addition, we provide the distribution of mortality incidence across the range of Os
concentrations in Chapter 7 to inform discussions of uncertainty in the results.
9.2.4.1 Urban study area results
Figure 9-6 (reproduced from Figures 7-4 and 7-5) show the results of the mortality and
adult (ages 65 and older) respiratory hospital admissions risk assessments for all 12 urban study
areas, showing the effect on the incidence per 100,000 population just meeting the existing 75
ppb standard and alternative Os standards of 70, 65, and 60 ppb in 2007 and 2009.
9-20
-------
2007 Simulation year
Trend in ozone-related mortality across standard
levels (deaths per 100,000)
-Atlanta, GA
-Baltimore, MD
Boston, MA
-Cleveland, OH
-Denver, CO
Detroit, Ml
-Houston, TX
-Los Angeles, CA
New York, NY
-Philadelphia, PA
-Sacramento, CA
St. Louis, MO
75ppb
70ppb
65ppb
SOppb
2009 Simulation year
Trend in ozone-related mortality across standard levels
(deaths per 100,000)
2 16
g
I"
| 12
| 10
I 8
1 6
1 a
-Atlanta, GA
-Baltimore, MD
Boston, MA
-Cleveland, OH
-Denver, CO
Detroit, Ml
- Houston, TX
-Los Angeles, CA
New York, NY
-Philadelphia, PA
Sacramento, CA
St. Louis, MO
2007 Simulation year
Trend in ozone-related HA across standard levels
(HA per 100,000)
-Atlanta, GA
-Baltimore, MD
Boston, MA
-Cleveland, OH
Denver, CO
Detroit, Ml
- Houston, TX
-Los Angeles, CA
New York, NY
-Philadelphia, PA
Sacramento, CA
St. Louis, MO
2009 Simulation year
Trend in ozone-related HA across standard levels
(HA per 100,000)
Atlanta, GA
Baltimore, MD
Boston, MA
leveland, OH
Denver, CO
Detroit, Ml
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Figure 9-6. Effect of Just Meeting Existing (75 ppb) and Alternative Standard Levels on Mortality Risk per 100,000
Population (left panels) and on Adult (ages 65 and older) Respiratory Hospital Admissions Risk per 100,000 Population, 2007
(top panels) and 2009 (bottom panels) Air Quality.
9-21
-------
In some urban study areas which have large NOx emissions (e.g. from heavy downtown
traffic), Os levels are artificially low because the NOx emissions remove Os through chemical
reactions (see section 9.2.1 and Chapter 4). In these places, when NOx emissions are decreased to
reduce peak Os concentrations across the entire CBS A, which often includes locations outside of
the urban core areas, lower concentrations of Os can go up. This can also happen in other areas
on the lowest Os days. This phenomenon occurs in some locations when meeting lower
alternative standards as well.
The overall trend across urban study areas is small decreases in mortality and morbidity
risk as Os concentrations are adjusted to just meet incrementally lower alternative standard
levels. In New York, there are somewhat greater decreases in these risks, reflecting the relatively
large emission reductions used to adjust air quality to just meet the 65 ppb alternative standard,
and the substantial change in the distribution of Os concentrations that resulted. We were not
able to adjust Os concentrations to just meet the 60 ppb alternative standard in the New York
urban study area. Risks vary substantially across urban study areas; however, the general pattern
of reductions across the alternative standards is similar between urban study areas. Because of
the generally lower baseline Os concentrations in 2009, risks are generally slightly lower in 2009
relative to 2007; however, the patterns of reductions in risk are very similar between the two
years.
Compared with exposures of concern and lung function risks, mortality and morbidity
risks based on epidemiology studies generally do not show as large responses to meeting existing
or alternative levels of the standard for several reasons. First, these risks are based on C-R
functions that are approximately linear along the full range of concentrations, and therefore
reflect the impact of changes in Os along the complete range of 8-hr average Os concentrations.
This includes days with low baseline6 Os concentrations that are predicted to have increases in
Os concentrations, as well as days with higher starting Os concentrations that are predicted to
have decreases in Os concentrations as a result of just meeting existing and alternative standards.
Second, these risks reflect changes in the urban-area wide monitor average, which will not be as
responsive to air quality adjustments as the design value monitor, and which includes monitors
with both decreases and increases in 8-hr concentrations. Third, the days and locations with
predicted increases in Os concentrations (generally those with low to midrange starting Os
concentrations) resulting from just meeting the existing or alternative standard levels generally
are frequent enough to offset days and locations with predicted decreases in Os. The heat maps
presented in Figures 7-2 and 7-3 demonstrate that just meeting progressively lower alternative
6 By low baseline concentrations, we mean area-wide average O3 concentrations between approximately 10 and 40
ppb prior to adjustments to just meet the existing and alternative standards.
9-22
-------
standard levels narrows the distribution of risk across the range of Os concentrations. In addition,
the distribution of risk tends to be more centered on area-wide average concentrations in the
range of 25 to 55 ppb after just meeting an alternative standard of 60 ppb. The focus of the
epidemiological studies on urban study area-wide average Os concentrations, and the lack of
thresholds coupled with the linear nature of the C-R functions mean that in this analysis, the
impact of a peak-based standard (which seeks to reduce peak concentrations regardless of effects
on low or mean concentrations) on estimates of mortality and morbidity risks based on results of
those studies is relatively small. For example, for mortality and hospital admissions (based on
applying C-R functions from the Medina-Ramon et al multicity study), we find a less than or
equal to 5, 10, and 17 percent reduction in risk for most urban study areas (excluding New York)
when just meeting the 70 ppb, 65 ppb, and 60 ppb alternative standards, respectively, compared
to just meeting the existing standard. The general pattern for other morbidity risks is similar to
hospital admissions. However, we are not able to draw strong conclusions about the results
across urban study areas, because of the limited number of urban study areas represented for
most of the endpoints.
We have applied city-specific mortality effect estimates to each urban study area based
on the largest multi-city epidemiological study. However, for many of the urban study areas, the
risk estimates have wide confidence intervals that can include zero, due to the lower statistical
power of some of the city-specific effect estimates relative to the national combined effect
estimate across cities. Furthermore, there is significant variability in these effect estimates across
the 12 urban study areas, with some urban study areas having effect estimates from 5 to 7 times
greater than other cites (see Chapter 7, section 1 A.I)1 The variability in effect estimates, along
with differences in Os concentrations, is a driver for the overall variability in the risk results
across urban study areas. Smith et al (2009) reports an overall significant national mortality
effect estimate with confidence intervals that do not include zero, reflecting the much greater
statistical power available when pooling information across urban study areas.
We also evaluated mortality risks in the 12 urban study areas associated with long-term
Os exposures (based on the seasonal average (April to September) of the peak daily 1-hr
maximum concentrations). Mortality risks from long-term exposures after just meeting the
existing standard are substantially greater than risks from short-term exposures, ranging from 16
to 20 percent of respiratory mortality across urban study areas. This compares with mortality
7 This substantial heterogeneity in effect estimates can reflect a number of factors including differences in population
susceptibility and behavior related to O3 exposure and risk (e.g., proximity to roadways, use of air conditioning,
commuting patterns, time spent outdoors) and differences in the degree to which the O3 monitoring network used
in the epidemiological study reflects patterns of population exposure.
9-23
-------
risks from short-term exposures which range from 1 to 4 percent of total mortality. However, the
percent reductions in long-term mortality risks are similar to those for mortality from short-term
exposures. For example, we find less than or equal to 5 and 10 percent reductions in risk relative
to just meeting the existing standard in most areas (excluding New York) when just meeting the
70 ppb and 65 ppb alternative standards, respectively, and a less than 17 percent reduction when
just meeting the 60 ppb alternative standard level. Risk reductions for the New York urban study
area are much greater when just meeting the 65 ppb alternative standard compared to just
meeting the existing standard, with a 24 percent reduction in risk in 2007.
New York and Los Angeles have characteristics that make epidemiological risk estimates
particularly uncertain. In the case of New York, the expansion of the urban study area definition
to the CBS A adds uncertainty due to the large and diverse nature of the CBS A. The New York
CBS A includes two urban study areas which have separate effect estimates available from the
Smith et al. (2009) study. These separate effect estimates (for Newark, NJ and Jersey City, NJ)
are smaller than the effect estimate for New York, however, they are also based on much smaller
populations, and have relatively wider confidence bounds, reflecting low statistical power. For
consistency with other urban study areas and to allow for comparison between the CBSA-based
risk estimates and the smaller study area based estimates (see the sensitivity analyses in Chapter
7), we elected to apply the New York city effect estimate, which is based on a very large
population and has high precision, to all of the counties in the New York CBSA. While this adds
substantial uncertainty to the absolute incidence of mortality for the New York CBSA, it does
not affect the pattern of risk reductions when just meeting alternative standards. In addition, as
noted earlier, the Os adjustments to meet existing and alternative standards in New York and Los
Angeles also have additional uncertainties relative to the other 10 urban study areas.
We conducted a number of sensitivity analyses based on a population normalized
mortality risk metric, e.g. mortality risk per 100,000 population. Maintaining the general linear,
no-threshold functional form, mortality risks per 100,000 population are generally robust to
alternative specifications of the C-R functions, although in several urban study areas, using effect
estimates from Smith et al. (2009) which were derived using regional priors rather than national
priors results in higher risk estimates.8 Using the effect estimates from Zanobetti and Schwartz
: In Bayesian modeling, effect estimates are "updated" from an assumed prior value using observational data. In the
Smith et al (2009) approach, the prior values are either a regional or national mean of the individual effect
estimates obtained for each individual city. The Bayesian adjusted city-specific effect estimates are then
calculated by updating the selected prior value based on the relative precision of each city-specific estimate and
the variation observed across all city-specific individual effect estimates. City-specific estimates are pulled
towards the prior value if they have low precision and/or there is low overall variation across estimates. City-
specific estimates are given less adjustment if they are precisely estimated and/or there is greater overall variation
across estimates.
9-24
-------
(2008) has no consistent effect on risk results across the urban study areas. Using effect estimates
based on a co-pollutant model with PMio, mortality risks are higher in some locations and lower
in others. However, in all locations the confidence intervals are substantially wider using the co-
pollutant model with PMio (due to fewer days with both pollutants measured), which makes it
difficult to determine whether the increases and decreases in estimates relative to the core
estimates are real or the result of statistical error.
We selected the CBS A as the spatial definition for the urban study areas. We made this
selection to address a downward bias that we identified resulting from a mismatch between the
smaller urban core areas used in the epidemiology studies and the larger areas where Os
concentrations are expected to change as a result of meeting the existing and alternative standard
levels (see Chapter 7). We included a sensitivity analysis evaluating the result of using a smaller
geographic area including only the counties used in the epidemiology study. As expected, using a
smaller geographic extent for the urban study areas results in smaller, and in some cases negative
risk reductions when compared to using the CBSA definitions. This reflects the fact that the
controlling monitor9 in many of the 12 urban study areas is located outside of the small set of
counties included in the Smith et al. (2009) urban study area definitions, and some of the
monitors that are within that more limited spatial extent are more prone to Cb titration due to
local NOx emission sources. As a result, those monitors are more likely to see increases in Os
which will, if other monitors with higher concentrations in the broader regions are not included,
lead to estimated increases in risk due to the application of a linear, no threshold C-R function.
This bias can be substantial, especially in St. Louis and several urban study areas in the
Northeast, including Boston, New York, and Philadelphia, where the highest concentration
monitors are outside the Smith et al. (2009) urban study area definitions.
Sensitivity analyses were conducted for scenarios of just meeting existing and alternative
standards using combinations of NOx and VOC emissions reductions (as compared to NOx
reductions alone). The addition of VOC emissions reductions had little impact with the exception
of New York and Los Angeles, where risk was decreased relative to the NOx-only reduction
scenario.
9.2.4.2 National-scale assessment results
The national-scale assessment evaluated only mortality associated with recent Os
concentrations across the entire U.S for 2006-2008. The national-scale assessment is a
complement to the urban scale analysis, providing both a broader geographic assessment of Os-
related health risks across the U.S., as well as an evaluation of how well the 12 urban study areas
' The controlling monitor is the monitor with the highest design value within a defined non-attainment area.
9-25
-------
represented the full distribution of Os-related health risks in the U.S. The national-scale
assessment demonstrates that there are Os risks across the U.S, not just in urban study areas,
even though the Os concentrations in many areas were lower than the existing standard level.
While we did not assess the changes in risk at a national level associated with just meeting
existing and alternative standards, just meeting existing and alternative standards would likely
reduce Os concentrations both in areas that are not meeting those standards and in locations
surrounding those areas, leading to risk reductions that are not included in the urban-scale
analysis.
9.3 COMPARISON OF RESULTS ACROSS EXPOSURE, LUNG FUNCTION RISK,
AND EPIDEMIOLOGY-BASED MORTALITY AND MORBIDITY RISK
ANALYSES
In considering the overall results across the human exposure, lung function risk, and
epidemiology-based risk assessments, we focus on the key policy-relevant metrics and levels for
each type of assessment. For the human exposure assessment, we selected exposures above the
60 ppb exposure benchmark for all children (ages 5-18). We select this exposure metric because
children represent a key at-risk lifestage, and the 60 ppb exposure benchmark is the lowest
exposure level associated with significant findings in controlled human exposure studies. For the
lung function risk assessment, we selected the results for lung function decrements greater than
or equal to 10 percent for all children (ages 5-18). We select this lung function risk metric
because children represent a key at-risk lifestage, and a 10 percent lung function decrement
represents a potentially more adverse event in asthmatic children. For the epidemiology-based
risk assessment we selected the core short-term exposure mortality results and the respiratory
hospital admission results, because these endpoints were estimated for all of the 12 urban study
areas. Generally speaking, these metrics provide the most differentiation between the alternative
standards, helping to inform policy-relevant questions regarding adequacy of the existing
standard, and public health impacts of meeting alternative standards. The other metrics analyzed
in this FIREA (e.g. other exposure benchmarks and other lung function decrements) show less
response to just meeting the existing standard or potential alternative standard levels.
As discussed in Chapter 2, we designed the exposure and risk assessment to help inform
two fundamental questions related to the adequacy of the existing standard in protecting public
health and the degree of exposure and risk reductions associated with alternative standards
compared with the existing standard. The following discussion evaluates the three types of
analyses we conducted in terms of the consistency of the information provided to inform these
questions.
9-26
-------
9.3.1 Evaluation of Exposures and Risks after Just Meeting the Existing Standard
To compare the results of the three assessments in urban study areas, we plot the key
metrics from each analysis across urban study areas for the two common years of analysis (i.e.,
2007 and 2009). For three urban study areas (i.e., Chicago, Dallas, and Washington D.C.) we
have only the exposure and lung function risk assessments, as these urban study areas did not
have sufficient information to estimate epidemiology-based risks. The epidemiology-based
metrics are the percent of baseline short-term exposure mortality, based on the core estimates
using the C-R functions from Smith et al. (2009), and respiratory hospital admissions based on
the core estimates using the C-R functions from Medina-Ramon (2006), attributable to Ch.
Figure 9-7 presents the exposures and risks after just meeting the existing standard of 75 ppb.
Each row represents one of the key analytical results; each column gives the results for 2007 and
2009 for each urban study area. The scale of each analytical metric for each analysis differs, and
thus the comparisons across analyses should focus on overall patterns rather than on direct
comparisons of numeric estimates.
All of the metrics show substantial variability among urban study areas, although there
appears to be less variability in lung function risk and hospital admission risk compared with the
exposure metric and mortality risk. The differences between estimates for 2007 and 2009 are
much higher for some urban study areas (e.g., Baltimore and Philadelphia) for the exposure
metric than any of the risk metrics. This may reflect the explicit threshold nature of the exposure
metric, which focused on daily maximum 8-hr average exposures above a benchmark level of 60
ppb. Differences between years in exposures at or above the 60 ppb benchmark after adjusting
air quality to just meet the existing standard are dependent on the number of days during each
year with decreases in higher Os concentrations, as well as the magnitude of the decreases in Os
on those higher Os concentration days. These in turn are sensitive to the shape of the Os
distribution in the analytical year prior to just meeting the existing standard (which determines
the starting number of days above 60 ppb) and the response to emissions reductions applied in
meeting the existing standard for 2007 or 2009. There is some consistency between metrics in
the urban study areas with highest values for the exposure and lung function risk metrics.
However, there were still differences, especially for Los Angeles, which had one of the higher
values for lung function risk in 2009, but had one of the lower percentages of school-age
children exposed at or above the 60 ppb benchmark. This again points to the importance of the
threshold nature of the exposure metric, combined with the tendency for more substantial
decreases in peak Os concentrations relative to mid-range and low concentrations when adjusting
air quality to just meet the existing standard.
There is little consistency within urban study areas between the epidemiology risk
metrics and the exposure and lung function risk metrics, and there is also little consistency
9-27
-------
between the mortality and hospital admission risks. Houston has the lowest metric values in 2007
(except for mortality risk), but in 2009 has some of the higher risk metrics (except for hospital
admission risk). New York has the highest mortality risk in 2007 and 2009 but has among the
lowest hospital admission risks in both years.
9-28
-------
Chicago
Los Angeles
Philadelphia
Washington
I
I
D
D
D
I
D
I
I
D
Year
pi 2007
B2009
DD
D
15
81
D
.
8|
•3
Figure 9-7. Comparison of Exposure (Row 1), Lung Function Risk (Row 2) and Epidemiology-Based Risk (Rows 3 and 4)
Metrics after Just Meeting the Existing 75 ppb Standard.
9-29
-------
9.3.2 Reductions in Exposure and Risk Metrics after Just Meeting Alternative Standards
To compare the results of the three assessments for urban study areas after adjusting air
quality to just meet alternative standards relative to the existing standard, we express each result
as a percent of the metric value when just meeting the existing standard. Figure 9-8 presents the
percent reduction in exposures and risks after just meeting alternative standards relative to just
meeting the existing standard of 75 ppb. In this plot, each row represents one of the key
analytical results and each column gives the results for 2007 and 2009 for each urban study area.
The scales are the same between analyses, and as such, it is informative to examine both the
overall patterns of change between alternative standards, and also the absolute value of the
percent reductions in risk metrics between analyses. In interpreting this chart, higher values
mean greater reductions in risk or exposure relative to just meeting the existing standards.
Because these are percent reductions, the maximum value is one hundred percent, which if
reached would indicate that risks or benchmark exposures are completely eliminated when the
alternative standard is met in the urban study area as was seen for the 60 ppb exposure
benchmark.
Many of the differences in results across the metrics are driven by how each metric is
affected by the Os data input to the analysis. In general, the impact of the HDDM adjustments to
Os vary based on three main considerations: (1) the degree to which the exposure or risk metric
is sensitive to changes across the various ranges of Os concentrations (e.g. high, mid-range, low);
(2) whether the exposure or risk metric uses individual census tract concentrations or area-wide
average concentrations; and (3) changes in the distribution of Os concentrations in the year of
analysis between recent Os concentrations and adjusted (meeting the existing or alternative
standards) Os scenarios. With respect to consideration (1), the exposure benchmark metric,
which focuses only on exposures above 60 ppb, will not be sensitive at all to changes in Os
concentrations in the range below 60 ppb. The lung function risk metric, which depends on the
dose rate and individuals' characteristics, does not have a concentration threshold. However,
because of the logistic form of the response function, it is less sensitive to lower Os
concentrations and has very few FEVi responses greater than 10 percent when exposure
concentrations are below 20 ppb and very few FEVi responses greater than 15 percent when
exposure concentrations are below 40 ppb. On the other hand, the mortality and hospital
admission risk metrics are based on non-threshold, approximately linear C-R functions, and
therefore will be sensitive to changes in Os along the full range of Os. As discussed in Chapter 4,
because Os at lower concentrations may increase following HDDM adjustment in some locations
9-30
-------
and on some days to just meet alternative standards,10 this can lead to increases in risk on some
days, which can lead to a net increase or decrease in risk over the entire year, depending on
whether the days with increased risk exceed days with decreased risk (generally due to a
preponderance of days with lower Os concentrations). With respect to consideration (2), the
exposure and lung-function risk metrics are based on concentrations at individual census tracts
since they depend on Os exposure modeled by moving each individual through their
environment. Because of this, the exposure and lung-function risk metrics are most affected by
the spatial and temporal variability of Os concentrations across the urban study area. The
mortality and hospital admission risk metrics are calculated applying C-R functions to area-wide,
daily maximum 8-hr average Os concentrations. As a result, the spatial variability in Os
concentrations between the monitors will only influence the epidemiology-based risk estimates
in how they influence the area-wide average. With respect to consideration (3), all three metrics
are influenced by how the distribution of Os concentrations changes between recent Os
conditions and after adjustment to just meet existing and alternative Os standard levels.
The exposure and lung function risk metrics are most affected by the reductions in the
individual monitors' peak Os concentrations, including the magnitude of these reductions and the
number of days that experience these reductions. In contrast, the mortality and hospital
admission risk metrics are affected by changes in the mean of the seasonal, area-wide average Os
concentrations, where the mean is determined by the frequency and magnitude of increases
versus decreases in area-wide, daily maximum 8-hr Cb concentrations.11 In addition to Cb
concentrations, there are other factors that affect the variability across urban study areas for these
three metrics, such as activity data and exposure factors for the exposure and lung function risk
metrics and the study-specific C-R functions for the mortality and hospital admission risk
metrics.
10 The frequency and magnitude of increases in spatially averaged mean concentrations in an urban study area occur
during a season when adjusting air quality to just meet a standard vary considerably between the existing and
alternative standards. The highest frequency of occurrence of days with increasing Os happens when adjusting air
quality to just meet the existing standards, and decreases as air quality is further adjusted to just meet lower
alternative standard levels.
11 As noted previously, changes in the spatial extent of the urban study areas over which monitors are averaged can
change the magnitude and sign of the change in the spatial average O3 concentration for an urban study area. For
example, we found that we bias the risk estimates low when using urban study area definitions that include only
urban core counties and not the counties with monitors experiencing the most reductions in Os
9-31
-------
75-
I
•o
S 50-
co
g
«> 25-
imore Boston Chicago Cleveland Dallas Denver Detroit Houston Los Angeles New York Philadelphia Sacramento St Louis Washington
=££1,
-^ Alternative Standard
2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009
Analytical Year
Figure 9-8. Comparison of the Percent Reduction in Key Risk Metrics for Alternative Standard Levels Relative to Just
Meeting the Existing 75 ppb Standard.
9-32
-------
One clear observation is that the percent reductions in risk from meeting alternative
standard levels relative to meeting the existing standard for the two epidemiology-based
endpoints are much smaller than for the exposure benchmark and lung function risk endpoints.
The maximum percent reduction in the mortality and hospital admissions risk relative to just
meeting the existing standard across years, locations, and alternative standards is less than 25
percent, and for many years/locations, the reductions in these risks when just meeting the lowest
alternative standard, 60 ppb, are less than 10 percent. The exposure benchmark results show the
most reductions when comparing just meeting the existing standard to just meeting alternative
standards. Just meeting the 65 ppb standard results in reductions in the percent of children
exceeding the 60 ppb exposure benchmark by over 50 percent in all urban study areas, and by
over 75 percent in 12 of the 15 urban study areas evaluated. For most locations and years, just
meeting the 60 ppb alternative standard reduced the percent of children exceeding the 60 ppb
exposure benchmark by over 90 percent compared to just meeting the existing standard.
Reductions in lung function risk were also much higher than reductions in mortality and hospital
admissions risk. Air Quality just meeting the 65 ppb standard results in reductions in lung
function risk by over 25 percent in most locations and years, and just meeting the 60 ppb
standard results in reductions by over 40 percent in most locations and years.
There is general consistency in the study area-to-study area patterns of reductions in the
exposure and lung function risk metrics, although the decreases in lung function risk are less than
half as large as the reductions in the percent of all school-age children exceeding the 60 ppb
exposure benchmark (with the clear exception of the New York study area, which we will
discuss further below). The patterns of reductions in mortality and hospital admission risk are
generally consistent with the patterns for exposure and lung function risk for 2007, with the
exception of Houston and Philadelphia. However, for 2009, the patterns for mortality and
hospital admission risk are quite different, both from the 2007 results, and from the exposure and
lung function risk results. This is due to the generally lower Cb concentrations in 2009, which
results in a greater number of days with predicted increases in Os concentrations at low
concentrations, fewer days with very high concentrations where predicted reductions in Os occur,
and a smaller predicted decrease in Os concentrations on those high days. This affects the
mortality and hospital admissions risk more than the exposure and lung function risk metrics
because those metrics incorporate thresholds, and therefore are not responsive to changes in Os
concentrations below those thresholds.
Additional considerations are important in interpreting the reduction in exposure and risk
between the existing standard and alternative standards. The HREA analyses focus on reducing
peak Os concentrations, in particular the 4th high Cb concentration averaged over 3-years so as to
simulate meeting the existing standard or various alternative standards. In addition, the air
9-33
-------
quality adjustments are based on applying reductions in U.S. anthropogenic emissions. In this
way, the adjusted air quality reflects day-to-day Os concentrations that could occur when
focusing on reducing high Os concentrations rather than on reducing mean Cb concentrations. In
addition, because the analyses do not include reductions of Os precursor emissions from sources
other than U.S. anthropogenic emissions (e.g. international emissions, biogenics, etc), the Os
concentrations in the adjusted air quality account for Os created from natural and international
sources, even if 100 percent emissions reductions are applied to U.S. anthropogenic sources in
adjusting air quality scenarios.
Finally, with respect to the epidemiology based analyses, we note that 2007, which had
generally higher Os concentrations than 2009, had more days where Os concentrations decreased
as a result of adjusting peak Os concentrations to just meet alternative standards. Thus just
meeting alternative standards resulted in net decreases in risk in all locations, with the exception
of Houston for just meeting the 70 ppb alternative standard. In contrast, 2009, which had
generally lower concentrations than 2007, had more days in the range where Os concentrations
were increased as a result of adjusting peak Os concentrations to just meet alternative standards,
and thus the patterns reflect some locations where mortality and hospital admissions risk
increases. However, for 2009, in all locations, when just meeting the lowest alternative standard
of 60 ppb, mortality and hospital admission risks are decreased relative to just meeting the
existing standard.
9.4 OVERALL ASSESSMENT OF REPRESENTATIVENESS OF EXPOSURE AND
RISK RESULTS
9.4.1 Representativeness of Selected Urban Study Areas in Reflecting Area across the
Nation with Elevated Risk
We selected urban study areas for the exposure and risk analyses based on several criteria
(e.g. recent elevated Os concentrations and presence of at-risk populations and lifestages) we
identified as likely indicators of areas and populations likely to experience high Os exposures
and risks (see Section 7.3.1). We then conducted several analyses to determine the extent to
which our selected urban study areas actually represent the highest mortality and morbidity risk
areas. We compared the distributions of risk characteristics12 and mortality risk (based on recent
Os concentrations) for the 12 urban study areas used in the epidemiology-based risk assessment
with the corresponding national distributions. We also evaluated the degree to which our selected
12 In this context, risk characteristics are the elements of populations, air quality, and inputs to the C-R functions that
are expected to be correlated with estimated mortality risks (see Chapter 8).
9-34
-------
urban study areas represent the patterns of Os concentration changes experienced by the overall
U.S. population.
Based on the comparisons of distributions of risk characteristics, the selected urban study
areas represent urban study areas that are among the most populated in the U.S., have relatively
high peak Os concentrations, and capture well the range of city-specific mortality risk effect
estimates. These three factors alone would suggest that the urban study areas should capture well
the overall risk for other heavily populated urban study areas in the nation, with a potential for
better characterization of the high end of the risk distribution. The selected urban study areas do
not include those with the highest numbers of some at-risk populations or lifestages, specifically
older people with high baseline mortality rates. However, most locations in the U.S. (except
Florida) with high percentages of older people have low overall populations, less than 50,000
people in a county, or low Cb concentrations. This suggests that while the risk per exposed
person per ppb of Os may be higher in these locations, the overall risk to the population is likely
to be within the range of risks represented by the urban study locations.
Based on the comparisons of distributions of short-term Os exposure mortality risk (using
the percent of mortality metric) for recent Os concentrations, the 12 selected urban study areas
are representative of the full distribution of U.S. Os-related mortality risk in urban study areas.
Two of the selected areas, New York and Philadelphia are representative of the highest end of
the distribution of short-term Os mortality risk. Overall, Os mortality risk for short-term Os
exposures in the 12 urban study areas are representative of the full distribution of U.S. urban Os-
related mortality, representing both high end and low end risk counties. For the long-term Os
exposure mortality risk metric (again using the percent of mortality), the 12 urban study areas are
representative of the central portion of the distribution of risks across all U.S. counties; however,
the selected 12 urban study areas do not capture the very highest (greater than 98th percentile) or
lowest (less than 25th percentile) ends of the national distribution of long-term exposure-related
Os-related risk.
9.4.2 Representativeness of Selected Urban Study Areas in Reflecting Responsiveness of
Risk to Just Meeting Existing and Alternative Os Standards
While we selected urban study areas to represent those populations likely to experience
elevated risks from Os exposure, we did not include among the selection criteria the
responsiveness of Os in the urban study area to decreases in Cb precursor emissions that would
be needed to just meet existing or alternative standards.
In our preliminary evaluations of risk modeling results, we observed a consistent
presence of days with low to midrange starting Os concentrations for which Os concentrations
(using the 8-hr maximum metric) increased after adjustments to just meet the existing and
9-35
-------
alternative standards across the selected urban study areas. As noted above, this led to estimates
of increased risk on those days, and in some cases, estimates of increased risk over the course of
the Os season, reflecting both the magnitude and frequency of the predicted increases relative to
the predicted decreases in Os concentrations. As explained above, this pattern was more
pronounced when using a more spatially limited definition of the urban study areas, but even
when using the CBS A definitions, there were still days when the area-wide average Os increased,
primarily due to predicted increases in Os in the core counties of the urban study areas.
In order to better understand how prevalent this type of air quality response was across
the U.S., we conducted several additional analyses of Os concentrations. These included
evaluations of trends at Cb monitors during a period of time with significant Os precursor
emission reductions, and evaluations of temporal and spatial patterns of Os changes across the
U.S., based on air quality modeling results, to simulate how Os would change across the U.S. in
response to NOX (and VOC) emissions reductions (relative to recent 2007 levels) similar to those
used in the HDDM adjustments (see section 9.2.1 above). The latter analysis includes an
assessment of the association of different types of Os responses with population counts to help
characterize the degree to which populations in the U.S. experience Os conditions like those in
the selected 15 urban study areas (see Chapter 8).
Overall, both types of analyses showed that decreases in Os precursor emissions lead to
decreases in Os concentrations in areas with higher starting Os concentrations, which tend to be
rural or suburban study areas, and on days with higher Os concentrations. The analyses also
indicate that in urban core areas (those with high levels of fresh NOX emissions), decreases in
NOx emissions can lead to increases in Os, primarily for days when initial Os concentrations are
suppressed due to NOx titration. This is consistent with conclusions of the Os ISA (U.S. EPA,
2013, Sections 3.2.4, 3.6.2.1) that Os can be suppressed in high NOx environments. The
observed widespread decreases of median Os in suburban and rural locations when NOx
emissions are decreased suggest the efficacy of large NOx emissions reductions on reducing Os
over large regions of the country.
These results suggest that many of the urban study areas may show Os responses that are
typical of other large urban study areas in the U.S., but may not represent the response of Os in
other populated areas of the U.S., including suburban study areas, smaller urban study areas, and
rural areas. These smaller urban study areas would be more likely than our urban study areas to
experience area-wide average decreases in mean Os concentrations as Os standards are met. Even
though large urban study areas throughout the U.S. have high population density, 73 percent of
9-36
-------
the U.S. population lives outside of these high population density areas13, and thus, a large
proportion of the population is likely to experience greater mortality and morbidity risk
reductions in response to reductions in 8-hr Os concentrations than are predicted by our modeling
in the selected urban study areas. The analyses presented in Section 8.2.3.2 show that
populations in the study areas we selected are approximately twice as likely to experience
increasing mean Os concentrations as populations in the U.S. as a whole. Because our selection
strategy for risk modeling was focused on identifying areas with high risk, we tended to select
large urban population centers. As discussed in the previous section, this strategy was largely
successful in including those urban study areas in the upper end of the Os risk distribution.
However, this also has led to an overrepresentation of the populations living in locations where
we estimate increasing mean seasonal Os in response to adjusting air quality to just meet the
existing and alternative standards using NOx emissions reductions. The implication of this is that
our estimates of mortality and morbidity risk reductions for the selected urban study areas are
likely to understate the average risk reduction that would be experienced across the population
and should not be seen as representative of potential risk reductions for most of the U.S.
population.
9.5 OVERALL ASSESSMENT OF CONFIDENCE IN EXPOSURE AND RISK
RESULTS
As with any complex analysis using estimated parameters and inputs from numerous data
sources and models, there are many sources of uncertainty that may affect our exposure and risk
estimates. These sources of uncertainty are discussed in each of the chapters related to air
quality, exposure, lung function risk, and epidemiology based mortality and morbidity risk. The
overall effect of the combined set of uncertainties on confidence in the interpretation of the
results of the analyses is difficult to quantify. However, we provide our judgment of our overall
confidence here, with an understanding that alternative judgments may also be supported.
The degree to which each analysis was able to incorporate quantitative assessments of
uncertainty differed, due to differences in available information on uncertain parameters and
complexities in propagating uncertainties through the models. In general, we followed the World
Health Organization tiered approach to uncertainty characterization (WHO, 2008), which
includes both quantitative and qualitative assessments. Each chapter includes a table identifying
and characterizing the potential impact of key uncertainties on risk estimates, including the
degree to which we were able to quantitatively address those uncertainties.
13 High population density areas are defined here as locations with population densities greater than 1000
people/km2.
9-37
-------
In considering our overall confidence in the results, there are several key considerations
discussed below related to sources of uncertainty which we were not able to fully quantify, but
which may have a large impact on both overall confidence and confidence in individual analyses.
9.5.1 Uncertainties in Modeling Ozone Responses to Meeting Standards
There is inherent uncertainty in all deterministic air quality models, such as CMAQ, the
photochemical grid model used to develop the model-based Os adjustment methodology.
Evaluations of air quality models against observed pollutant concentrations build confidence that
the model performs with reasonable accuracy despite both structural and parametric
uncertainties. A comprehensive model performance evaluation provided in Appendix 4-B shows
generally acceptable model performance that is equivalent to or better than typical state-of-the
science regional modeling simulations described in Simon et al. (2013). Two additional sources
of uncertainties in the HDDM adjustment methodology are the applicability of HDDM
sensitivities over large emissions perturbations and the variability in data used to create
regressions which allowed the application of these sensitivities to ambient data.
Both sources of uncertainty are shown to be reasonably small in chapter 4 with the first
having a mean error of less than Ippb for 50% NOx cuts and less than 4 ppb for 90% NOx cuts.
The regressions, which were developed to relate Os response to emissions perturbations with
ambient Os concentrations for every season, hour-of-the-day, and monitor location, are purely
empirical so applying Os responses to ambient data based on this modeled relationship adds
uncertainty. Preliminary work showed that the relationships developed with these regressions
were generally statistically significant for most season, hour-of-the-day, and monitor location
combinations for 2005 modeling in Detroit and Charlotte (Simon et al, 2013). Despite the
empirical nature of the regressions, the statistically significant fits and the operational evaluation
of the model which shows satisfactory ability of the model to predict changes in ozone under
varying meteorology and emissions conditions (Appendix 4-B) build confidence in the
application of these modeled relationships to ambient data. The uncertainty introduced from the
application of regressions to determine sensitivities were quantified by propagating uncertainties
in the sensitivities through to uncertainties in the final predicted Os concentrations which had
standard errors less than 1.4 ppb for all adjustment scenarios. New York and Los Angeles had
the largest uncertainties in these two areas due to the fact that they required the largest reductions
in NOx emissions. Uncertainties stemming from the application of 8-months of model data to 5-
years of ambient data and the across-the-board emissions cut assumptions are further discussed
in chapter 4 but are not expected to substantially degrade confidence in the air quality results.
9-38
-------
9.5.2 Uncertainties in Modeling Exposure and Lung Function Risk
With regard to the exposure and lung-function risk estimates, the modeling explicitly
incorporates population variability in many of the modeling inputs. We did not attempt to
probabilistically incorporate the many sources of uncertainty in model parameters or input data
due to limitations in the ability to specify distributions characterizing our confidence in those
variables. To explore the impacts of some of the more important sources of uncertainty, we
conducted a limited set of sensitivity analyses. For the exposure assessment, the estimate of
repeated exposures above exposure benchmarks is based on the limited set of diaries of activity
data available in the Consolidated Human Activity Database (CHAD) database (see Chapter 5).
The method for constructing activity patterns over the course of an Os season may not fully
capture the behavior of children who have systematically high outdoor activity levels. As a
result, while we are able to report the percent of children with two or more exposures, modeling
of the distribution of multiple exposures is limited, and the ability to identify the percent of the
population with unusually high numbers of multiple exposures is not possible.
For the lung function risk assessment, sensitivity analyses indicate that the MSS model
parameter related to the impact of the ventilation rate was most influential in determining the
estimated number of children with FEVi decrements greater than 10 percent. Estimates of lung
function decrements are also influenced by how much variability in individual response is
assumed in the MSS model. Sensitivity analyses indicate that when a greater amount of
variability is allowed in the MSS model, the percent of children ages 5-18 with FEVi decrements
greater than 10 percent can increase substantially. In addition, we performed analyses to
understand the age-related factors in APEX that could influence the estimated FEVi decrements.
It was found that the four most influential factors influencing the relationship between the
predicted FEVi decrement and age are the decreasing level of exertion, the decreasing equivalent
ventilation rate (with increasing age), the higher time spent outdoors by children, and the higher
exposure concentration experienced by children while outdoors. These all lead to children having
higher FEVi decrements than adults, and are more influential than the MSS model age term.
9.5.3 Uncertainties in Modeling Epidemiological-Based Risk
A major issue in using the results of the epidemiology studies in estimating risk is the
narrow geographic definition used for urban study areas in the epidemiology studies. In many of
the urban study areas, we observe two distinct patterns of Os response to the reductions in
precursor emissions we evaluated to just meet the existing and alternative standard levels. The
first pattern generally occurs in areas outside the urban core (e.g. suburban and rural areas), and
on days when Os concentrations are on the higher end of the distribution of Os concentrations,
and is characterized by predicted decreases in 8-hr Os concentrations. These tend to be the
9-39
-------
locations where the highest 8-hr design values occur. The second pattern generally occurs in the
urban core, and on days when Os concentrations are on the lower end of the distribution of Os
concentrations, and is characterized by predicted increases in 8-hr Os concentrations. The narrow
definitions of urban study areas used in the epidemiological studies generally included the urban
core areas, but did not include all of the suburban or rural areas. The narrow geographic
definitions led to a clear downward bias in the estimates of risk changes that would be associated
with just meeting the standards in the urban study areas, because the risk changes would reflect
the locations with a tendency towards increases in 8-hr Os, but would not include locations
outside the urban core with decreases in Os. In many cases, the narrowly defined geographic
definitions used in the epidemiology studies did not even include the location with the monitor
that was violating the standard. We addressed this bias by expanding the urban study area to the
CBS A. However, this adds additional uncertainty to the risk estimates, and reduces our
confidence that we have a good match between the basis of the C-R function (just urban core
locations) and the risk analysis context (including both urban core counties and other counties in
the CBSA). A clear implication of this decision is that the absolute incidence estimates will be
larger than if the analysis was limited to a smaller number of counties. For this reason, we have
placed more emphasis on risk metrics that have been normalized for population size (e.g. risks
per 100,000 population and percent risk), so as to facilitate comparisons between study areas of
different population sizes and to reduce the influence of population size on the risk metrics.
The epidemiology studies used as the source for C-R functions for short-term exposure
mortality and morbidity endpoints all use time-series approaches to estimate the effect of daily
variations in Os concentrations on daily mortality or morbidity incidence. The effect estimates
developed in these epidemiology studies were based on air quality and health information
observed from 1987-2000. These effect estimates were based on day-to-day variations in area-
wide observed Os ambient monitoring concentrations that reflect a specific set of emissions and
atmospheric conditions. In our analyses, we apply these effect estimates to adjusted air quality
scenarios that are reflective of substantial changes in Os concentrations across an area due to, in
some cases, large decreases in NOX and VOC emissions reductions. The resulting spatial and
temporal patterns of Os modeled here may not be the same as the spatial and temporal patterns of
Os that existed at the time of the epidemiology study. The potential for different spatial and
temporal patterns in Cb concentrations between the adjusted air quality scenarios and the air
quality observed during the epidemiology study period potentially adds uncertainly to the
estimates of risk, as it is not clear the degree to which the exposure surrogate used in the
epidemiology study correlates with the exposure surrogate used in the risk analysis. However,
the impact of these differences in temporal and spatial patterns is not likely to be great, and is
likely to be less than other sources of uncertainty in the analysis. The use of Bayesian adjusted
9-40
-------
effect estimates in the C-R functions for each urban study area should make each individual C-R
function less sensitive to the particular spatial-temporal pattern present during the place and time
used for the epidemiological analysis.
There is also additional uncertainty in the specific quantitative estimates of mortality risk
from long-term Os exposures resulting from applying the non-threshold, two-pollutant model
from Jerret et al. (2009). The results of a sensitivity analysis evaluating alternative threshold
models showed that these mortality risk estimates are highly sensitive to the specific threshold
concentration. Given that the authors of the Jerrett et al (2009) study concluded that there was
substantial uncertainty about both the existence and level of a potential threshold, we
acknowledge that there is substantially more uncertainty in the estimates of mortality associated
with long-term exposures compared to those associated with short-term exposures.
9.6 OVERALL INTEGREATED CHARACTERIZATION OF RISK IN THE
CONTEXT OF KEY POLICY RELEVANT QUESTIONS
Our analyses set out to inform two questions: (1) what are the magnitudes of exposures of
concern and risks for Os-related health effects that are estimated to occur with Os concentrations
that just meet the existing Os standard?; and (2) to what extent do alternative standards reduce
estimated exposures and risks of concern attributable to Os, focusing on at-risk populations and
lifestages? In evaluating risk, we did not limit the assessment to just the absolute risk that is
attributable to U.S. or North American emissions, as this is not relevant to answer the two
questions. Instead, we estimated total risk from all Os concentrations and the distribution of risk
over the range of Os concentrations. Our estimates of changes in risk from meeting alternative Os
standard levels relative to meeting the existing standard reflect only the impact of reductions in
U.S. precursor emissions on Os distributions, recognizing that these emissions are most likely to
be affected by implementation of the standards.
To inform these questions, we conducted air quality, exposure, and risk analyses for
selected urban study areas. We evaluated changes in the distribution of Os concentrations along
the full range of Os concentrations down to zero. We have utilized a new method (compared to
the Os NAAQS review completed in 2008) for estimating Os concentrations consistent with just
meeting existing and alternative standards, based on modeling the response of Os concentrations
to reductions in U.S. anthropogenic NOx and VOC emissions, using the HDDM capabilities in
CMAQ. This modeling incorporates all known emissions, including emissions from non-
anthropogenic sources and anthropogenic emissions from sources in and outside of the U.S. As a
result, background Os concentrations are directly modeled and, therefore, do not need to be
separately specified. Application of this approach also addresses the recommendation by the
National Research Council (NRC, 2008) to explore how emissions reductions might affect
9-41
-------
temporal and spatial variations in Os concentrations, and to include information on how NOx
versus VOC control strategies might affect exposure to Os and potential risks.
We estimated exposures and risks using several different metrics. Consistent with the
available evidence, we estimated the percentages of different study populations and lifestages
with exposures exceeding several health-based exposure benchmarks. We estimated lung
function risks based on a model of individual risk of lung function decrements that incorporates a
dose-equivalent threshold and individual exposures, activity levels, and physiology. We
estimated mortality and morbidity risks based on non-threshold C-R functions derived from
epidemiology studies. These three different analyses result in differing sensitivities of results to
changes in the Os concentration distribution. Because the three metrics are affected differently in
the analyses by changes in Os at low concentration levels, it is important to understand these
changes in Os at low concentrations in interpreting differences in the results across metrics.
We also evaluated the degree to which exposures of concern and lung function risk were
reduced in the portions of urban study areas (urban core areas) that were more likely to
experience an increase in low concentrations of Os, and in some cases an overall net increase in
epidemiology based mortality and morbidity risk (results for this assessment are presented in
Appendix 9A). We compared these estimates of changes in exposures and lung function risk to
estimates of changes in exposures and lung function risk in the areas outside of the urban core
areas to judge whether for exposures of concern and lung function risk we see the same pattern
of risk reduction between those areas.
Both exposures of concern and lung-function risk estimates in the core urban study areas
showed similar patterns compared with the areas outside the urban cores when just meeting the
existing and potential alternative standards. Thus, we observe that in urban core areas which in
some cases showed overall increases in epidemiology-based mortality and morbidity risk when
looking across these same air quality scenarios (see section 9.5.3), we generally see reductions in
exposures of concern and lung function risk. These findings illustrate that populations within
core urban study areas are likely to experience risk reductions for health endpoints reflected in
the exposure and lung function analyses.
The mortality and morbidity risk assessment is the analysis that is most sensitive to the
increases in Os in the lower part of the distribution of initial Os concentrations at some monitors
and on some days after meeting the existing and alternative standards in some urban study areas.
As demonstrated in the heat maps (Figures 7-2 and 7-3), the increases in Os (and resulting
estimated increases in risk) occur largely on days with initial Os concentrations in the range of 10
to 40 ppb. In addition, mean Os concentrations for the urban study areas change little between
air quality scenarios for meeting the existing and alternative standards, because mean
concentrations reflect both the increases in Os at lower concentrations and the decreases in Os
9-42
-------
occurring on days with high Os concentrations. This leads to small net changes in mortality and
morbidity risk estimates for many of the urban study areas. For New York, we find there is a
larger decrease relative to other urban study areas (nearly five times as large as the next largest
result for Los Angeles), in mortality and respiratory hospital admissions when just meeting the
65 ppb alternative standard compared to just meeting the existing standard, reflecting the large
degree of air quality adjustment needed to meet the standard at all monitors in New York. Both
the net change in risk and the distribution of risk across the range of Os concentrations in the
urban study areas may be relevant in considering the degree of additional protection provided by
just meeting existing and alternative standards.
The dampened response of short-term mortality risk can be contrasted with lung function
risk estimates based on application of results from controlled human exposure studies. The lung
function risk estimates primarily reflect changes in the upper end of the Os distribution and
reflect counts of exceedances of lung function decrement benchmarks, rather than summing risks
across all days in the season. In addition, lung function risks are based on detailed
microenvironmental exposure modeling that uses tract-level concentrations instead of composite
monitor values, thereby resulting in less dampening of spatial variability in Os within a given
urban study area.
The exposure benchmark analysis is the least sensitive to changes in Os in the lower part
of the distribution of initial Os concentrations, because the lowest of the exposure benchmarks is
60 ppb, well above the portion of the distribution where initial Os concentrations tended to
increase. Since the modeled exposures will always be less than or equal to the ambient
concentrations, a benchmark of exposure at 60 ppb is above the range of Os concentrations
where the HDDM approach estimates increases in concentrations. Thus, this metric is most
reflective of the decreases in Os at high concentrations that are expected to result from just
meeting the existing and alternative standards.
The lung function risk analysis is less sensitive than the epi-based risk assessments to
increases at very low concentrations of Os, because the epi-based risk function is logistic and
shows little response at lower Os dose rates that tend to occur when ambient concentrations are
lower (generally less than 20 ppb for the 10 percent FEVi decrement and generally less than 40
ppb for the 15 percent FEVi decrement). However, because there are still some increases in Os
concentrations that occur in the 50 to 60 ppb range where the estimated risk is more responsive,
there may be some reduction in the magnitude of the risk decrease (this is evident when
comparing the lung function risk metric with the exposure benchmark metric in Figure 9-7).
The exposure-based lung function risk assessment is based on controlled human exposure
studies which studied responses in healthy adults. Although the lung function model based on
this population shows less responsiveness at lower ambient concentrations, the applicability of
9-43
-------
this model to the responses of more sensitive populations and lifestages, including children and
asthmatics, is uncertain. In addition, although the most complete information for generating an
E-R function is available for FEVi as a measure of lung function, there are other, potentially
more public health relevant effects, such as lung inflammation, which have also been shown to
respond to Os. As such, the lung function risk analysis should be seen as providing useful but not
complete information on risks of health responses to Os.
Exposures above health benchmarks and risks remain after adjusting Os to just meet the
existing standard. The percentage of children with at least one 8-hr Os exposure exceeding 60
ppb is greater than 10 percent in at least one of the five analytical years for all of the 15 urban
study areas. The percent of children with a predicted decrement in lung function greater than or
equal to 10 percent is greater than 16 percent in at least one of the five analytical years for all of
the 15 urban study areas, and for a 15 percent decrement is less than 7 percent for all years and
areas. Os-attributable mortality is slightly less than one percent up to four percent of total
mortality across the 12 urban study areas, with little variation between 2007 and 2009. Os-
attributable respiratory hospital admissions are between 2 and 3 percent across the 12 urban
study areas, with little variation between 2007 and 2009. The percent attributable risk for other
morbidity endpoints is somewhat higher than for respiratory hospital admissions, but we only
estimated these endpoints for a more limited set of urban study areas due to data limitations.
The degree of reduction in exposures and risks when adjusting Os from just meeting the
existing standard to just meeting lower alternative standard levels varies considerably between
metrics. The greatest degree of reduction occurs in exposures above the 60 ppb exposure
benchmark, followed by reductions in lung function decrements greater than or equal to 10
percent, with the smallest changes in mortality and respiratory hospital admissions. Although the
magnitude of reduction differs between the different exposure and risk metrics, there are
generally the same patterns of reductions for the exposure benchmark and lung function risk
metrics, showing consistent reductions across all 15 urban study areas. Risk reductions also
occur in most of the urban study areas for mortality and respiratory hospital admissions.
However, these reductions are small, and reflect net changes in risk that include days with risk
increases as well as risk decreases. For most urban study areas, the greatest incremental
reductions in exposures above the 60 ppb benchmark occurred when just meeting 70 ppb
compared to just meeting the existing standard. Just meeting lower standards of 65 ppb and 60
ppb had incrementally smaller reductions in the percent of children exposed above 60 ppb.
Incremental lung function risk reductions are more even between alternative standards, with
similar or greater incremental reductions for the 65 ppb and 60 ppb alternatives compared with
the incremental reductions for just meeting 70 ppb. Incremental reductions in mortality and
respiratory hospital admissions risk are small between alternative standards, but more urban
9-44
-------
study areas have somewhat larger risk reductions when comparing just meeting the 60 ppb
alternative to just meeting the 65 ppb standard, than when comparing 65 ppb to 70 ppb or 70 ppb
to 75 ppb. Long-term exposure mortality risk results show larger absolute estimates of mortality
risk and more consistent reductions across urban study areas. However, percent changes in long-
term exposure mortality are similar to those for short-term exposure mortality.
In conclusion, we have estimated that exposures and risks remain after just meeting the
existing standards and that in many cases, just meeting alternative standard levels results in
reductions in those exposures and risks. Meeting alternative standards has larger impacts on
metrics that are not sensitive to changes in lower Os concentrations. When meeting the 70, 65,
and 60 ppb alternative standards, the percent of children experiencing exposures above the 60
ppb health benchmark falls to less than 20 percent, less than 10 percent, and less than 3 percent
in the worst Os year for all 15 study urban study areas, respectively. Lung function risk also
drops considerably as lower standards are met. When meeting the 70, 65, and 60 ppb alternative
standards, the percent of children with lung function decrements greater than or equal to 10
percent falls to less than 21 percent, less than 18 percent, and less than 14 percent in the worst Os
year for all 15 urban study areas, respectively. When meeting the 70 and 65 ppb alternative
standards, mortality risk attributable to short-term Os exposures falls by up to 5 percent and up to
22 percent, respectively across all 12 urban study areas, and when meeting the 60 ppb alternative
standards, falls by up to 14 percent, excluding New York, where we were unable to adjust air
quality to meet the 60 ppb alternative standard level.
While there remain uncertainties in each of the analytical areas, we have sufficient
confidence in the overall results for them to be useful in informing the policy assessment. Our
assessment suggests that the highest confidence should be placed in the results of the human
exposure and lung function risk results, largely because they are based on results of controlled
human exposure studies and a physiology-based risk model. Medium to high confidence should
be placed in the results of the assessment of epidemiology-based risks associated with short-term
Os exposures, because while the large number of studies supporting the C-R relationships used
provides increased confidence, there remain uncertainties related to unexplained heterogeneity
between locations, exposure measurement errors, and interpretation of the shape of the C-R
function at lower Os concentrations. Lower confidence should be placed in the results of the
assessment of epidemiology-based mortality risks associated with longer-term Os exposures,
primarily because that analysis is based on only one well designed study, and because of the
uncertainty in that study about the existence and location of a potential threshold in the C-R
function.
9-45
-------
9.7 REFERENCES
Frey, C. and J. Samet. 2012. CASAC Review of the EPA 's Policy Assessment for the Review of
the Ozone National Ambient Air Quality Standards (First External Review Draft -
August 2012). U.S. Environmental Protection Agency Science Advisory Board. EPA-
CASAC-13-003.
Medina-Ramon, M.; A. Zanobetti and J. Schwartz. 2006. The effect of Os and PMio on hospital
admissions for pneumonia and chronic obstructive pulmonary disease: a national
multicity study American Journal of Epidemiology. 163(6):579-588.
McDonnell, W.F.; P.W. Stewart; M.V. Smith; C.S. Kim C.S. andE.S. Schelegle. 2012.
Prediction of lung function response for populations exposed to a wide range of O3
conditions. Inhalation Toxicology. 24:619-633.
Simon, H.; K.R. Baker; F. Akhtar; S.L. Napelenok; N. Possiel; B. Wells and B. Timin. 2013. A
direct sensitivity approach to predict hourly O3 resulting from compliance with the
National Ambient Air Quality Standard. Environmental Science and Technology.
47:2304-2313.
Smith, R.L.; B. Xu and P. Switzer. 2009. Reassessing the relationship between O3 and short-
term mortality in U.S. urban communities. Inhalation Toxicology. 21:37-61.
U.S. EPA. 2007. Ozone Health Risk Assessment for Selected Urban Case Study Areas. Research
Triangle Park, NC: Office of Air Quality Planning and Standards. (EPA document
number EPA 452/R-07-009).
.
U.S. EPA. 2013. Integrated Science Assessment for Os and Related Photochemical Oxidants:
Final. Research Triangle Park, NC: U.S. Environmental Protection Agency. (EPA
document number EPA/600/R-10/076F).
World Health Organization. 2008. Part 1: Guidance Document on Characterizing and
Communicating Uncertainty in Exposure Assessment, Harmonization Project Document
No. 6. Published under joint sponsorship of the World Health Organization, the
International Labour Organization and the United Nations Environment Programme.
WHO Press, World Health Organization, 20 Avenue Appia, 1211 Geneva 27,
Switzerland.
Zanobetti, A. and J. Schwartz. 2008. Mortality displacement in the association of ozone with
mortality: an analysis of 48 cities in the United States. American Journal of Respiratory
and Critical Care Medicine. 177:184-189.
9-46
-------
This page left intentionally blank
9-47
-------
United States Office of Air Quality Planning and Standards Publication No. EPA-452/R-14-004a
Environmental Protection Health and Environmental Impacts Division August 2014
Agency Research Triangle Park, NC
-------
|