&EPA
United State
EirviroiiwiU Protection
Agnncy
Health Risk and Exposure Assessment
for Ozone
Second External Review Draft
-------
DISCLAIMER
This draft 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. This draft document is being circulated to
facilitate discussion with the Clean Air Scientific Advisory Committee to inform the EPA's
consideration of the ozone National Ambient Air Quality Standards.
This information is distributed for the purposes of pre-dissemination peer review under
applicable information quality guidelines. It has not been formally disseminated by EPA. It
does not represent and should not be construed to represent any Agency determination or policy.
Questions related to this preliminary draft 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).
11
-------
EPA-452/P-14-004a
February 2014
Health Risk and Exposure Assessment for Ozone
Second External Review Draft
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
in
-------
This page left intentionally blank.
IV
-------
TABLE OF CONTENTS
TABLE OF CONTENTS v
TABLE OF FIGURES xi
TABLE OF TABLES xxv
1 INTRODUCTION 1-1
1.1 HISTORY 1-3
1.2 CURRENT RISK AND EXPOSURE ASSESSMENT: GOALS AND PLANNED
APPROACH 1-5
1.3 ORGANIZATION OF DOCUMENT 1-6
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 Os Chemistry and Response to Changes in Precursor Emissions 2-5
2.2.2 Sources of Os and O3 Precursors 2-6
2.2.3 Simulation of Meeting Existing and Alternative Standards 2-7
2.2.4 Consideration of Health Evidence 2-8
2.2.5 Exposures of Concern 2-9
2.2.6 Health Endpoints 2-9
2.2.7 Exposure and Concentration-response Functions for Health Endpoints 2-13
2.2.8 At-risk Populations 2-14
2.3 URBAN-SCALE MODELING OF INDIVIDUAL EXPOSURE 2-15
2.3.1 Microenvironmental O3 Concentrations 2-16
2.3.2 Human Activity Patterns 2-17
2.3.3 Modeling of Exposures Associated with Simulating Just Meeting O3
Standards 2-19
2.3.4 Considerations in Selecting Urban Case Study Areas for the Exposure
Analysis 2-19
2.4 RISK ASSESSMENT 2-19
2.4.1 Attributable Risk 2-20
2.4.2 Modeling of Risk for Total Exposure to O3 2-21
2.4.3 Distributions of Risk Across O3 concentrations 2-22
2.5 MODELING OF RISKS ASSOCIATED WITH SIMULATING JUST MEETING O3
STANDARDS 2-22
2.6 CONSIDERATIONS IN SELECTING URBAN CASE STUDY AREAS FOR THE
RISK ANALYSIS 2-23
2.7 RISK CHARACTERIZATION 2-23
2.8 REFERENCES 2-25
-------
3 SCOPE 3-1
3.1 OVERVIEW OF EXPOSURE AND RISK ASSESSMENTS FROM LAST REVIEW 3-2
3.1.1 Overview of Exposure Assessment from Last Review 3-2
3.1.2 Overview of Risk Assessment from Last Review 3-3
3.2 PLAN FOR THE CURRENT EXPOSURE AND RISK ASSESSMENTS 3-5
3.3 CHARACTERIZATION OF UNCERTAINTY AND VARIABILITY IN THE
CONTEXT OF THE O3 EXPOSURE AND RISK ASSESSMENT 3-7
3.4 AIR QUALITY CHARACTERIZATION 3-10
3.5 EXPOSURE ASSESSMENT 3-13
3.6 URBAN-SCALE LUNG FUNCTION RISK ANALYSES BASED ON APPLICATION
OF RESULTS FROM CONTROLLED HUMAN EXPOSURE STUDIES 3-15
3.7 URBAN CASE STUDY AREA EPIDEMIOLOGY-BASED RISK ASSESSMENT.. 3-17
3.8 NATIONAL-SCALE MORTALITY RISK ASSESSMENT 3-22
3.9 PRESENTATION OF EXPOSURE AND RISK ESTIMATES TO INFORM THE O3
NAAQS POLICY ASSESSMENT 3-24
3.10 REFERENCES 3-26
4 AIR QUALITY CONSIDERATIONS 4-1
4.1 INTRODUCTION 4-1
4.2 OVERVIEW OF O3 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 Case Study Areas 4-5
4.3.2 Recent Air Quality 4-9
4.3.3 Air Quality Adjustments for "Just Meeting" Existing and Potential Alternative O3
Standards 4-14
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 ..4-38
4.6 REFERENCES 4-52
5 CHARACTERIZATION OF HUMAN EXPOSURE TO O3 5-1
5.0 OVERVIEW 5-1
5.1 SYNOPSIS OF O3 EXPOSURE AND EXPOSURE MODELING 5-2
5.1.1 Human Exposure 5-2
5.1.2 Estimating O3 Exposure 5-3
5.1.3 Modeling O3 Exposure Using APEX 5-4
5.2 SCOPE OF THE EXPO SURE ASSESSMENT 5-6
5.2.1 Urban Areas Selected 5-6
5.2.2 Time Periods Simulated 5-7
5.2.3 Ambient Concentrations Used 5-9
vi
-------
5.2.4 Meteorological Data Used 5-10
5.2.5 Populations Simulated 5-11
5.2.6 Key Physiological Processes And Personal Attributes Modeled 5-15
5.2.7Microenvironments Modeled 5-16
5.2.8 Model Output 5-17
5.3 EXPOSURE ASSESSMENT RESULTS 5-23
5.3.1 Overview 5-23
5.3.2 Exposure Modeling Results for Base Air Quality 5-24
5.3.3 Exposure Modeling Results for Simulations of Just Meeting Existing and Alternative
O3 Standards 5-25
5.4 TARGETED EVALUATION OF EXPOSURE MODEL INPUT AND OUTPUT DATA
5-35
5.4.1 Analysis of Time-Locaton-Activity Data 5-35
5.4.1.1 General Evaluation of CHAD Study Data: Historical and Recently Acquired Data
5-36
5.4.1.2 Exposure-Relevant Personal Attributes Included in CHAD and APEX Simulated
Individuals 5-36
5.4.1.3 Evaluation of Afternoon Time Spent Outdoors for CHAD and Survey Participants
5-37
5.4.1.4 Evaluation of Afternoon Time Spent Outdoors for ATUS Survey
Participants 5-38
5.4.1.5 Evaluation of Outdoor Time and Exertion Level for Asthmatics and Non-
Asthmatics in CHAD 5-39
5.4.2 Characterization of Factors Influencing High Exposures 5-40
5.4.3 Exposure Results for additional at-risk populations and Lifestages, Exposure
scenarios, and Air Quality Input Data Used 5-41
5.4.3.1 Exposures Estimated for All School-age Children During Summer Months,
Neither Attending School or Performing Paid Work 5-41
5.4.3.2 Exposures Estimated for Outdoor Workers During Summer Months 5-43
5.4.3.3 Exposures Estimated for All School-age Children When Accounting for Averting
Behavior 5-45
5.4.3.4 Comparison of APEX Estimated Exposures Using Three Different Base Case Air
Quality Data Sets: AQS, VNA, andEVNA 5-46
5.4.3.5 Comparison of APEX Estimated Exposures Using Two Different Adjusted Air
Quality Data Sets: Quadratic Rollback and HDDM 5-48
5.4.4 Limited Performance Evaluations 5-49
5.4.4.1 Personal Exposure Comparisons 5-49
5.4.4.2 Ventilation Rate Comparisons 5-51
5.4.4.3 Evaluation of Longitudinal Profile Methodology 5-55
5.5 VARIABILITY AND UNCERTAINTY 5-56
vn
-------
5.5.1 Treatment of Variability 5-56
5.5.2 Characterization of Uncertainty 5-57
5.6 KEY OBSERVATIONS 5-65
5.7 REFERENCES 5-71
6 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 6-3
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
6.2.4 The McDonnell-Stewart-Smith (MSS) Model 6-8
6.2.5 The Exposure-Response Function Approach Used in Prior Reviews 6-16
6.3 O3 RISK ESTIMATES 6-21
6.3.1 Lung Function Risk Estimates Based on the McDonnell-Stewart-Smith Model 6-22
6.3.2 Lung Function Risk Estimates Based on the Exposure-Response Functions
Approach Used in Prior Reviews 6-28
6.3.3 Comparison of the MSS Model with the Exposure-Response Function
Approach 6-29
6.4 EVALUATION OF THE MSS MODEL 6-35
6.4.1 Summary of Published Evaluations 6-35
6.4.2 Children 6-35
6.4.3 Threshold vs. Non-Threshold Models 6-36
6.5 CHARACTERIZATION OF UNCERTAINTY 6-37
6.5.1 Statistical Model Form 6-38
6.5.2 Convergence of APEX Results 6-41
6.5.3 Application of Model for All Lifestages 6-42
6.5.4 Application of Model for Asthmatic Children 6-43
6.5.5 Interaction Between Os and Other Pollutants 6-43
6.5.6 Qualitative Assessment of Uncertainty 6-43
6.6 DISCUSSION 6-46
6.7 REFERENCES 6-50
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 Os-Related Health Effects Incidence 7-11
7.2 AIR QUALITY CONSIDERATIONS 7-13
viii
-------
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 7-30
7.3.4 Population (demographic) Data 7-31
7.4 ADORES SING VARIABILITY AND UNCERTAINTY 7-31
7.4.1 Treatment of Key Sources of Variability 7-34
7.4.2 Qualitative Assessment of Uncertainty 7-38
7.4.3 Description of Core and Sensitivity Analyses 7-45
7.5 URBAN STUDY AREA RESULTS 7-47
7.5.1 Assessment of Health Risk After Just Meeting the Existing 75 ppb standard 7-65
7.5.2 Assessment of Health Risk Associated with Simulating Meeting Potential
Alternative Standards of 70, 65, and 60 ppb 7-67
7.5.3 Sensitivity Analyses Designed to Enhance Understanding of the Core Risk
Estimates 7-72
7.6 KEY OBSERVATIONS REGARDING OVERALL CONFIDENCE IN THE RISK
ASSESSMENT AND RISK ESTIMATES 7-83
7.7 REFERENCES 7-86
8 NATIONAL SCALE MORTALITY RISK BURDEN BASED ON APPLICATION OF
RESULTS FROM EPIDEMIOLOGICAL STUDIES 8-1
8.1 NATIONAL-SCALE ASSESSMENT OF MORTALITY RELATED TO O3
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 8-19
8.2 EVALUATING THE REPRESENTATIVENESS OF THE URBAN STUDY AREAS IN
THE 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 Os-Related
Mortality Risk 8-45
8.2.3 Analysis Based on Consideration of National Responsiveness of Os Concentrations
to Emissions Changes 8-49
8.2.5 Discussion 8-77
8.3 REFERENCES 8-78
9 SYNTHESIS 9-1
9.1 INTRODUCTION 9-2
9.2 SUMMARY OF KEY RESULTS 9-2
9.2.1 Air Quality Considerations (Chapter 4) 9-2
9.2.2 Human Exposure Modeling (Chapter 5) 9-8
ix
-------
9.2.3 Health Risks Based on Controlled Human Exposure Studies (Chapter 6) 9-13
9.2.4 Health Risks Based on Epidemiological Studies (Chapters 7 and 8) 9-18
9.3 COMPARISON OF RESULTS ACROSS EXPOSURE, LUNG FUNCTION RISK,
AND EPIDMEIOLOGY-BASED MOTALITY AND MORBIDITY RISK
ANALYSES 9-27
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 9-31
9.4 OVERALL ASSESSMENT OF REPRESENTATIVENESS OF EXPOSURE AND
RISK RESULTS 9-36
9.4.1 Representativeness of Selected Urban Case Study Areas in Reflecting Areas
Across the Nation with Elevated Risk 9-36
9.4.2 Representativeness of Selected Urban Case Study Areas in Reflecting
Responsiveness of Risk to Just Meeting Existing and Alternative Os
Standards 9-37
9.5 OVERALL ASSESSMENT OF CONFIDENCE IN EXPOSURE AND RISK
RESULTS 9-38
9.5.1 Uncertainties in Modeling Oj Responses to Meeting Standards 9-39
9.5.2 Uncertainties in Modeling Exposure and Lung-function Risk 9-40
9.5.3 Uncertainties in Modeling Epidemiological-based Risk 9-40
9.6 OVERALL INTEGRATED CHARACTERIZATION OF RISK IN THE CONTEXT OF
KEY POLICY RELEVANT QUESTIONS 9-42
9.7 REFERENCES 9-47
-------
TABLE OF FIGURES
Figure 2-1. Overview of Exposure and Risk Assessment Design 2-2
Figure 2-2. Causal Determinations for O3 Health Effects 2-11
Figure 3-1. Conceptual Diagram for Air Quality Characterization in the Health REA 3-10
Figure 3-2. Conceptual Diagram for Population Exposure Assessment 3-13
Figure 3-3. Conceptual Diagram of O3 Lung Function Health Risk Assessment Based on
Controlled Human Exposure Studies 3-15
Figure 3-4. Conceptual Diagram of Urban Case Study Area Health Risk Assessment Based on
Results of Epidemiology Studies 3-18
Figure 3-5. Conceptual Diagram of National O3 Mortality Risk Assessment Based on Results of
Epidemiology Studies 3-23
Figure 4-1. Map of Monitored 8-hour O3 Design Values for the 2006-2008 Period 4-3
Figure 4-2. Map of Monitored 8-hour O3 Design Values for the 2008-2010 Period 4-4
Figure 4-3. Flowchart of Air Quality Data Processing for Different Parts of the Urban-scale Risk
and Exposure Assessments 4-5
Figure 4-4. Trends in Annual 4th Highest 8-hour Daily Maximum O3 Concentrations in ppb for
the 15 Urban Case Study Areas for 2006-2010. Urban areas are grouped into 3
regions: Eastern (top), Central (middle), and Western (bottom) 4-7
Figure 4-5a. Maps of the 5 Eastern U.S. Urban Case Study Areas Including O3 Monitor
Locations 4-11
Figure 4-5b. Maps of the 5 Central U.S. Urban Case Study Areas Including O3 Monitor
Locations 4-12
Figure 4-5 c. Maps of the 5 Western U.S. Urban Case Study Areas Including O3 Monitor
Locations 4-13
Figure 4-6. Flowchart of HDDM adjustment methodology to inform risk and exposure
assessment 4-17
Figure 4-7. Distributions of composite monitor 8-hour daily maximum O3 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 4-21
XI
-------
Figure 4-8. Distributions of composite monitor 8-hour daily maximum O^ 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 4-22
Figure 4-9. Distributions of composite monitor 8-hour daily maximum values for the 12 urban
case study areas in the epidemiology-based risk assessment. Plots depict values
based on ambient measurements (base), and values obtained with the HDDM
adjustment methodology showing attainment of 75, 70, 65 and 60 ppb standards.
Values shown are based on CBSAs for April-October of 2007. Note that the
FIDDM adjustment technique was not able to adjust air quality to show attainment
of a 60 ppb standard in New York, so no boxplot is shown for that case 4-25
Figure 4-10. Distributions of composite monitor 8-hour daily maximum values for the 12 urban
case study areas in the epidemiology-based risk assessment. Plots depict values
based on ambient measurements (base), and values obtained with the HDDM
adjustment methodology showing attainment of 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 FIDDM adjustment
technique was not able to adjust air quality to show attainment of a 60 ppb
standard in New York, so no boxplots are shown for those cases 4-26
Figure 4-11. Maps showing the 4th highest (top) and May-September average (bottom) daily
maximum 8-hour Oi concentrations in Atlanta based on 2006-2008 ambient
measurements (left), HDDM adjustment to meet the existing standard (center),
and FIDDM 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
Figure 4-12. Maps showing the 4 highest (top) and May-September average (bottom) daily
maximum 8-hour 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 4-29
xn
-------
Figure 4-13. Maps of 4th highest (top) and May-September average (bottom) daily maximum 8-
hour 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 4-30
Figure 4-14. May-September average 8-hour daily maximum O3 concentrations in ppb, based on
a Downscaler fusion of 2006-2008 average monitored values with a 12km 2007
CMAQ model simulation 4-32
Figure 4-15. June-August average 8-hour daily 10am-6pm mean O3 concentrations in ppb, based
on a Downscaler fusion of 2006-2008 average monitored values with a 12km
2007 CMAQ model simulation 4-33
Figure 4-16. April-September average 1-hour daily maximum O3 concentrations in ppb, based on
a Downscaler fusion of 2006-2008 average monitored values with a 12km 2007
CMAQ model simulation 4-34
Figure 4-17. Frequency and Cumulative Distributions of the Three Fused Seasonal Average O3
Surfaces Based on all CMAQ 12 km Grid Cells 4-35
Figure 4-18. 2006-2008 O3 Design Values Versus 2006-2008 Fused Seasonal Average O3 Levels
for the CMAQ 12km Grid Cells Containing O3 Monitors 4-37
Figure 5-1. Conceptual Framework Used for Estimating Study Area Population O3 Exposure
Concentrations 5-8
Figure 5-2. Percent of asthmatic school-age children in all study areas with at least one O3
exposure at or above 60 ppb-8hr while at moderate or greater exertion using base
air quality (2006-2010), stratified by year (top left panel) or by study area (bottom
left panel) 5-20
Figure 5-3. Percent of asthmatic school-age children in Atlanta with at least one O3 exposure at
or above 60 ppb-8hr (left top panel), 70 ppb-8hr (middle top panel), and 80 ppb-
8hr (right top panel while at moderate or greater exertion, years 2006-2010 air
quality adjusted to just meet the existing and alternative O3 standard levels. The
multi-panel display (bottom) illustrates the same exposure results expanded to
Xlll
-------
reflect individual data points by year, standard averaging period, and benchmark
level 5-22
Figure 5-4. Percent of asthmatic school-age children in Atlanta with multiple 63 exposures at or
above 60 ppb-8hr while at moderate or greater exertion, years 2006-2010 air
quality adjusted to just meet the existing and alternative 63 standard levels.... 5-23
Figure 5-5. Percent of all school-age children with at least one daily maximum 8-hr average 63
exposure at or above 60, 70, and 80 ppb while at moderate or greater exertion,
years 2006-2010, air quality adjusted to just meet the existing and potential
alternative standards 5-30
Figure 5-6. Percent of asthmatic school-age children with at least one daily maximum 8-hr
average 63 exposure at or above 60, 70, and 80 ppb while at moderate or greater
exertion, years 2006-2010, air quality adjusted to just meet the existing and
potential alternative standards 5-31
Figure 5-7. Percent of all asthmatic adults with at least one daily maximum 8-hr average 63
exposure at or above 60, 70, and 80 ppb-8hr while at moderate or greater exertion,
years 2006-2010, air quality adjusted to just meet the existing and potential
alternative standards 5-32
Figure 5-8. Percent of all older adults with at least one daily maximum 8-hr average 03 exposure
at or above 60, 70, and 80 ppb-8hr while at moderate or greater exertion, years
2006-2010, air quality adjusted to just meet the existing and potential alternative
standards 5-33
Figure 5-9. Percent of all school-age children with multiple daily maximum 8-hr average Oi
exposures at or above 60 ppb while at moderate or greater exertion, years 2006-
2010, air quality adjusted to just meet the existing and potential alternative
standards 5-34
Figure 5-10. Comparison of the percent of all school-age children having daily maximum 8-hr
average 63 concentration at or above 60 ppb during June, July, and August in
Detroit 2007: using any available CHAD diary ("All CHAD Diaries") or using
CHAD diaries having no time spent in school or performing paid work ("No
School/Work Diaries") 5-42
xiv
-------
Figure 5-11. Percent of workers between age 18-35 experiencing exposures at or above selected
benchmark levels while at moderate or greater exertion using an outdoor worker
approach (left panel) and a general population-based approach (right panel)
considering air quality adjusted to just meet the existing standard in Atlanta, GA,
Jun-Aug, 2006 5-45
Figure 5-12. Percent of all school-age children (left panel) and asthmatic school-age children
(right panel) having daily maximum 8-hr average Os concentration at or above
benchmark levels during a 2-day simulation in Detroit, base air quality, August 1-
2, 2007. Red bars indicate exposure results when considering effect of averting....
5-46
Figure 5-13. Comparison of APEX exposure results generated for three study areas (Atlanta,
Detroit, and Philadelphia) using three different 2005 air quality input data sets:
AQS, VNA, and eVNA 5-47
Figure 5-14. 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 (right panel) 5-48
Figure 5-15. Distribution of daily average O^ exposures (top panels) and daily afternoon outdoor
time (bottom panels) and for DEARS study participants (left panels) and APEX
simulated individuals (right panels) in Wayne County, MI, July-August 2006 5-50
Figure 5-16. Means (and range) of 6-day average personal 63 exposures, measured and modeled
(APEX), Upland Ca. Obtained from Figure 8-22 of US EPA (2007b) 5-51
Figure 5-17. Comparison of body mass normalized mean daily ventilation rates estimated by
APEX (closed symbols) and by Brochu et al., 2006 (open symbols) 5-52
Figure 5-18. 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 5-53
Figure 5-19. Comparison of body mass normalized daily mean ventilation rates in school-age
children (5-18) estimated using APEX and literature reported values 5-55
Figure 5-20. Incremental increases in percent of all school-age children exposed to 63 at or
above 60 ppb-8hr for each study area, year 2006-2010 air quality. Average
xv
-------
percent (left panels), maximum percent (right panels), at least one exposure (top
panels), at least two exposures (bottom panels) per year 5-69
Figure 6-1. Two-Compartment Model 6-1
Figure 6-2. Distribution of Responses (Lung Function Decrements in FEV1) Predicted by the
MSS Model for 20-Year Old Individuals. Exposure to 100 ppb O3 at Moderate
Exercise (40 L/min, BSA=2 m2) Under the Conditions of a Typical 6.6-hour
Clinical Study 6-13
Figure 6-3. Median Response (Lung Function Decrements in FEV1) Predicted by the MSS
Model for 20-Year Old Individuals. Exposure to 100 ppb O3 at Moderate Exercise
(40 L/min, BSA=2 m2) Under the Conditions of a Typical 6.6-hour Clinical
Study 6-14
Figure 6-4. Median Response (FEV1 Decrements) Predicted by the MSS Threshold and Non-
Threshold Models for 20-Year Old Individuals, Constant 100 ppb O3 Exposure, 2
Hours Heavy Exercise (30 L/min-m2 BSA) 6-15
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, 2 Hours
Heavy Exercise (30 L/min-m2 BSA) 6-15
Figure 6-6. Probabilistic Exposure-Response Relationships for FEV1 Decrements > 10% for 8-
Hour Exposures At Moderate Exertion, Ages 18-35. Values associated with data
points are the number of subject-exposures at each exposure concentration .... 6-20
Figure 6-7. Risk results for all school-aged children with > 1 occurrences of FEV1 decrements >
10, 15, 20% for all cities, year, and scenarios (y-axis is percent of children
affected) 6-23
Figure 6-8. Risk results for all school-aged children with > 1 occurrences of FEV1 decrements >
10% under the 0.07 ppm alternative standard showing variability across cities
(horizontally) and years (vertically) 6-24
Figure 6-9. Distribution of Daily FEV1 Decrements > 10% Across Ranges of 8-hour Average
Ambient O3 Concentrations (Los Angeles, 2006 recent air quality) 6-27
Figure 6-10. Comparison of E-R and MSS Model (restricted to 8-hour average EVR > 13)
Response Functions (Atlanta 2006 base case, ages 18-35) 6-33
xvi
-------
Figure 6-11. Distribution of Daily Maximum 8-hour Average EVR For Values of EVR > 13
(L/min-m2) (midpoints on vertical axis) (Atlanta 2006 base case, ages 18-35)
6-34
Figure 6-12. Sensitivity (Percent Change) of Population With One or More FEV1 Decrements >
10% to a 5% Increase in Individual MSS Model Parameter Estimates 6-40
Figure 6-13. Lung Function Risk Results, Incremental Increases In Risk For Increasing Standard
Levels: Percent of All School-aged Children With FEV1 Decrement > 10%,
Highest Value For Each Study area Over Years 6-48
Figure 6-14. Lung Function Risk Results, Incremental Increases In Risk For Increasing Standard
Levels: Percent of All School-aged Children With FEV1 Decrement > 10%, Mean
Value For Each Study Area Over Years 6-49
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 (Just meeting existing standard
and risk reductions from just meeting alternative standards) (2007) (Smith et al.,
2009 C-R functions) 7-55
Figure 7-3. Heat Maps for Short Term Os-attributable Mortality (Just meeting existing standard
and risk reductions from just meeting alternative standards) (2009) (Smith et al.,
2009 C-R functions) (see Key at bottom of figure) 7-56
Figure 7-4. Plots of Short-Term Os-attributable All-Cause Mortality for Meeting Existing
standard and Alternative Standards (Smith et al., 2009) (Simulation year 2007 and
2009) 7-57
Figure 7-5. Plots of Short-Term Os-attributable Respiratory HA for Meeting Existing standard
and Alternative Standards (Medina-Ramon, et al., 2006) (Simulation year 2007
and 2009) 7-61
Figure 7-6. Plots of Long-Term Os-attributable Respiratory Mortality for Meeting Existing
standard and Alternative Standards (Jerrett et al., 2009) (Simulation year 2007
and 2009) 7-64
Figure 7-7. Sensitivity Analysis: Short-Term Os-attributable Mortality (air quality-related factors
including study area size and method used to simulate attainment of existing and
alternative standard levels) (2009) SAl-smaller (Smith-based) study area, SA2-
alternative method for simulating standards 7-79
xvn
-------
Figure 7-8. Sensitivity Analysis: Short-Term O3-attributable Mortality (C-R function
specification) (2009) SAl-regional Bayes-based adjustment; SA2-copollutant
model (PMio); SA3-Zanobetti and Schwartz-based effect estimates 7-80
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 average 8-hr daily maximum O3 levels
by 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 average 8-hr daily mean (10am-6pm) O3 levels
by 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 average 1-hr daily max O3
levels by 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 O3 levels by 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 O3 levels by 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 O3 levels by county using Jerrett et al. (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 O3 for the U. S 8-13
Figure 8-9. Cumulative percentage of total O3 deaths by baseline O3 concentration. O3
concentrations are reported as May-September average 8-hr daily maximum for
results based on Smith et al. (2009) effect estimates, June-August average 8-hr
mean (10am to 6pm) for results based on Zanobetti and Schwartz (2008) effect
estimates, and April-September average 1-hr daily maximum for results based on
Jerrett et al. (2009) effect estimates 8-15
xvin
-------
Figure 8-10.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:
Sametetal. 2000) 8-18
Figure 8-11. (Vattributable 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 in the main results 8-19
Figure 8-12. Comparison of county-level populations of urban case study area counties to the
frequency distribution of population in 3,143 U.S. counties 8-32
Figure 8-13. Comparison of county-level seasonal mean 8-hr daily maximum O^ concentrations
in urban case study area counties to the frequency distribution of seasonal mean
8-hr daily maximum Os concentrations in 671 U.S. counties with Os monitors
8-33
Figure 8-14. Comparison of 2007 county-level 4th high 8-hr daily maximum Oj concentrations in
urban case study area counties to the frequency distribution of 2007 4* high 8-hr
daily maximum Os concentrations in 725 U.S. counties with Os monitors 8-34
Figure 8-15. Comparison of county-level all-cause mortality in urban case study area counties to
the frequency distribution of all-cause mortality in 3,137 U.S. counties 8-35
Figure 8-16. Comparison of county-level non-accidental mortality in urban case study area
counties to the frequency distribution of non-accidental mortality in 3,135 U.S.
counties 8-36
Figure 8-17. Comparison of city-level all-cause mortality risk coefficients from Zanobetti and
Schwartz (2008) in urban case study areas to the frequency distribution of all-
cause mortality risk coefficients from Zanobetti and Schwartz (2008) in 48 U.S.
cities 8-37
Figure 8-18. Comparison of city-level national prior Bayes-shrunken non-accidental mortality
risk coefficients from Smith et al. (2009) in urban case 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 8-38
xix
-------
Figure 8-19. Comparison of county-level percent of population 0 to 14 years old in urban case
study area counties to the frequency distribution of percent of population 0 to 14
years oldin 3,141 U.S. counties 8-39
Figure 8-20. Comparison of county-level percent of population 0 to 14 years old in urban case
study area counties to the frequency distribution of percent of population 0 to 14
years oldin 3,141 U.S. counties 8-40
Figure 8-21. Comparison of county-level income per capita in urban case study areas to the
frequency distribution of income per capita in 3,141 U.S. counties 8-41
Figure 8-22. Comparison of county-level July temperature in urban case study area counties to
the frequency distribution of July temperature in all U.S. counties 8-42
Figure 8-23. Comparison of city-level asthma prevalence in urban case study areas to the
frequency distribution of asthma prevalence in 184 U.S. cities 8-43
Figure 8-24. Comparison of city-level air conditioning prevalence in urban case study areas to
the frequency distribution of air conditioning prevalence in 76 U.S. cities 8-44
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 8-47
Figure 8-26. Cumulative distribution of county-level percentage of June-August all-cause
mortality for all ages attributable to 2006-2008 average 63 for the U.S. and the
locations of the selected urban study areas along the distribution, using Zanobetti
and Schwartz (2008) effect estimates 8-48
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) 8-49
Figure 8-28. Change in 50th percentile summer season (April-October) daily 8-hr maximum Os
concentrations between 2001-2003 and 2008-2010 8-52
Figure 8-29. Change in 95th percentile summer season (April-October) daily 8-hr maximum Oj
concentrations between 2001-2003 and 2008-2010 8-53
xx
-------
Figure 8-30. Population density at each O?, monitor 8-54
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 8-56
Figure 8-32. Distributions of 63 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....
8-57
Figure 8-33. Distributions of 63 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 8-58
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 8-hr daily maximum
Os values, center panel shows trends in annual mean 8-hr daily maximum Os
values, and right panel shows trends in annual median 8-hr daily maximum Os
values 8-59
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
xxi
-------
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 4* highest 8-hr daily maximum
Os values, center panel shows trends in annual mean 8-hr daily maximum Os
values, and right panel shows trends in annual median 8-hr daily maximum 63
values 8-60
Figure 8-36. Ratio of April-October seasonal average 63 concentrations in the brute force 50%
NOx emissions reduction CMAQ simulations to April-October seasonal average
Os concentrations in the 2007 base CMAQ simulation 8-64
Figure 8-37. Ratio of April-October seasonal average 03 concentrations in the brute force 90%
NOx emissions reduction CMAQ simulations to April-October seasonal average
Os concentrations in the 2007 base CMAQ simulation 8-65
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 8-66
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 8-67
Figure 8-40. Histograms of U.S. population living in locations with increasing and decreasing
mean Os. 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 8-69
Figure 8-41. Histograms of U.S. population living in locations with increasing and decreasing
mean Os. 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
xxn
-------
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 63, middle plots show
changes in seasonal mean June-August 63, and bottom plots show changes in
seasonal mean April-October 63 8-70
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 8-71
Figure 8-43. Population (as % of total case-study area population) living in locations of
increasing April-October seasonal mean Oi in the 90% NOx reduction CMAQ
simulation 8-72
Figure 8-44. Histograms of U.S. population living in locations with increasing and decreasing
mean 03. 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 case study areas. Bottom plots show histograms for
low-mid population density areas while top plots show histograms for high
population density areas 8-74
Figure 8-45. Histograms of U.S. population living in locations with increasing and decreasing
mean 03. Values on the x-axis represent the change in seasonal mean (April-
October) O3 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 urban case study areas. Bottom plots show histograms for
xxin
-------
low-mid population density areas while top plots show histograms for high
population density areas 8-75
Figure 9-1. Distributions of composite monitor 8-hour daily maximum Os concentrations from
ambient measurements (black), quadratic rollback (blue), and the HDDM
adjustment methodology (red) for meeting the existing standard 9-7
Figure 9-2. Effects of just meeting existing (columns 1 and 2) and alternative (columns 3 through
8) standards on percent of children (ages 5-18) with at least one Os exposure at or
above 60, 70, and 80 ppb-8hr while at moderate or greater exertion, years 2006-
2010 9-11
Figure 9-3. Effects of just meeting existing (75 ppb) and alternative standards on percent of
children (ages 5-18) exceeding 60 ppb exposure benchmark, highest value across
years for each urban case study area, 2006-2010 9-13
Figure 9-4. Effects of just meeting existing (column 1) and alternative (columns 2-4) standards
on percent of children (ages 5-18) with FEV1 decrement > 10, 15, and 20%, years
2006-2010 9-16
Figure 9-5. Impact of just meeting existing (75 ppb) and alternative standards on percent of
children (ages 5-18) with FEVi decrement > 10%, highest value for each urban
case study area, 2006-2010 9-18
Figure 9-6. Impacts of just meeting existing (75 ppb) and alternative standard levels on mortality
risk per 100,000 population for 2007 and 2009 9-21
Figure 9-7. Impacts of just meeting existing and alternative standard levels on adult (ages 65 and
older) respiratory hospital admissions risk per 100,000 population for 2007 and
2009 9-22
Figure 9-8. 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 9-30
Figure 9-9. Comparison of Percent Reduction in Key Risk Metrics for Alternative Standard
Levels Relative to Just Meeting the Existing 75 ppb Standard 9-33
xxiv
-------
TABLE OF TABLES
Table 3-1. Short-term 63 Exposure Health Endpoints Evaluated in Urban Case Study Areas
3-20
Table 4-1. Monitor and Area Information for the 15 Urban Case Study Areas in the Exposure
Modeling and Clinical Study Based Risk Assessment 4-6
Table 4-2. Monitor and Area Information for the 12 Urban Case Study Areas in the
Epidemiology Based Risk Assessment 4-9
Table 4-3. Summary Statistics Based on the Three Fused Seasonal Average 63 Surfaces Based
onallCMAQ 12 km Grid Cells 4-36
Table 4-4. Correlation Coefficients Between the Three Fused Seasonal Average 63 Surfaces
Based on all CMAQ 12 km Grid Cells 4-36
Table 4-5. 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 O^ Monitors 4-38
Table 4-6. Summary of Qualitative Uncertainty Analysis of Key Air Quality Elements in the O^
NAAQS Risk Assessment 4-42
Table 5-1. General Characteristics of the Population Exposure Modeling Domain Comprising
Each Study Area 5-9
Table 5-2. Asthma Prevalence for Children and Adults Estimated by APEX in Each Simulated
Study Area 5-12
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 (b,) and residuals distributions (e\) 5-16
Table 5-5. Microenvironments Modeled, Calculation Method Used, and Variables Included
5-17
Table 5-6. Characterization of Key Uncertainties in Historical and Current APEX Exposure
Assessments 5-58
Table 5-7. Mean and Maximum Percent of all School-age Children Estimated to Experience at
Least One Daily Maximum 8-hr Average Exposure to Os at or Above Selected
Health Benchmark Levels 5-68
xxv
-------
Table 5-8. Mean and Maximum Percent of All School-age Children Estimated to Experience at
Least Two Daily Maximum 8-hr Average Exposures to Os At or Above Selected
Health Benchmark Levels 5-70
Table 6-1. Estimated Parameters in theMSS Models 6-11
Table 6-2. Age Term Parameters for Application of the 2012 MSS Threshold Model to All Ages
6-12
Table 6-3. Study-specific Ozone Exposure-response Data for Lung Function Decrements Based
on Correcting Individual Responses for the Effect on Lung Function of Exercise
in Clean Air, Ages 18-35 6-16
Table 6-4. Ranges of percents of population experiencing one or more days during the ozone
season with lung function decrement (AFEVi) more than 10%. The numbers in
this table are the minimum and maximum percents estimated over all cities and
years 6-25
Table 6-5. Ranges of percents of population experiencing one or more days during the ozone
season with lung function decrement (AFEVi) more than 15%. The numbers in
this table are the minimum and maximum percents estimated over all cities and
years 6-26
Table 6-6. Percents of the General Population and Outdoor Workers (ages 19-35) Experiencing
1 or More and 6 or More FEVi Decrements > 15% (based on Atlanta 2006 APEX
simulations) 6-28
Table 6-7. Ranges of percents of school-aged children experiencing one or more days during the
ozone season with lung function decrement (AFEVi) more than 10 and 15%. The
numbers in this table are the minimum and maximum percents estimated over all
cities and years 6-29
Table 6-8. Comparison of responses from the MSS model with responses from the population
exposure-response (E-R) method. 2006 existing standard, ages 5 to 18 6-30
Table 6-9. Comparison of MSS Model and E-R Model of Previous Reviews for Atlanta, Mar 1-
Oct30, 2006, ages 18-35 6-32
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 6-32
XXVI
-------
Table 6-11. Comparison of Responses from the MSS 2010 Model with Responses from
McDonnell etal. (1985) 6-36
Table 6-12. Percents of the population by age group with one or more days during the ozone
season with lung function (FEVi) decrements more than 10, 15, and 20% (Atlanta
2006 base case). MSS Threshold model, monitors air quality 6-37
Table 6-13. Percents of the population by age group with one or more days during the ozone
season with lung function (FEVi) decrements more than 10, 15, and 20% (Atlanta
2006 base case). MSS No-Threshold model, monitors air quality 6-37
Table 6-14. MSS threshold model estimated parameters with confidence intervals 6-39
Table 6-15. 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 ozone season with lung function (FEVi) decrements
more than 10, 15, and 20%. Minimum and maximum values and ranges over 40
APEX runs 6-42
Table 6-16. Summary of Qualitative Uncertainties of Key Modeling Elements in the Os Lung
Function Risk Assessment 6-44
Table 7-1. Information on the 12 Urban Case Study Areas in the Risk Assessment 7-14
Table 7-2. Overview of Epidemiological Studies Used in Specifying C-R Functions 7-23
Table 7-3. CBSA-based Study Areas with Multiple Effect Estimates from the Smith et al., 2009
Study 7-27
Table 7-4. Summary of Qualitative Uncertainty Analysis of Key Modeling Elements in the Oi
NAAQS Risk Assessment 7-41
Table 7-5. Specification of the Core and Sensitivity Analyses (air quality simulation) 7-46
Table 7-6. Specification of the Core and Sensitivity Analyses (alternative C-R function
specification) 7-47
Table 7-7. Short-Term O3-attributable All Cause Mortality Incidence (2007 and 2009) (Smith et
al., 2009 C-R Functions) 7-53
Table 7-8. Percent of Total All-Cause Mortality Attributable to 63 and Percent Change in 63.
Attributable Risk (2007 and 2009) (Smith et al., 2009 C-R functions) 7-54
XXVll
-------
Table 7-9. Short-Term Os-attributable Morbidity Incidence, Percent of Baseline and Reduction
in Os-attributable Risk - Respiratory-Related Hospital Admissions (2007 and
2009) 7-58
Table 7-10. Short-Term (Vattributable Morbidity Incidence, Percent of Baseline and Reduction
in Ozone-attributable Risk - Emergency Room Visits (2007 and 2009) 7-59
Table 7-11. Short-Term (Vattributable Morbidity Incidence, Percent of Baseline and Reduction
in Ozone-attributable Risk - Asthma Exacerbations (2007 and 2009) 7-60
Table 7-12. Long-Term Os-attributable Respiratory Mortality Incidence (2007 and 2009) (Jerrett
etal., 2009 C-R Functions) 7-62
Table 7-13. Long-Term Os-attributable Respiratory Mortality Percent of Baseline Incidence and
Percent Reduction in Os-attributable Risk (simulation years 2007 and 2009)
(Jerrett etal., 2009 C-R Functions) 7-63
Table 7-14. Sensitivity Analysis for Long-Term Os-attributable Respiratory Mortality -
Alternative C-R Function Specification (regional effect estimates) % of baseline
all-cause mortality and change in Os-attribuable risk (2009) (Smith et al., 2009,
Os season) 7-81
Table 7-15. Sensitivity Analysis for Long-Term Os-attributable Respiratory Mortality -
Alternative C-R Function Specification (national Os-only effect estimates) % of
baseline all-cause mortality and change in Os-attribuable risk (2009) (Smith et al.,
2009, O3 season) 7-82
Table 8-1. Estimated annual Os-related premature mortality in 2007 associated with 2006-2008
average Oj concentrations (95th percentile confidence interval) 8-7
Table 8-2. Mean, median, 2.5 percentile, and 97.5 percentile of the estimated percentage of
mortality attributable to ambient Os for all 3087 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 gridcells between 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
xxvin
-------
estimates, as compared with the national prior Bayes-shrunken city-specific and
national average effect estimates as in the main results 8-18
Table 8-5. Data Sources for 63 risk-related Attributes 8-25
Table 8-6. Summary Statistics for Selected 63 Risk-related Attributes 8-29
Table 8-7. Broad Regional Annual Trends of Concurrent Oi Concentrations and Emissions of
NOx and VOCs over the 2000-2011 Time Period 8-61
Table 9-1. Area and Monitoring Information for the 15 Case Study Areas 9-3
Table 9-2. General Patterns in Seasonal (May-Sept) Mean of Daily Maximum 8-hour Os
Concentrations after Adjusting to Meet Existing and Alternative Standards 9-5
XXIX
-------
LIST OF ACRONYMS/ABBREVIATIONS
AER
AHRQ
APEX
AQI
AQS
ATUS
BenMAP
BRFSS
BSA
CAA
CASAC
CDC
CDF
CH4
CHAD
CI
CMAQ
CO2
C-R
ED
EGU
EPA
ER
eVNA
EVR
FEM
FEV1
FRM
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
Center for Disease Control and Prevention
cumulative distribution functions
methane
Consolidated Human Activity Database
confidence interval
Community Multi-scale Air Quality
carbon dioxide
Concentration Response (function)
emergency department
electric generating unit
U.S. Environmental Protection Agency
emergency room
enhanced Voronoi Neighbor Averaging
equivalent ventilation rate
Federal Equivalent Method
one-second forced expiratory volume
Federal Reference Method
XXX
-------
FVC
HA
HDDM
HNO3
H02
HUCP
IPCC
IRP
ISA
LML
MATS
METs
MSA
MT
NAAQS
NCDC
NEI
NO
NO2
NOX
03
OAQPS
OH
PA
PDI
PI
PM
ppb
ppm
PRB
REA
forced vital capacity
hospital admissions
Higher-order Decoupled Direct Method
nitric acid
hydro-peroxy radical
Healthcare Cost and Utilization Program
Intergovernmental Panel on Climate Change
Integrated Review Plan
Integrated Science Assessment
lowest measured level
Modeled Attainment Test Software
metabolic equivalents of work
Metropolitan Statistical Area
metric ton
National Ambient Air Quality Standards
National Climatic Data Center
National Emissions Inventory
nitric oxide
nitrite
nitrogen oxides
Ozone
Office of Air Quality Planning and Standards
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
xxxi
-------
RR
SAB
SEDD
SES
SID
SO2
STE
TRIM Expo
VE
VNA
voc
WHO
relative risk
Science Advisory Board
State Emergency Department Databases
socioeconomic status
State Inpatient Databases
sulfur dioxide
stratosphere-troposphere exchange
Total Risk Integrated Methodology Inhalation Exposure
ventilation rate
Voronoi Neighbor Averaging
volatile organic carbon
World Health Organization
xxxn
-------
1 1 INTRODUCTION
2 The U.S. Environmental Protection Agency (EPA) is presently conducting a review of
3 the national ambient air quality standards (NAAQS) for ozone (63), and related photochemical
4 oxidants. The NAAQS review process includes four key phases: planning, science assessment,
5 risk/exposure assessment, and policy assessment/rulemaking.l This process and the overall plan
6 for this review of the Os NAAQS is presented in the Integrated Review Plan for the Ozone
1 National Ambient Air Quality Standards (IRP, U. S. EPA, 2011 a). The IRP additionally presents
8 the schedule for the review; identifies key policy-relevant issues; and discusses the key scientific,
9 technical, and policy documents. These documents include an Integrated Science Assessment
10 (ISA), Risk and Exposure Assessments (REAs), and a Policy Assessment (PA). This draft Health
11 REA is one of the two quantitative REAs developed for the review by the EPA's Office of Air
12 Quality Planning and Standards (OAQPS); the second is a Welfare REA. This draft Health REA
13 focuses on assessments to inform consideration of the review of the primary (health-based)
14 NAAQS for O3..
15 The existing primary (health-based) NAAQS for Os is set at a level of 75 ppb (0.075
16 ppm), based on the annual fourth-highest daily maximum 8-hour average concentration,
17 averaged over three years, and the secondary standard is identical to the primary standard (73 FR
18 16436). The EPA initiated the current review of the O3 NAAQS on September 29, 2008, with an
19 announcement of the development of an 63 ISA and a public workshop to discuss policy-
20 relevant science to inform EPA's integrated plan for the review of the Os NAAQS (73 FR
21 56581). Discussions at the workshop, held on October 29-30, 2008, informed identification of
22 key policy issues and questions to frame the review of the Os NAAQS. Drawing from the
r\
23 workshop discussions, the EPA developed a draft and then final IRP (U.S. EPA, 2011). In early
24 2013, the EPA completed the Integrated Science Assessment for Ozone and Related
25 Photochemical Oxidants (U.S. EPA, 2013). The ISA provides a concise review, synthesis and
26 evaluation of the most policy-relevant science to serve as a scientific foundation for the review
27 of the NAAQS. The scientific and technical information in the ISA, including that newly
1 For more information on the NAAQS review process see http://www.epa.gov/ttn/naaqs/review.html.
2 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.
1-1
-------
1 available since the previous review on the health effects of O?, includes information on exposure,
2 physiological mechanisms by which O^ might adversely impact human health, an evaluation of
3 the toxicological and controlled human exposure study evidence, and an evaluation of the
4 epidemiological evidence, including information on reported concentration-response (C-R)
5 relationships for Os-related morbidity and mortality associations, and also includes information
6 on potentially at-risk populations and life-stages.3
7 This REA is a concise presentation of the conceptual model, scope, methods, key results,
8 observations, and related uncertainties associated with the quantitative analyses performed. This
9 REA builds upon the health effects evidence presented and assessed in the ISA, as well as
10 CAS AC advice (Samet, 2011), and public comments on a scope and methods planning document
11 for the REA (here after, "Scope and Methods Plan," U.S. EPA, 2011). Preparation of this second
12 draft REA draws upon the final ISA and reflects consideration of CASAC and public comments
13 on the first draft REA (Frey and Samet, 2012). This second draft health REA is being released,
14 concurrently with the second draft welfare REA and second draft PA for review by the CASAC
15 Os Panel at a public meeting scheduled for March 25-27, 2014, and for public comment.
16 The second draft PA presents a staff evaluation and preliminary staff conclusions of the
17 policy implications of the key scientific and technical information in the ISA, and second draft
18 REAs. When final, the PA is intended to help "bridge the gap" between the Agency's scientific
19 assessments presented in the ISA and REAs, and the judgments required of the EPA
20 Administrator in determining whether it is appropriate to retain or revise the NAAQS. The PA
21 integrates and interprets the information from the ISA and REAs to frame policy options for
22 consideration by the Administrator. In so doing, the PA recognizes that the selection of a specific
23 approach to reaching final decisions on primary and secondary NAAQS will reflect the
24 judgments of the Administrator. The development of the various scientific, technical and policy
25 documents and their roles in informing this NAAQS review are described in more detail in the
26 second draft PA.
3 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 draft of a second REA titled Ozone Welfare Effects Risk and Exposure Assessment
(U.S. EPA, 2013).
1-2
-------
1 1.1 HISTORY
2 As part of the last O3 NAAQS review completed in 2008, EPA's OAQPS conducted
3 quantitative risk and exposure assessments to estimate exposures above health benchmarks and
4 risks of various health effects associated with exposure to ambient 63 in a number of urban study
5 areas, selected to illustrate the public health impacts of this pollutant (U.S. EPA 2007a, U.S.
6 EPA, 2007b). The assessment scope and methodology were developed with considerable input
7 from CASAC and the public, with CASAC generally concluding that the exposure assessment
8 reflected generally accepted modeling approaches, and that the risk assessments were well done,
9 balanced and reasonably communicated (Henderson, 2006a). The final quantitative risk and
10 exposure assessments took into consideration CASAC advice (Henderson, 2006a; Henderson,
11 2006b), and public comments on two drafts of the risk and exposure assessments.
12 The exposure and health risk assessment conducted in the last review developed exposure
13 and health risk estimates for 12 urban areas across the U.S., based on 2002 to 2004 air quality
14 data. That assessment provided annual or Os season-specific exposure and risk estimates for
15 these years of air quality and for air quality scenarios, simulating just meeting the then-existing
16 8-hour Os standard set in 1997 at a level of 0.08 ppm and several alternative 8-hour standards.
17 The strengths and limitations in the assessment were characterized, and analyses of key
18 uncertainties were presented.
19 Exposure estimates from the last assessment were used as an input to the risk assessment
20 for lung function responses (a health endpoint for which exposure-response functions were
21 available from controlled human exposure studies). Exposure estimates were developed for the
22 general population and population groups including school age children with asthma as well as
23 all school age children. The exposure estimates also provided information on exposures to
24 ambient O?, concentrations at and above specified benchmark levels (referred to as "exposures of
25 concern"), to provide some perspective on the public health impacts of health effects associated
26 with Oj, exposures in controlled human exposure studies that could not be evaluated in the
27 quantitative risk assessment (e.g., lung inflammation, increased airway responsiveness, and
28 decreased resistance to infection). For several other health endpoints, Os-related risk estimates
29 were generated using concentration-response relationships reported in epidemiological or field
30 studies, together with ambient air quality concentrations, baseline health incidence rates, and
31 population data for the various locations included in the assessment. Health endpoints included
1-2
-------
1 in the assessment based on epidemiological or field studies included: hospital admissions for
2 respiratory illness in four urban areas, premature mortality in 12 urban areas, and respiratory
3 symptoms in asthmatic children in 1 urban area.
4 Based on the 2006 Air Quality Criteria for Ozone (U.S. EPA, 2006), the Staff Paper
5 (U.S. EPA, 2007), and related technical support documents (including the REAs), the proposed
6 decision was published in the Federal Register on July 11, 2007 (72 FR 37818). The EPA
7 proposed to revise the level of the primary standard to a level within the range of 0.075 to 0.070
8 ppm. Two options were proposed for the secondary standard: (1) replacing the current standard
9 with a cumulative seasonal standard, expressed as an index of the annual sum of weighted hourly
10 concentrations cumulated over 12 daylight hours during the consecutive 3-month period within
11 the Os season with the maximum index value (W126), set at a level within the range of 7 to 21
12 ppm-hours, and (2) setting the secondary standard identical to the revised primary standard. The
13 EPA completed the review with publication of a final decision on March 27, 2008 (73 FR
14 16436), revising the level of the 8-hour primary Os standard from 0.08 ppm to 0.075 ppm, as the
15 3-year average of the fourth highest daily maximum 8-hour average concentration, and revising
16 the secondary standard to be identical to the revised primary standard.
17 Following promulgation of the revised Os standard in March 2008, state, public health,
18 environmental, and industry petitioners filed suit against EPA regarding that final decision.
19 At EPA's request, the consolidated cases were held in abeyance pending EPA's
20 reconsideration of the 2008 decision. A notice of proposed rulemaking to reconsider the
21 2008 final decision was issued by the Administrator on January 6, 2010. Three public
22 hearings were held. The Agency solicited CASAC review of the proposed rule on January
23 25, 2010, and additional CASAC advice on January 26, 2011. On September 2, 2011, the
24 Office of Management and Budget returned the draft final rule on reconsideration to EPA for
25 further consideration. EPA decided to coordinate further proceedings on its voluntary
26 rulemaking on reconsideration with this ongoing periodic review, by deferring the
27 completion of its voluntary rulemaking on reconsideration until it completes its statutorily-
28 required periodic review. In light of that, the litigation on the 2008 final decision proceeded.
29 On July 23, 2013, the Court ruled on the litigation of the 2008 decision, denying the
30 petitioners suit except with respect to the secondary standard, which was remanded to the
31 Agency for reconsideration. The second draft PA provides additional description of the court
1-4
-------
1 ruling with regard to the secondary standard.
2 1.2 CURRENT RISK AND EXPOSURE ASSESSMENT: GOALS AND PLANNED
3 APPROACH
4 The goals of the current quantitative exposure and health risk assessments are to provide
5 information relevant to answering questions regarding the adequacy of the existing 63 standard
6 and the potential improvements in public health from meeting alternative standards. To meet
7 these goals, this assessment provides results from several analyses, including (1) estimates of the
8 number of people in the general population and in at-risk populations and lifestages with 63
9 exposures above benchmark levels, while at moderate or greater exertion levels; (2) estimates of
10 the number of people in the general population and in at-risk populations and lifestages with
11 impaired lung function resulting from exposures to O^; and (3) estimates of the potential
12 magnitude of premature mortality and selected morbidity health effects in the population,
13 including at-risk populations and lifestages, where data are available to assess these groups. For
14 each of the analyses, we provide estimates for recent ambient levels of 63 and for air quality
15 conditions simulated to just meet the existing 63 standard and alternative standards.
16 In presenting these results, we evaluate the influence of various inputs and assumptions
17 on the exposure and risk estimates to more clearly differentiate alternative standards that might
18 be considered, including potential impacts on various at-risk populations and lifestages. We also
19 evaluate the distribution of risks and patterns of risk reduction and uncertainties in those risk
20 estimates. In addition, we have conducted an assessment to provide nationwide estimates of the
21 potential magnitude of premature mortality associated with recent ambient 63 concentrations, to
22 more broadly characterize this risk on a national scale. This assessment includes an evaluation of
23 the distribution of risk across the U.S., to assess the extent to which we have captured the upper
24 end of the risk distribution with our urban study area analyses.
25 This current quantitative risk and exposure assessment builds on the approach used and
26 lessons learned in the last Os risk and exposure assessment, and focuses on improving the
27 characterization of the overall confidence in the exposure and risk estimates, including related
28 uncertainties, by incorporating a number of enhancements, in terms of both the methods and data
29 used in the analyses. This risk assessment considers a variety of health endpoints for which, in
30 staff s judgment, there is adequate information to develop quantitative risk estimates that can
31 meaningfully inform the review of the primary Os NAAQS.
1-5
-------
1 1.3 ORGANIZATION OF DOCUMENT
2 The remainder of this document is organized as follows. Chapter 2 provides a conceptual
3 framework for the risk and exposure assessment, including discussions of 63 chemistry, sources
4 of 63 precursors, exposure pathways and microenvironments where 63 exposure can be high, at-
5 risk populations and lifestages, and health endpoints associated with 63. This conceptual
6 framework sets the stage for the scope of the risk and exposure assessments. Chapter 3 provides
7 an overview of the scope of the quantitative risk and exposure assessments, including a summary
8 of the previous risk and exposure assessments, and an overview of the current risk and exposure
9 assessments. Chapter 4 discusses air quality considerations relevant to the exposure and risk
10 assessments, including available 63 monitoring data, and important inputs to the risk and
11 exposure assessments. Chapter 5 describes the inputs, models, and results for the human
12 exposure assessment, and discusses the literature on exposure to 63, exposure modeling
13 approaches using the Air Pollution Exposure Model (APEX), the scope of the exposure
14 assessment, inputs to the exposure modeling, sensitivity and uncertainty evaluations, and
15 estimation of results. Chapter 6 describes the estimation of health risks based on application of
16 the results of controlled human exposure studies, including discussions of health endpoint
17 selection, approaches to calculating risk, and results. Chapter 7 describes the estimation of health
18 risks in selected urban areas based on application of the results of observational epidemiology
19 studies, including discussions of air quality characterizations, model inputs, variability and
20 uncertainty, and results. Chapter 8 describes the national scale risk characterization and urban
21 area representativeness analysis. Chapter 9 provides an integrative discussion of the exposure
22 and risk estimates generated in the analyses drawing on the results of the analyses based on both
23 clinical and epidemiology studies, and incorporating considerations from the national scale risk
24 characterization.
1-6
-------
1 2 OVERVIEW OF EXPOSURE AND RISK ASSESSMENT DESIGN
2 In this chapter, we summarize our framework for assessing exposures to 63 and the
3 associated risks to human populations. Figure 2-1 provides an overview of the general design of
4 this exposure and risk assessment, which includes air quality characterization, review of relevant
5 scientific evidence on health effects, modeling of exposure, modeling of risk, and risk
6 characterization. Each element identified in the diagram is described in a specific, identified
7 chapter of this exposure and risk assessment.
8 In this OT, exposure and risk assessment, modeling of personal exposure and estimation of
9 risks which rely on personal exposure estimates, are implemented using the Air Pollution
10 Exposure model (APEX)1 (U.S. EPA, 2012 a, b). Modeling of population level risks for
11 endpoints based on application of results of epidemiological studies, is implemented using the
12 environmental Benefits Mapping and Analysis Program (BenMAP),2 a peer reviewed software
13 tool for estimating risks and impacts associated with changes in ambient air quality (U.S. EPA,
14 2013). The overall characterization of risk draws from the results of the exposure assessment and
15 both types of risk assessment.
16 The remainder of this chapter includes summary discussions of each of the main elements
17 of Figure 2-1, including policy-relevant exposure and risk questions (Section 2.1),
18 characterization of ambient 63, including important sources of 63 precursors, and its relation to
19 population exposures, as well as simulation of just meeting existing and potential alternative Os
20 standards (Section 2.2), review of health evidence identified in the literature describing
21 associations with ambient 63 (Section 2.3), key components of exposure modeling (Section 2.4),
22 key components of risk modeling (Section 2.5), and risk characterization (Section 2.6).
23 Specific details related to the scope of the exposure and risk assessments and how each
24 element will be addressed in the quantitative exposure and risk analysis are provided in Chapter
25 3.
26
1 APEX is available for download at http://www.epa.gov/ttn/fera/human_apex.html
2 BenMAP is available for download at http://www.epa.gov/ai^enmap/
2-1
-------
Policy Relevant Exposure
and Risk Questions
(Chapter2)
Exposure Assessment
APEX
Urban Scale
Assessment of
Individual Exposure
(Chapters)
Air Quality Characterization
(Chapter4)
Review of Health Evidence
(Chapter2)
Urban Scale Risk
Analyses Based on
Application of
Results from
Controlled Human
Exposure Studies
(ChapterB)
r '
RiskAssessment
BenMAP
Urban Scale Risk
Analyses Based on
Application of
Results from
Epidemiological
Studies
(Chapter?)
National Scale Risk
Burden Basedon
Application of
Results from
Epidemiological
Studies
(Chapters)
Risk Characterization
(Chapter 9)
1
2 Figure 2-1 Overview of Exposure and Risk Assessment Design
3
4 2.1 POLICY-RELEVANT EXPOSURE AND RISK QUESTIONS
5 The first step in the design is to determine the set of policy-relevant exposure and risk
6 questions that will be informed by the assessment. Consistent with recommendations from the
7 recent National Academy of Sciences report "Science and Decisions: Advancing Risk
8 Assessment" (NAS, 2009), these exposure and risk assessments have been designed to address
9 the risk questions identified in the Integrated Review Plan for the Ozone National Ambient Air
10 Quality Standards (U.S. EPA, 2011). We have focused on designing the exposure and risk
11 assessments to inform consideration of those risk-related policy-relevant questions in the
12 separately developed Os NAAQS Policy Assessment. The risk-related policy-relevant questions
13 identified in the Integrated Review Plan are related to two main activities, evaluation of the
14 adequacy of the existing standards and, if appropriate, evaluation of potential alternative
15 standards (U.S. EPA, 2011). With regard to evaluation of the adequacy of the existing standards,
16 the risk-related policy-relevant questions are:
2-2
-------
1 "To what extent do risk and/or exposure analyses suggest that exposures of
2 concern for Os-related health effects are likely to occur with existing ambient
3 levels of 03 or with levels that just meet the Oj standard? Are these
4 risks/exposures of sufficient magnitude such that the health effects might
5 reasonably be judged to be important from a public health perspective? What are
6 the important uncertainties associated with these risk/exposure estimates? "
1 With regards to evaluation of potential alternative standards, the risk-related policy-relevant
8 questions are:
9 "To what extent do alternative standards, taking together levels, averaging times
10 and forms, reduce estimated exposures and risks of concern attributable to O3
11 and other photochemical oxidants, and what are the uncertainties associated with
12 the estimated exposure and risk reductions? What conclusions can be drawn
13 regarding the health protection afforded at-riskpopulations? "
14
15 This risk and exposure assessment is designed to inform consideration of these questions
16 through application of exposure and risk modeling for a set of urban case study areas. Exposure
17 and risk estimates will be generated for recent O3 concentrations, O3 concentrations after
18 simulating just meeting the existing standards, and O3 concentrations after simulating just
19 meeting potential alternative standards. Careful consideration will be given to addressing
20 variability and uncertainty in the estimates, and to the degree to which at-risk populations
21 experience exposures and risks. Exposure modeling is discussed in Chapter 5 (Urban-Scale
22 Assessment of Individual Exposure), while risk modeling is discussed in Chapter 6
23 (Characterization of Health Risks Based on Clinical Studies) and Chapter 7 (Characterization of
24 Health Risks Based on Epidemiological Studies). Chapter 8 (National-Scale Risk Assessment
25 and Representativeness Analysis) provides a national-scale assessment of risks under recent O3
26 concentrations to provide context for the urban-scale analyses and to help characterize the
27 representativeness of the urban-scale analyses.
28 In order to inform consideration of the risk-related policy-relevant questions, the first step
29 for all of the exposure and risk analyses is simulation of meeting the existing and alternative
30 standards. To do this, recent air quality measurements of Os are adjusted such that they mimic a
31 realistic and general atmospheric response to changes in precursor emissions for the specific
32 urban area and so that they just meet the existing and alternative standard levels. Conceptually,
33 there is an almost infinite set of combinations of precursor emissions reductions that will result
34 in just meeting the existing or alternative standards. The specific combinations of reductions that
35 might actually be implemented are not relevant for the exposure and risk analyses, as those will
36 result from the implementation processes which follow the establishment of a standard.
2-3
-------
1 However, it is appropriate to ask the question of how the patterns of ambient O?, on multiple
2 temporal scales (hourly, daily, monthly, seasonally) and across each urban area, may respond to
3 precursor emissions reductions that result in meeting the existing and potential alternative
4 standards, and how these different patterns of Os could affect the exposure and risk results. The
5 answers to these questions are critical inputs to the exposure and risk analyses. Consideration of
6 the available methods for simulating just meeting existing and alternative standards is discussed
7 in Chapter 4 (Air Quality Characterization).
8 Analyses presented in this document to inform the policy-relevant risk questions
9 regarding potential alternative standards, are focused on alternative levels for an 8-hour standard.
10 Other elements of the standard (indicator, averaging time, and form),3 are addressed in the Policy
11 Assessment as part of the overall evaluation of the health protection afforded by the primary Os
12 standards.
13 With regard to potential alternative levels for an 8-hour 63 standard, the quantitative risk
14 assessment evaluates the range of levels in 5 ppb increments from 60 to 70 ppb. These levels
15 were selected based on the evaluations of the evidence provided in the first draft PA, which
16 received support from the CAS AC in their advisory letter on the first draft PA (Frey and Samet,
17 2012). For a subset of urban areas, we also evaluated a standard level of 55 ppb, consistent with
18 recommendations from CAS AC to also give consideration to evaluating a level somewhat below
19 60 ppb. Thus, for most areas, we evaluate exposures and risks for potential alternative standard
20 levels of 70, 65, and 60 ppb. Some additional analyses were also included for evaluation of
21 exposures and risks for a potential alternative 8-hour standard level of 55 ppb.
22 2.2 AIR QUALITY CHARACTERIZATION
23 In order to address the policy-relevant questions discussed in Section 2.1, the first step is
24 characterizing 63 concentrations relevant to estimation of exposure and risk. This requires
25 characterization of recent OT, concentrations, Os concentrations after simulating just meeting the
26 existing standards, and 63 concentrations after simulating just meeting potential alternative
27 standards. This section provides conceptual information on 63 formation and responsiveness of
28 Os to changes in precursor emissions, that inform the simulations of just meeting existing and
29 alternative standards.
3 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
-------
1 2.2.1 Os chemistry and response to changes in precursor emissions
2 O3 occurs naturally in the stratosphere where it provides protection against harmful solar
3 ultraviolet radiation, and it is formed closer to the surface in the troposphere from precursor
4 emissions from both natural and anthropogenic sources. O3 is created when its two primary
5 precursors, volatile organic compounds (VOC) and oxides of nitrogen (NOX), combine in the
6 presence of sunlight. VOC and NOX are, for the most part, emitted directly into the atmosphere.
7 Carbon monoxide (CO) and methane (CH/j) can also be important for O3 formation (U.S. EPA,
8 2013, section 3.2.2).
9 Rather than varying directly with emissions of its precursors, O3 changes in a nonlinear
10 fashion with the concentrations of its precursors. NOX emissions lead to both the formation and
11 destruction of O3, depending on the local concentrations of NOX, VOC, and radicals such as the
12 hydroxyl (OH) and hydroperoxy (HO2) radicals. In areas dominated by fresh emissions of NOX,
13 these radicals are removed via the production of nitric acid (HNO3), which lowers the O3
14 formation rate. In addition, the depletion of O3 by reaction with NO is called "titration" and is
15 often found in downtown metropolitan areas, especially near busy streets and roads, and in
16 power plant plumes. This "titration" results in O3 concentrations that can be much lower than in
17 surrounding areas. Titration is usually confined to areas close to strong NOX sources, and the
18 NO2 formed can lead to O3 formation later and further downwind. Consequently, O3 response to
19 reductions in NOX emissions is complex and may include O3 decreases at some times and
20 locations and increases of O3 in other times and locations. In areas with low NOX concentrations,
21 such as those found in remote continental areas and rural and suburban areas downwind of urban
22 centers, the net production of O3 typically varies directly with NOX concentrations, and increases
23 with increasing NOX emissions.
24 In general, the rate of O3 production is limited by either the concentration of VOCs or
25 NOX, and O3 formation, using these two precursors relies on the relative sources of OH and NOX.
26 When OH radicals are abundant and are not depleted by reaction with NOX and/or other species,
27 O3 production is referred to as being "NOx-limited" (U.S. EPA, 2013, section 3.2.4). In this
28 situation, O3 concentrations are most effectively reduced by lowering NOX emissions, rather than
29 lowering emissions of VOCs. When the abundance of OH and other radicals is limited either
30 through low production or reactions with NOX and other species, O3 production is sometimes
31 called "VOC-limited" or "radical limited" or "NOX-saturated" (Jaegle et al., 2001), and O3 is
32 most effectively reduced by lowering VOCs. However, even in NOx-saturated conditions, very
33 large decreases in NOX emissions can cause the O3 formation regime to become NOx-limited.
34 Consequently, reductions in NOX emissions (when large), can make further emissions reductions
35 more effective at reducing O3. Between the NOx-limited and NOx-saturated extremes there is a
36 transitional region, where O3 is less sensitive to marginal changes in either NOX or VOCs. In
2-5
-------
1 rural areas and downwind of urban areas, Os production is generally NOx-limited. However,
2 across urban areas with high populations, conditions may vary. For contrast, while data from
3 monitors in Nashville, TN, suggest NOx-limited conditions exist there, data from monitors in Los
4 Angeles suggest NOx-saturated conditions (U.S. EPA, 2013, Figure 3-3).
5 2.2.2 Sources of Os and Os Precursors
6 Os precursor emissions can be divided into anthropogenic and natural source categories,
7 with natural sources further divided into biogenic emissions (from vegetation, microbes, and
8 animals), and abiotic emissions (from biomass burning, lightning, and geogenic sources). The
9 anthropogenic precursors of Os originate from a wide variety of stationary and mobile sources.
10 In urban areas, both biogenic and anthropogenic VOCs, as well as CO, are important for
11 Os formation. Hundreds of VOCs are emitted by evaporation and combustion processes from a
12 large number of anthropogenic sources. Based on the 2005 national emissions inventory (NEI),
13 solvent use and highway vehicles are the two main anthropogenic sources of VOCs, with
14 roughly equal contributions to total emissions (U.S. EPA, 2013, Figure 3-2). The emissions
15 inventory categories of "miscellaneous" (which includes agriculture and forestry, wildfires,
16 prescribed burns, and structural fires), and off-highway mobile sources are the next two largest
17 contributing emissions categories with a combined total of over 5.5 million metric tons a year
18 (MT/year).
19 On the U.S. and global scales, emissions of VOCs from vegetation are much larger than
20 those from anthropogenic sources. Emissions of VOCs from anthropogenic sources in the 2005
21 NEI were -17 MT/year (wildfires constitute -1/6 of that total), compared to emissions from
22 biogenic sources of 29 MT/year. Vegetation emits substantial quantities of VOCs, such as
23 isoprene and other terpenoid and sesqui-terpenoid compounds. Most biogenic emissions occur
24 during the summer because of their dependence on temperature and incident sunlight. Biogenic
25 emissions are also higher in southern and eastern states than in northern and western states for
26 these reasons and because of species variations.
27 Anthropogenic NOX emissions are associated with combustion processes. Based on the
28 2005 NEI, the three largest sources of NOX are on-road and off-road mobile sources (e.g.,
29 construction and agricultural equipment), and electric power generation plants (EGUs) (U.S.
30 EPA, 2013, Figure 3-2). Emissions of NOX therefore are highest in areas having a high density of
31 power plants and in urban areas having high traffic density. However, it is not possible to make
32 an overall statement about their relative impacts on Os in all local areas because EGUs are
33 sparser than mobile sources, particularly in the west and south and because of the nonlinear
34 nature of Os chemistry discussed in Section 2.2.1.
2-6
-------
1 Major natural sources of NOX in the U.S. include lightning, soils, and wildfires. Biogenic
2 NOX emissions are generally highest during the summer and occur across the entire country,
3 including areas where anthropogenic emissions are low. It should be noted that uncertainties in
4 estimating natural NOX emissions are much larger than for anthropogenic NOX emissions.
5 63 concentrations in a region are maintained by a balance between photochemical
6 production and transport of Os into the region; and loss of Os by chemical reactions, deposition
7 to the surface and transport out of the region. 63 transport occurs on many spatial scales
8 including local transport between cities, regional transport over large regions of the U.S. and
9 international/long-range transport. In addition, Os is also transfered into the troposphere from the
10 stratosphere, which is rich in 63, through stratosphere-troposphere exchange (STE). STE occurs
11 in tropopause "foldings" that occur behind cold fronts, bringing stratospheric air with them (U.S.
12 EPA, 2013, section 3.4.1.1). Contributions to Os concentrations in an area from STE are defined
13 as being part of background 63 (U.S. EPA, 2013, section 3.4).
14 2.2.3 Simulation of Meeting Existing and Alternative Standards
15 Conceptually, simulation of meeting existing and alternative standards should reflect the
16 physical and chemical processes of 63 formation in the atmosphere and estimate how hourly
17 values of Os at each monitor in an urban area would change in response to reductions in
18 precursor emissions, allowing for nonlinearities in response to emissions reductions and allowing
19 for nonlinear interactions between reductions in NOX and VOC emissions. For this assessment,
20 we have employed sophisticated air quality models to conduct simulations of hourly 03
21 responses to reductions in precursor emissions. This modeling incorporates all known emissions,
22 including emissions from both natural and anthropogenic sources within and outside of the U.S.
23 By using the model-adjustment methodology we are able to more realistically simulate the
24 temporal and spatial patterns of 63 response to precursor emissions. We chose to simulate just
25 meeting the existing and alternative standards, by applying equal proportional decreases in U.S.
26 anthropogenic emissions of NOx and VOC, in order to avoid any suggestion that we are
27 approximating a specific emissions control strategy that a state or urban area might adopt to meet
28 a standard. These analyses allow us to apply an adjustment to ambient 03 measurements in the
29 urban case study areas, to better represent how air quality concentrations at each monitor would
30 change to meet the existing and alternative standard levels. The details of the specific approach
31 used to simulating attainment for the existing and alternative standards, are discussed in greater
32 detail in Chapter 4 and in the Chapter 4 appendices.
33 It is fundamentally a policy decision, as to which sources of precursor emissions are most
34 appropriate to decrease to simulate just meeting existing and alternative 03 standards. In
35 addressing the policy-relevant questions regarding the evaluation of alternative standards,
2-7
-------
1 consistent with previous reviews of the O?, standards, this analysis is focused on simulating
2 reductions in risk associated with precursor emissions originating from anthropogenic sources
3 within the U.S. In doing so, we recognize that the CAA provides mechanisms primarily for
4 reducing emissions from U.S. emissions sources. As such, we estimate changes in exposure and
5 risks likely to result from just meeting alternative standards relative to just meeting the existing
6 standards, by simulating changes in atmospheric concentrations that represent atmospheric
7 response to reductions in U.S. anthropogenic emissions. However, we recognize that, in this
8 approach, we are simulating attainment of existing and alternative standard levels, based on
9 recent air quality concentrations and the chemical environment and emissions in those years. We
10 have not mimicked the future-year atmospheric conditions and emissions inventory as would be
11 done for the implementation process.
12 In addition, while it is possible to decrease O^ concentrations using decreases in either
13 NOx or VOC or both NOX and VOC, the specific combination of the reductions in those
14 emissions is a policy decision, with recognition that atmospheric chemistry considerations will
15 make NOX and VOC decreases more or less effective in specific urban areas, depending on the
16 degree to which 63 formation is NOX or VOC limited. As discussed above, in most locations,
17 decreases in NOX are the most effective means to decrease ambient Os concentrations. However,
18 in some downtown urban areas, 63 formation is VOC-limited, and therefore smaller decreases in
19 NOX will not decrease O?,.
20 2.2.4 Consideration of Health Evidence
21 A critical input for both the exposure and risk assessments is the health evidence
22 summarized in the Integrated Science Assessment (ISA) (U.S. EPA, 2013). This health evidence
23 provides the basis for evaluating the significance of exposures to Os, by informing health
24 benchmarks for estimating exposures of concern. The evidence also provides the basis for
25 selecting health endpoints that will be modeled in the risk assessment. This evidence includes
26 controlled human exposure studies and observational epidemiology studies. The health evidence
27 is also the source of the specific studies that are used to develop exposure-response (E-R) and
28 concentration-response (C-R) functions, used in the risk assessment. Finally, the health evidence
29 provides information on at-risk populations to guide the selections of study populations used in
30 the exposure and risk assessments. The following subsections summarize key conceptual aspects
31 regarding exposures of concern, health endpoints, E-R and C-R functions, and at-risk
32 populations.
2-8
-------
1 2.2.5 Exposures of Concern
2 The 63 ISA identifies health effects associated with exposures to varying concentrations
3 of 63. However, not all of the evidence is suitable for evaluation in a quantitative risk
4 assessment. Estimating exposures to ambient 03 concentrations at and above benchmark levels
5 where health effects have been observed in studies provides a perspective on the public health
6 impacts of Cb-related health effects that have been demonstrated in human clinical and
7 toxicological studies but cannot currently be evaluated in quantitative risk assessments, such as
8 lung inflammation, increased airway responsiveness, and decreased resistance to infection.
9 To inform the selection of benchmark levels for Os exposure, it is appropriate to consider
10 the evidence from clinical studies which have evaluated individual controlled levels of 63
11 exposure. There is substantial clinical evidence demonstrating a range of (Vrelated effects
12 including lung inflammation and airway responsiveness in healthy individuals at an exposure
13 level of 0.080 ppm. There is additional evidence that asthmatics have larger and more serious
14 effects than healthy people at 0.070 ppm, as well as a substantial body of epidemiological
15 evidence of associations with 03 levels that extend well below 0.080 ppm. There is a more
16 limited set of evidence based on clinical studies of healthy individuals exposed at 0.060 ppm in
17 which Os-related effects have been observed. This is the lowest level at which any Os-related
18 effects have been observed in clinical studies of healthy individuals (U.S. EPA, 2013, section
19 6.2.1).
20 Thus, benchmark levels of 0.060 ppm, 0.070 ppm, and 0.080 ppm are used in this
21 assessment to characterize exposures of concern for a range of potential health effects in healthy
22 and at-risk populations exposed to Os.
23 2.2.6 Health Endpoints
24 The OsISA identifies a wide range of health outcomes associated with short-term
25 exposure to ambient Os, including an array of morbidity effects as well as
26 premature mortality. The ISA also identifies several morbidity effects and some
27 evidence for premature mortality associated with longer-term exposures to Os. In
28 identifying health endpointsfor risk assessment, we have focused on endpoints
29 that pertain to at-risk populations, have public health significance, and for which
30 information is sufficient to support a quantitative concentration-response
2-9
-------
1 relationship, in the case of epidemiological studies, or exposure-response
2 relationship, in the case of controlled human exposure studies.4
3 In considering such endpoints for O3, we draw from two types of studies: controlled
4 human exposure and epidemiological studies. Each study type informs our characterization of
5 O3 risk and can do so in different ways. Estimates of risk based on results of controlled human
6 exposure studies are valuable because they provide clear evidence of the detrimental effects of
7 controlled (and measured) exposures to 63 over multiple hours on lung function at moderate
8 levels of exertion. Results of these studies can be applied to modeled estimates of population
9 exposure to provide insights into population exposure characteristics, including types of activity
10 patterns and microenvironments, which are associated with high levels of risk. Controlled human
11 exposure studies, however, cannot directly provide relationships for endpoints such as premature
12 death or hospitalizations, focusing more on intermediate biological endpoints including
13 inflammatory, blood, neurological, cardiovascular, and respiratory biomarkers or symptoms.
14 Estimates of risk based on concentration-response functions from observational epidemiology
15 studies can provide insights on risk for more serious or chronic health endpoints. For example,
16 epidemiological studies of 63 described in the ISA have evaluated associations between 63 and
17 various endpoints including respiratory symptoms, respiratory-related hospitalizations and
18 emergency department (ED) visits, and premature mortality (U.S. EPA, 2013, sections 6.2.9 and
19 6.3.4). Epidemiological studies also generally focus on a population residing in specific area,
20 which may reflect a broad range of susceptibilities and sensitivities. Controlled human exposure
21 studies typically involve a smaller number of individuals over a more limited range of health
22 status, in some cases focused on at-risk populations, such as asthmatics and individuals with
23 COPD. Lastly, while controlled human exposure studies directly measure the exposures eliciting
24 the recorded effects, epidemiology studies have not traditionally been based on observations of
25 personal exposure to ambient 03, relying instead on surrogate measures of population exposure.
26 Such surrogates are often based on simple averages of ambient 63 monitor observations. Thus,
27 with attention to their differing strengths and limitations, risk analyses based on each type of
28 study can inform the risk characterization.
29 The Os ISA makes overall causal determinations based on the full range of
30 evidence including epidemiological, controlled human exposure, and
4 The distinction between concentration-response and exposure-response 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
exposure-responserelationship, however, the epidemiology studies are actually providing a concentration-
response relationship, which captures the exposure-response relationship with errors in exposure measurement.
2-10
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
toxicological studies. Figure 2-1 shows the 0j health effects which have been
categorized by strength of evidence for causality in the Oj ISA (U.S. EPA, 2013,
chapter 2). The ISA determined there to be causal relationships between short-
term exposure to ambient Oj 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 Oj
and central nervous system effects. The ISA determined to also be a likely causal
relationship between long-term Oj 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 Oj exposures and total mortality as well as cardiovascular, reproductive and
developmental, and central nervous system effects.
14
15
16
17
OJ
IS]
O
Q.
X
OJ
m
O
E
o
LO
If]
OJ
l/l
o
Q.
X
OJ
m
O
E
i_
OJ
-t-i
tio
O
Central nervous
system effects
Cardiovascular effects
Total Mortality
Respi ratory effects
Suggestive
Likely
Cardiovascular effects Respi ratory effects
Reproductive and
developmental effects
Central nervous system
effects
Total Mortality
Figure 2-2 Causal Determinations for Os Health Effects
2-11
-------
1 The ISA identifies several specific respiratory responses to short-term O?, exposure that
2 have been evaluated in controlled human exposure studies (U.S. EPA, 2013, section 6.2.1).
3 These include decreased inspiratory capacity, decreased forced vital capacity (FVC) and forced
4 expiratory volume in 1 second (FEV1); mild bronchoconstriction; rapid, shallow breathing
5 patterns during exercise; symptoms of cough and pain on deep inspiration (PDI); and pulmonary
6 inflammation. While such studies document quantitative relationships between short-term Os
7 exposure and an array of respiratory-related effects, exposure-response data across a range of
8 concentrations sufficient for developing quantitative risk estimates are only available for 63-
9 related decrements in FEV1 (U.S. EPA, 2013, section 6.2.1).
10 Within the broad category of respiratory morbidity effects, the epidemiology literature
11 has provided effect estimates for a wide range of health endpoints associated with short-term Os
12 exposures which we have considered for risk assessment. These health endpoints include lung
13 function, respiratory symptoms and medication use, respiratory-related hospital admissions, and
14 emergency department visits. In the case of respiratory symptoms, the evidence is most
15 consistently supportive of the relationship between short-term ambient 63 metrics and
16 respiratory symptoms and asthma medication use in children with asthma, but not for a
17 relationship between O^ and respiratory symptoms in children without asthma (U.S. EPA, 2013,
18 section 6.2.9). In the case of hospital admissions, there is evidence of associations between short-
19 term ambient Os metrics and general respiratory-related hospital admissions as well as more
20 specific asthma-related hospital admissions (U.S. EPA, 2013, section 6.2.7.2).
21 With regard to mortality, studies have evaluated associations between short-term ambient
22 Os metrics and all-cause, non-accidental, and cause-specific (usually respiratory or
23 cardiovascular) mortality. The evidence from respiratory-related morbidity studies provides
24 strong support for respiratory-related mortality for which a causal determination has been made
25 (U.S. EPA, 2013, Table 2-3). There are also a number of large studies that have found
26 associations between 63 and all-cause and all non-accidental mortality for which a likely causal
27 determination has been made (U.S. EPA, 2013, Table 2-3). Thus, it is appropriate to assess risks
28 for respiratory-related mortality as well as for all-cause total mortality associated with 03
29 exposure. The ISA also reports a likely causal determination for short-term 63 and
30 cardiovascular effects, including cardiovascular mortality (U.S. EPA, 2013, Table 2-3). This
31 determination is supported by studies relating total and cardiovascular mortality, coupled with
32 evidence from animal toxicological studies and controlled human exposure studies which find
33 effects of 03 on systemic inflammation and oxidative stress. Cardiovascular mortality effects are
34 covered through the estimation of risks associated with total mortality, which is dominated by
35 cardiovascular mortality. There are not sufficient epidemiological studies of cardiovascular
2-12
-------
1 morbidity showing consistent associations to justify inclusion of any cardiovascular morbidity
2 endpoints in the quantitative risk assessment.
3 With regard to effects associated with long-term Os exposures, the ISA states that the
4 relationship between O^ and respiratory-related effects, including respiratory symptoms, new-
5 onset asthma, and respiratory mortality is likely causal (U.S. EPA, 2013, Table 2-3). This
6 suggests that for long-term exposures, when comparing the evidence for respiratory-related
7 mortality and total mortality, the evidence is strongest for respiratory-related mortality, which is
8 supported by the strong evidence for respiratory morbidity. As a result, it is appropriate to
9 include respiratory mortality rather than total mortality in the risk assessment and to give
10 consideration to inclusion of additional respiratory-related health endpoints.
11 2.2.7 Exposure and Concentration-response Functions for Health Endpoints
12 Estimation of risk requires characterization of the E-R and C-R functions along the full
13 range of potential exposures. For E-R functions, the evidence from individual controlled human
14 exposure studies provides responses for exposures at and above 60 ppb. McDonnell et al. (2012)
15 develop an integrated model of FEV1 response that is fit to the results from controlled human
16 exposure studies and find that a model with a threshold provides the best fit to the data. In
17 addition, the ISA notes that it is difficult to characterize the E-R relationship at and below 40 ppb
18 due to the dearth of data at these lower concentrations (U.S. EPA, 2013, section 2.5.4.4). Thus,
19 for the portion of the risk assessment based on application of results of controlled human
20 exposure studies, the threshold model is applied.
21 The evidence for a threshold in the C-R functions for mortality and morbidity outcomes
22 derived from the epidemiological literature is limited. In general, the epidemiological evidence
23 suggests a generally linear C-R function with no indication of a threshold. However, evaluation
24 of evidence for a threshold in the C-R function is complicated by the high degree of
25 heterogeneity between cities in the C-R functions and by the sparse data available at lower
26 ambient O3 concentrations (U.S. EPA, 2013, sections 2.5.4.4 and 2.5.4.5).
27 The ISA also evaluated whether the magnitude of the relationship between short-term
28 exposures to O^ and mortality changes at lower concentrations (e.g., whether the C-R function is
29 non-linear). The ISA concludes that epidemiologic studies that examined the shape of the C-R
30 curve and the potential presence of a threshold have indicated a generally linear C-R function
31 with no indication of a threshold in analyses that have examined 8-h max and 24-h avg O3
32 concentrations, and that the evidence supports less certainty in the shape of the C-R function at
33 the lower end of the distribution of Os concentrations, e.g., 24-hour average Os below 20 ppb,
34 due to the low density of data in this range (U.S. EPA, 2013, section 2.5.4.4). In the absence of
35 information in the scientific literature on alternative forms of C-R functions at low Oi
2-13
-------
1 concentrations, the best estimate of the C-R function is a linear, no-threshold function. The
2 scientific literature does not provide sufficient information with which to quantitatively
3 characterize any potential additional uncertainty in the C-R functions at lower Os concentrations
4 for use in the quantitative risk assessment.
5 Multiple exposures to elevated Os levels over the course of an Os season may result in
6 adaptation within exposed population. Evidence suggests that repeated or chronic exposures to
7 elevated 63 can result in morphologic and biochemical adaptation which reduces the impacts of
8 subsequent 63 exposures (U.S. EPA, 2013, section 6.2.1.1). This has implications for exposure
9 modeling, in that the effects of modeled repeat exposures on risk may be attenuated relative to
10 the effects of the initial exposures. The ISA notes that "neither tolerance nor attenuation should
11 be presumed to imply complete protection from the biological effects of inhaled O3, because
12 continuing injury still occurs despite the desensitization to some responses (U.S. EPA, 2013,
13 section 6.2.1.1)." The ISA reports that there are limited epidemiological studies evaluating
14 adaptation to the mortality effects of Os, although the limited evidence does suggest that
15 mortality effects are decreased in later months during the 63 season relative to earlier months
16 (U.S. EPA, 2013, section 6.3.3). The impact of this phenomenon on risks based on application of
17 results from epidemiological studies is likely to be small, because the relative risk estimates from
18 those studies already incorporate any adaptive phenomenon.
19 2.2.8 At-risk Populations
20 The Os ISA refers to "at-risk" populations as an all-encompassing term used for groups
21 with specific factors that increase the risk of an air pollutant- (e.g., 63) related health effect in a
22 population group (U.S. EPA, 2013, chapter 8). Populations or lifestages can experience elevated
23 risks from Os exposure for a number of reasons. These include high levels of exposure due to
24 activity patterns which include a high duration of time in high-Os locations, e.g., outdoor
25 recreation or work, high levels of activity which increase the dose of Os, e.g., high levels of
26 exercise, genetic or other biological factors, e.g., life stage, which predispose an individual to
27 sensitivity to a given dose of Os, pre-existing diseases, e.g., asthma or COPD, and
28 socioeconomic factors which may result in more severe health outcomes, e.g., low access to
29 primary care that can lead to increased emergency department visits or hospital admissions. To
30 consider risks to these populations, modeling of exposures to Os needs to incorporate
31 information on time spent by potentially at-risk populations in high Os locations. This requires
32 identification of populations with the identified exposure-related risk factors, e.g. children or
33 adults engaging in activities involving moderate to high levels of outdoor exertion, especially on
34 a repeated basis typical of student athletes or outdoor workers, as well as identifying populations
35 with high sensitivity to Os, e.g. asthmatic children. It also requires that information on Os
2-14
-------
1 concentrations be mapped to locations where at-risk populations are likely to be exposed, e.g.
2 near roadways where running may occur, or at schools or parks where children are likely to be
3 engaged in outdoor activities.
4 In addition to consideration of factors that lead to increased exposure to Os, modeling of
5 risk from 63 exposures should incorporate additional information on factors that can lead to
6 increased dose of Os for a given exposure, e.g., increased breathing rates during periods of
7 exertion. These factors are especially important for risk estimates based on application of the
8 results of controlled human exposure studies. For risk modeling based on application of
9 observational epidemiology results, it is also important to understand characteristics of study
10 populations that can impact observed relationships between ambient 63 and population health
11 responses.
12 The Os ISA identifies a number of factors which have been associated with modifications
13 of the effect of ambient 63 on health outcomes. Building on the causal framework used
14 throughout the OT, ISA, conclusions are made regarding the strength of evidence for each factor
15 that may contribute to increased risk of an Os-related health effect based on the evaluation and
16 synthesis of evidence across scientific disciplines. The 63 ISA categorizes potential risk
17 modifying factors by the degree of available evidence. These categories include "adequate
18 evidence," "suggestive evidence," "inadequate evidence," and "evidence of no effect." See
19 Table 8-1 of the O3 ISA for a discussion of these categories (U.S. EPA, 2013, chapter 8).
20 Factors categorized as having adequate evidence include asthma, lifestage (children less
21 than 18 years of age, adults older than 65 years of age), diets with nutritional deficiencies, and
22 working outdoors. For example, children are the group considered to be at greatest risk because
23 they breathe more air per unit of body weight, are more likely to be active outdoors when 63
24 levels are high, are more likely than adults to have asthma, and are in a critical time period of
25 rapid lung growth and organ development. Factors categorized as having suggestive evidence
26 include genetic markers, sex (some studies have shown that females are at greater risk of
27 mortality from OT, compared to males), low socioeconomic status, and obesity. Factors
28 characterized as having inadequate evidence include influenza and other respiratory infections,
29 COPD, cardiovascular disease, diabetes, hyperthyroidism, race, and smoking (U.S. EPA, 2013,
30 section 8.5, Table 8-6).
31 2.3 URBAN-SCALE MODELING OF INDIVIDUAL EXPOSURE
32 Estimates of human exposure to 63 provide important information to inform
33 consideration of policy-relevant questions identified in Section 2.2 regarding the occurrence of
34 exposures of concern under air quality conditions that meet existing and potential alternative
35 standards, and also to provide inputs to the portion of the risk assessment based on application of
2-15
-------
1 results of controlled human exposure studies. Studies that measure human exposure to Os are
2 limited. More commonly, human exposure is estimated using sophisticated models which
3 combine information on ambient Os concentrations in various microenvironments, e.g. near
4 roads, in schools, etc., with information on activity patterns for individuals sampled from the
5 general population or specific subpopulations, e.g. children with asthma.
6 Os exposure is highly dependent on the ambient Os concentrations in an urban area.
7 Given that these concentrations are variable from year to year, it is important to model multiple
8 years representing the range of variability on Os concentrations to provide a better
9 characterization of potential exposures of concern. In addition, other important sources of
10 variability and uncertainty affecting the exposure estimates should be characterized, including
11 uncertainty and variability in the data on time-activity patterns, Os concentrations, and
12 population inputs. This can be accomplished in part by modeling exposure for multiple urban
13 areas selected to represent variability in these underlying sources of variability.
14 This section briefly describes the conceptual foundation for key components of exposure
15 modeling, characterization of microenvironmental Os concentrations, and characterization of
16 human activity patterns, including behaviors intended to avert exposures to 63. In addition, a
17 brief discussion of key factors to consider in selecting urban case study areas for the exposure
18 analysis is provided. The specific exposure model used in this assessment, APEX, is described
19 more fully in Chapters 3 and 5. Characterization of ambient Os concentrations is discussed
20 earlier in this chapter and in greater detail in Chapter 4.
21 2.3.1 Microenvironmental Os Concentrations
22 Human exposure to Os involves the contact (via inhalation) between a person and the
23 pollutant in the various locations (or microenvironments) in which people spend their time. Os
24 concentrations in some indoor microenvironments, such as within homes or offices, are
25 considerably lower than Os concentrations in similarly located outdoor microenvironments,
26 primarily due to deposition processes and the transformation of Os into other chemical
27 compounds within those indoor microenvironments. Concentrations of Os may also be quite
28 different in roadway environments, such as might occur while an individual is in a vehicle.
29 Thus, three important classes of microenvironments that should be considered when
30 evaluating population exposures to ambient Os are indoors, outdoors, and in-vehicle. Within
31 each of these broad classes of microenvironments, there are many subcategories, reflecting types
32 of buildings, types of vehicles, etc. The Os ISA evaluated the literature on indoor-outdoor Os
33 concentration relationships and found that studies consistently show that indoor concentrations
34 of Os are often substantially lower than outdoor concentrations unless indoor sources are present.
35 This relationship is greatly affected by the air exchange rate, which can be affected by open
2-16
-------
1 windows, use of air conditioning, and other factors. Ratios of indoor to outdoor O^
2 concentrations generally range from about 0.1 to 0.4 (U.S. EPA, 2013, section 4.3.2). In some
3 indoor locations, such as schools, there can be large temporal variability in the indoor-outdoor
4 ratios because of differences in air exchange rates over the day. For example, during the school
5 day, there is an increase in open doors and windows, so the indoor-outdoor ratio is higher during
6 the school day compared with an overall average across all hours and days. In-vehicle
7 concentrations are also likely to be lower than ambient concentrations, although the literature
8 providing quantitative estimates is smaller. Studies of personal exposure to 63 have identified
9 that Os exposures are highest when individuals are in outdoor microenvironments, such as
10 walking outdoors midday, moderate when in vehicle microenvironments, and lowest in
11 residential indoor microenvironments (U.S. EPA, 2013, section 4.3.3). Thus the time spent
12 indoors, outdoors, and in vehicles is likely to be a critical component in estimating 03 exposures.
13 Because of localized chemistry, 63 concentrations on or near roadways can be much
14 lower than away from roadways. This is due to the high levels of NOx emissions from motor
15 vehicles, which can lead to NOx titration of 63, reducing 63 levels during times of peak traffic.
16 The ISA reports evidence that concentrations of NO, NC>2, and NOx are negatively correlated
17 with concentrations of O^ near busy roadways. Because few monitors are located in direct
18 proximity to roadways, it is important to account for differences between near-road Os
19 concentrations and ambient OT, measurements in modeling exposure.
20 2.3.2 Human Activity Patterns
21 Human exposure can be measured using several metrics. Exposure to ambient
22 concentrations is one such metric. It is also possible to model dose, which combines exposure
23 information with physiological parameters related to activity levels. In order to model exposure
24 to ambient concentrations, detailed information on the patterns of time spent in different
25 microenvironments is critical. In order to model Os dose, additional information on the activities
26 conducted while in those microenvironments is needed, along with data on physiological
27 parameters associated with different activities.
28 Several large-scale databases of human time-activity-location patterns have been
29 compiled. The most comprehensive of these databases in the Consolidated Human Activity
30 Database (CHAD), which has been the basis of several previous exposure analyses for previous
31 NAAQS reviews. These databases compile large numbers of diaries of time spent at different
32 activities in different locations collected as part of smaller studies. The ISA notes the high degree
33 of variability in activity patterns across the population, as well as the variability in time spent in
34 different microenvironments. Time-activity-location patterns vary by age group, as well as by
2-17
-------
1 region of the U.S. Children generally spend more time in outdoor locations and also generally
2 have higher activity levels in those environments.
3 The dose of O3 received for any given exposure in a microenvironment depends not only
4 on the activity levels and O^ concentrations in the microenvironment, but also on ventilation
5 rates, which are related to age, body weight, and other physiological parameters. Children
6 generally have lower ventilation rates than adults when considering the volume of air breathed
7 per unit time; however, they tend to have a greater oral breathing contribution than adults, and
8 due to smaller lung volumes and generally greater breathing frequencies, children breathe at
9 higher body mass or surface area normalized minute ventilation rates, relative to their lung
10 volumes. Both of these factors tend to increase their applied or intake dose normalized to lung
11 surface area. For example, when comparing daily body mass normalized ventilation rates,
12 children can have up to a factor of two greater ventilation rates when compared to that of adults.
13 During periods of high activity, ventilation rates for children and young adults can be nearly
14 double those during moderate activity. Thus, it is important to model levels of activity and
15 associated ventilation rate as well as time spent in different microenvironments.
16 In addition to modeling daily exposures, it may also be important to understand the
17 patterns of exposure over an 03 season, including multiple repeated exposures for the same
18 individuals. Some individuals or subpopulations may exhibit multiple high daily exposures due
19 to persistent patterns of high activity in microenvironments with high Os concentrations. For
20 example, children engaged in numerous outdoor sports over a summer O^ season may have
21 multiple exposures to elevated 63 levels.
22 Another important issue in characterizing exposure involves consideration of the extent
23 to which people in relevant population groups modify their behavior for the purpose of
24 decreasing their personal exposure to 63 based on information about predicted air quality levels
25 made public through the Air Quality Index (AQI). The AQI is the primary tool EPA has used to
26 communicate information on predicted occurrences of high levels of 63 and other pollutants. The
27 AQI provides both the predicted level of air quality in an area along with a set of potential
28 actions that individuals and communities can take to reduce exposure to air pollution and thus
29 reduce the risk of health effects associated with breathing ambient air pollution. There are
30 several studies, discussed in the Os ISA, that have evaluated the degree to which populations are
31 aware of the AQI and what actions individuals and communities take in response to AQI values
32 in the unhealthy range. These studies suggest that at-risk populations, such as children, older
33 adults, and asthmatics, modify their behavior in response to days with bad air quality, most
34 commonly by reducing their time spent outdoors or limiting their outdoor activity exertion level.
35 A challenge remains in how to consider existing averting behaviors within the assessment tools
36 we use and how best to use improved knowledge of participation rates, the varying types of
2-18
-------
1 actions performed particularly by potentially at-risk individuals, and the duration of these
2 averting behaviors to quantify the impact on estimated exposures and health risks.
3 2.3.3 Modeling of Exposures Associated with Simulating Just Meeting Os Standards
4 In order to address policy-relevant questions regarding changes in exposure associated
5 with potential alternative standards, the exposure assessment evaluates changes in the 63
6 concentrations, and the resulting changes in exposure, associated with simulating just meeting
7 alternative standards relative to just meeting the existing standards. The new, model-adjustment
8 methodology being implemented in this risk and exposure assessment provides for more realistic
9 responses of hourly Os concentrations to changes in the precursor emissions that lead to Os
10 formation. Characterization of exposure and changes in exposure when simulating just meeting
11 the alternative standards are discussed in greater detail in Chapter 5.
12 2.3.4 Considerations in Selecting Urban Case Study Areas for the Exposure Analysis
13 The goal of the urban area exposure analysis is to characterize the variability in exposures
14 for different locations, taking into account variability in essential factors that affect exposures.
15 Important factors identified earlier that may influence exposure include time activity patterns,
16 especially activities occurring in outdoor environments; demographics of the exposed
17 population, e.g., age and income level; and Os concentrations. In addition to these factors, the
18 selection of urban areas to include in the exposure analysis takes into consideration the location
19 of 63 epidemiological studies (for comparability with the risk assessments), the availability of
20 ambient Os data and specific exposure information (e.g., air conditioning prevalence), and the
21 desire to represent a range of geographic areas. To make the exposure analysis most useful in
22 addressing the key policy-relevant questions, urban case study areas were also chosen such that
23 most of them exceeded the existing 8-hr 03 standards and potential alternative standards during
24 the time period of interest.
25 2.4 RISK ASSESSMENT
26 Assessment of risk entails joint consideration of the exposure to a hazard, frequency of
27 adverse outcomes given exposure, and severity of resulting adverse outcomes. A risk assessment
28 for 63 requires characterization of exposures to ambient 63 for relevant populations,
29 identification of appropriate dose-response or concentration-response functions linking 03 with
30 adverse health outcomes, and characterizing risks for individuals and populations.
31 As discussed above, there are two classes of studies that have provided information to
32 inform the risk modeling: controlled human exposure studies and observational epidemiology
33 studies. The conceptual approach to risk assessment varies based on which type of study result is
34 being applied. This section briefly describes the conceptual foundation for several aspects of risk
2-19
-------
1 modeling, including the concept of attributable risk, modeling of total risk and incremental risk
2 reductions, development of risk estimates based on controlled human exposure studies, and
3 development of risk estimates based on results of observational epidemiology studies.
4 This section briefly describes the conceptual foundation for key elements of risk
5 modeling, including a discussion of the concept of attributable risk, modeling of risk for total 63
6 exposure and the distribution of risk over Os concentrations, modeling of risk reductions
7 associated with alternative standards, and key factors to consider in selecting urban case study
8 areas for the risk analysis. Characterization of ambient 63 concentrations is discussed earlier in
9 this chapter and in greater detail in Chapter 4. The specific risk models used in the urban case
10 study area risk analyses, APEX for analyses based on application of controlled human exposure
11 studies and BenMAP for analyses based on application of observational epidemiology studies,
12 are described more fully in Chapters 6 and 7, respectively. Chapter 8 provides an additional
13 national-scale assessment of mortality risk associated with recent 63 concentrations, to provide
14 context for evaluating the magnitude of health risks in the urban case study areas and to evaluate
15 the representativeness of the urban case study areas in estimating 63 risks.
16 2.4.1 Attributable Risk
17 This risk and exposure assessment relies on the concept of attributable risk in evaluating
18 both total risk and incremental changes in risk associated with just meeting existing and potential
19 alternative 63 standards. Attributable risk is defined as the difference in incidence of an adverse
20 effect between an exposed and unexposed population for a specific stressor. Attributable risk is
21 an important concept when addressing risks that are associated with multiple causes, such as
22 mortality and respiratory hospital admissions.
23 Estimates of attributable risk require either an exposure-response (E-R) function (for
24 analyses based on results of controlled human exposure studies) or a concentration-response (C-
25 R) function (for analyses based on results of epidemiology studies).
26 E-R functions require estimates of exposure, in this case supplied by the APEX modeling
27 described above. In the case of the lung function endpoint evaluated in this risk analysis, the E-R
28 function also requires information on age and exertion levels to predict the impact of O^
29 exposure on decrements in lung function. E-R functions may provide estimates of the incidence
30 of an endpoint or the probability of exceeding benchmark decrement levels.
31 C-R functions derived from relative risk estimates reported in the epidemiological
32 literature generally require estimates of ambient 63 concentrations, baseline incidence rates, and
33 estimates of exposed populations. Ambient OT, concentrations should generally be constructed to
34 match the spatial and temporal averaging used in the underlying epidemiology study; e.g., a
2-20
-------
1 study may have used a spatial average over a metropolitan statistical area of the 8-hour daily
2 maximum.
3 As with exposure, attributable risk is highly dependent on the ambient Os concentrations
4 in an urban area. Given that these concentrations are variable from year to year, it is important to
5 model multiple years representing the range of variability of Os concentrations to provide a
6 better characterization of risk. In addition, other important sources of variability and uncertainty
7 affecting the risk estimates should be characterized, including uncertainty and variability in the
8 C-R and E-R functions, Os concentrations and 63 exposure, and population inputs. This can be
9 accomplished in part by modeling risk for multiple urban areas selected to represent variability in
10 these underlying risk drivers.
11 2.4.2 Modeling of Risk for Total Exposure to O3
12 As discussed earlier in this chapter, ambient Os is contributed to by emissions from a
13 variety of sources, including natural, U.S. anthropogenic, and non-U.S. anthropogenic sources.
14 Once in the atmosphere, Os molecules created from these different sources of emissions are not
15 distinguishable. Individuals and populations are exposed to total 03 from all sources, and risks
16 associated with 63 exposure are due to total 63 exposure and do not vary for 63 exposure
17 associated with any specific source. Given the absence of a detectable threshold in the available
18 C-R functions, total risk attributable to 63 will thus be the risk associated with total exposure to
19 63, with no threshold or cutpoint applied. To address certain policy-related questions, it is
20 possible to approximately attribute risk to specific sources through the use of air quality
21 modeling techniques, and this is explored in the Policy Assessment. However, these techniques
22 are based on applying model results to total Os risk, rather than on directly modeling risk
23 attributable to specific sources.
24 As discussed earlier in this chapter, a critical policy-relevant risk question is the 63
25 attributable risk remaining after just meeting the existing Os standards. This risk includes risks
26 associated with 63 from all sources after we have simulated just meeting the existing daily 8-
27 hour maximum standard level of 75 ppb. The estimates of total risk remaining after meeting the
28 existing standard form the reference values for evaluating reductions in risk associated with just
29 meeting alternative levels of the standard.
30 In addition to providing risk estimates for urban case study areas, it is also useful to
31 evaluate Os risks across the entire U.S., both to better understand the total magnitude of the
32 health burden associated with Os and to evaluate the representativeness of selected urban case
33 study areas in characterizing the range and variability in risks across the U.S. The national-scale
34 risk assessment presented in Chapter 8 is focused on estimating risk associated with recent Os
35 concentrations, rather than on risk after just meeting existing or alternative standards. This is the
2-21
-------
1 appropriate focus for the national analysis, because the techniques used to simulate just meeting
2 existing and alternative standards in urban case study areas are less certain in a national context
3 due to concerns about interdependence between air quality responses in different urban areas;
4 e.g., just meeting a standard in one urban area would likely have impacts on O^ air quality in
5 surrounding urban areas. It is beyond the scope of this REA to attempt to simulate control
6 strategies that would result in national attainment of existing or alternative primary health
7 standards.
8 2.4.3 Distributions of Risk Across Os concentrations
9 Total Os risk for the Os season is calculated by summing daily risks across all days in the
10 Oj, season. Because of the high degree of variability in daily 63 concentrations across an 63
11 season, total 63 risk will include risks calculated for some days with high 63 concentrations as
12 well as for some days with very low O^ concentrations. Therefore it is appropriate to provide the
13 distribution of total risk over the range of daily 63 concentrations to allow for an understanding
14 of how Os concentrations on different days are contributing to the estimates of total risk. In
15 addition, as noted in the ISA and discussed above, because of the relatively lower density of data
16 on days with low concentrations of 63, there is decreased confidence in the shape of the C-R
17 function at lower Os concentrations, and therefore lower confidence in risk estimates for days
18 with lower 63 concentrations, especially in the range below 20 ppb. As a result, it is appropriate
19 to provide the distribution of total risk over the range of daily 63 concentrations to allow for
20 better characterization of confidence in the estimates of total risk.
21 2.5 MODELING OF RISKS ASSOCIATED WITH SIMULATING JUST MEETING O3
22 STANDARDS
23 In order to address policy-relevant questions regarding changes in risk associated with
24 potential alternative standards, the risk assessment evaluates changes in the distribution of 63
25 concentrations, and the resulting changes in risk, associated with simulating just meeting
26 alternative standards relative to just meeting the existing standards. The new, model-adjustment
27 methodology being implemented in this risk and exposure assessment provides for more realistic
28 responses of hourly O^ concentrations to changes in the precursor emissions that lead to O^
29 formation. As noted earlier there are multiple combinations of reductions in precursor emissions
30 that can result in just meeting alternative standards. As a result, there is variability in the
31 potential changes in the distribution of 63 concentrations and risk that would result from just
32 meeting existing and alternative standards. Characterization of this variability, as well as
33 uncertainties in the simulation of just meeting the standards, will be included in Chapters 6 and
34 7.
2-22
-------
1 2.6 CONSIDERATIONS IN SELECTING URBAN CASE STUDY AREAS FOR THE
2 RISK ANALYSIS
3 The goal of the urban area risk analysis is to characterize the magnitude of risk and the
4 impact on risk of meeting existing and potential alternative standards. The selection of specific
5 urban case study areas is based on a set of factors reflecting both variability in factors that affect
6 risk and availability of high quality input data, to provide risk estimates that have higher overall
7 confidence. Important factors identified earlier that may influence risk include 63 concentrations,
8 demographics, exposure factors, and magnitude of the effect estimate in the C-R function. In
9 addition to consideration of variability in these factors, urban areas are preferentially selected if
10 they have 63 concentrations that are above the existing standards and potential alternative
11 standards, if they have suitable epidemiological studies to provide C-R functions for mortality or
12 morbidity, if they have adequate monitoring data available to characterize population exposures,
13 and if they have appropriate baseline health incidence data available.
14 2.7 RISK CHARACTERIZATION
15 Risk characterization is the process of communicating the results of risk (and exposure)
16 modeling in metrics that have meaning to decision makers. In the specific context of this review,
17 this translates into providing metrics that are most useful in the Policy Assessment to assess the
18 adequacy of the existing 63 standards in protecting public health with an adequate margin of
19 safety and to evaluate the additional protection provided by potential alternative standards.
20 Risk characterization requires careful translation of very complex outputs of exposure
21 and risk models into simpler metrics, for example, translating hourly 63 exposures in various
22 microenvironments into estimates of population exposures above alternative exposure
23 benchmarks. Risk characterization also requires the condensation of a large number of analytical
24 steps and results to (a) summarize the results of the risk analysis, usually taking detailed results
25 and condensing them into a more aggregate interpretation while still providing information about
26 heterogeneity across space and time; (b) communicate the sensitivity of results to different
27 modeling assumptions; and (c) characterize the qualitative and quantitative uncertainty in results.
28 As described more fully in Chapter 5 and in the Policy Assessment, EPA has selected,
29 based on providing a reasonable measure of exposures of concern for at-risk populations and
30 lifestages, aggregate exposure metrics including the number and percent of certain highly
31 vulnerable populations exposed to levels of 63 above exposure levels that have been identified in
32 the scientific literature as associated with adverse respiratory responses. As noted in section
33 2.3.1, these benchmark exposure levels are 0.060 ppm, 0.070 ppm, and 0.080 ppm. Highly
34 vulnerable populations include active children, older adults, and outdoor workers.
2-23
-------
1 As described more fully in Chapters 6 and 7 and in the Policy Assessment, EPA has
2 selected, based on providing characterization of risks to the public including at-risk populations
3 and lifestages, aggregate risk metrics including the number and percent of vulnerable
4 populations experiencing adverse respiratory responses based on application of results of
5 controlled human exposure studies and the attributable incidence and percent of baseline
6 incidence of mortality and morbidity endpoints based on application of results of epidemiology
7 studies.
8 For all three types of metrics (exposure, risk based on controlled human exposure studies,
9 and risk based on epidemiology studies) and for the purpose of evaluating the adequacy of the
10 existing standards, the focus is on the exposure and risk remaining upon just meeting the existing
11 standards. For the purpose of evaluating alternative standards, the focus in on the changes in
12 exposure and risk after simulating just meeting the alternative standards, compared to exposures
13 and risk after simulating just meeting the existing standards.
14 As detailed in Chapter 3, quantitative sensitivity analyses are provided to evaluate the
15 impacts of critical inputs to the exposure and risk modeling. Limited quantitative uncertainty
16 analyses are also included, along with a comprehensive qualitative uncertainty assessment. The
17 overall treatment of uncertainty is guided by the WHO guidelines for uncertainty assessment
18 (World Health Organization, 2008). These guidelines recommend a tiered approach in which
19 progressively more sophisticated methods are used to evaluate and characterize sources of
20 uncertainty depending on the overall complexity of the risk assessment.
21 In order to inform considerations of overall confidence in the risk estimates derived from
22 application of C-R functions derived from the epidemiological literature, we provide the
23 distributions of total risk across the entire range of daily 8-hour maximum Os concentrations. In
24 addition, we provide an assessment of the representativeness of the urban areas selected for the
25 risk and exposure analysis in characterizing the overall distribution of risk across the U.S. This
26 assessment evaluates how well the selected urban areas capture important characteristics that are
27 associated with risk, including demographics, air quality levels, and factors affecting exposure
28 such as air conditioning prevalence.
2-24
-------
1 2.8 REFERENCES
2 Frey, C. and J. Samet. 2012. "CASAC Review of the EPA's Policy Assessment for the Review
3 of the Ozone National Ambient Air Quality Standards (First External Review Draft -
4 August 2012)." U.S. Environmental Protection Agency Science Advisory Board, EPA-
5 CASAC-13-003.
6 U.S. Environmental Protection Agency. 2012a. Total Risk Integrated Methodology (TRIM) - Air
7 Pollutants Exposure Model Documentation (TREVI.Expo / APEX, Version 4.4) Volume I:
8 User's Guide. Research Triangle Park, NC: EPA Office of Air Quality Planning and
9 Standards. (EPA document number EPA-452/B-12-001a).
10 .
11 U.S. EPA. 2012 b. "Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
12 Documentation (TREVI.Expo / APEX, Version 4.4) Volume II: Technical Support
13 Document." Research Triangle Park, NC: Office of Air Quality Planning and Standards.
14 (EPA document number EPA-452/B-12-001b).
15 < http://www.epa.gov/ttn/fera/human_apex.html>.
16 U.S. EPA. 2013. Integrated Science Assessment of Ozone and Related Photochemical Oxidants
17 (Final Report). Washington, DC: EPA Office of Air and Radiation. (EPA document
18 number EPA/600/R-10/076F).
19 U.S. EPA. 2013. "Environmental Benefits Mapping Analysis Program Community Edition
20 (BenMAP-CE vl.O)," posted on December 03, 2013,
21 .
22 World Health Organization. 2008. "Part 1: Guidance Document on Characterizing and
23 Communicating Uncertainty in Exposure Assessment, Harmonization Project Document
24 No. 6." Published under joint sponsorship of the World Health Organization, the
25 International Labor Organization and the United Nations Environment Programme. WHO
26 Press, World Health Organization, 20 Avenue Appia, 1211 Geneva 27, Switzerland (tel.:
27 +41 22 791 2476).
28
2-25
-------
1 3 SCOPE
2 This chapter provides an overview of the scope and key design elements of this
3 quantitative exposure and health risk assessment. The design of this assessment began with a
4 review of the exposure and risk assessments completed during the last 63 NAAQS review (U.S.
5 EPA, 2007a,b), with an emphasis on considering key limitations and sources of uncertainty
6 recognized in that analysis.
7 As an initial step in the current Os NAAQS review in October 2009, EPA invited outside
8 experts, representing a broad range of expertise (e.g., epidemiology, human and animal
9 toxicology, statistics, risk/exposure analysis, atmospheric science), to participate in a workshop
10 with EPA staff to help inform EPA's plan for the review. The participants discussed key policy-
11 relevant issues that would frame the review and the most relevant new science that would be
12 available to inform our understanding of these issues. One workshop session focused on planning
13 for quantitative risk and exposure assessments, taking into consideration what new research
14 and/or improved methodologies would be available to inform the design of quantitative exposure
15 and health risk assessment. Based in part on the workshop discussions, EPA developed a draft
16 IRP (U.S. EPA, 2009) outlining the schedule, process, and key policy-relevant questions that
17 would frame this review. On November 13, 2009, EPA held a consultation with CASAC on the
18 draft IRP (74 FR 54562, October 22, 2009), which included opportunity for public comment.
19 The final IRP incorporated comments from CASAC (Samet, 2009) and the public on the draft
20 plan, as well as input from senior Agency managers. The final IRP included initial plans for
21 quantitative risk and exposure assessments for both human health and welfare (U.S. EPA, 201 la,
22 chapters 5 and 6).
23 As a next step in the design of these quantitative assessments, OAQPS staff developed
24 more detailed planning documents, the OT, National Ambient Air Quality Standards: Scope and
25 Methods Plan for Health Risk and Exposure Assessment (Health Scope and Methods Plan, U. S.
26 EPA, 201 Ib) and the Os National Ambient Air Quality Standards: Scope and Methods Plan for
27 Welfare Risk and Exposure Assessment (Welfare Scope and Methods Plan, U.S. EPA, 201 Ic).
28 These Scope and Methods Plans was the subject of a consultation with CASAC on May 19-20,
29 2011 (76 FR 23809, April 28, 2011). Based on consideration of CASAC (Samet, 2011) and
30 public comments on the Scope and Methods Plans, and information in the second draft ISA, we
31 modified the scope and design of the quantitative risk assessment and provided a memo with
32 updates to information presented in the Scope and Methods Plans (Wegman, 2012). The Scope
33 and Methods Plans together with the update memo provide the basis for the discussion of the
34 scope of this exposure and risk assessment provided in this chapter. This chapter also reflects
3-1
-------
1 comments received from CAS AC based on their review of the first draft Risk and Exposure
2 Assessment on September 11-12, 2012 (Frey and Samet, 2012).
3 In presenting the scope and key design elements of the current risk assessment, this
4 chapter first provides a brief overview of the quantitative exposure and risk assessment
5 completed for the previous 63 NAAQS review in section 3.1, including key limitations and
6 uncertainties associated with that analysis. The remaining sections describe the current exposure
7 and risk assessment, following the general conceptual framework described in Chapter 2. Section
8 3.2 provides a summary of the design of the urban-scale exposure assessment. Section 3.3
9 provides a summary of the design of the urban-scale risk assessment based on application of
10 results of human clinical studies. Section 3.4 provides a summary of the design of the urban-
11 scale risk assessment based on application of results of epidemiology studies. Section 3.5
12 provides a summary of the design of the national-scale risk burden assessment and
13 representativeness analysis.
14 3.1 OVERVIEW OF EXPOSURE AND RISK ASSESSMENTS FROM LAST REVIEW
15 The exposure and health risk assessment conducted in the review, completed in March
16 2008, developed exposure and health risk estimates for 12 urban areas across the U.S. which
17 were chosen based on the location of Os epidemiological studies and availability of ambient Os
18 data and to represent a range of geographic areas, population demographics, and 63 climatology.
19 That analysis was in part based upon the exposure and health risk assessments included in the
20 review completed in 1997.l The exposure and risk assessment incorporated air quality data (i.e.,
21 2002 through 2004), and provided annual or 63 season-specific exposure and risk estimates for
22 these recent years of air quality and for air quality scenarios simulating just meeting the existing
23 8-hour O^ standard and several alternative 8-hour 03 standards.
24 3.1.1 Overview of exposure assessment from last review
25 Exposure estimates were used as an input to the risk assessment for lung function
26 responses (a health endpoint for which exposure-response functions were available from
27 controlled human exposure studies). Exposure estimates were developed for the general
28 population and population groups including school-age children with asthma as well as all
29 school-age children. The exposure estimates also provided information on population exposures
1 In the 1994-1997 O3 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 as
is air quality and for just meeting the existing 1-hour standard and several alternative 8-hour standards.
Several reports that describe these analyses can be found at:
http://www.epa.gov/ttn/maqs/standards/O3/s_O3_pr.html.
O O
3-2
-------
1 exceeding potential health effect benchmark levels that were identified based on the observed
2 occurrence of health endpoints not explicitly modeled in the health risk assessment (e.g., lung
3 inflammation, increased airway responsiveness, and decreased resistance to infection) associated
4 with 6-8 hour exposures to O^ in controlled human exposure studies.
5 The exposure analysis took into account several important factors including the
6 magnitude and duration of exposures, frequency of repeated high exposures, and breathing rate
7 of individuals at the time of exposure. Estimates were developed for several indicators of
8 exposure to various levels of 63 air quality, including counts of people exposed one or more
9 times to a given O^ concentration while at a specified breathing rate and counts of person-
10 occurrences (which accumulate occurrences of specific exposure conditions over all people in
11 the population groups of interest over an Os season).
12 As discussed in the 2007 Staff Paper (U.S. EPA, 2007c) and in Section II a of the O3
13 Final Rule (73 FR 16440 to 16442, March 27, 2008), the most important uncertainties affecting
14 the exposure estimates were related to modeling human activity patterns over an Os season,
15 modeling of variations in ambient concentrations near roadways, and modeling of air exchange
16 rates that affect the amount of 63 that penetrates indoors. Another important uncertainty,
17 discussed in more detail in the Staff Paper (U.S. EPA, 2007c, section 4.3.4.7), was the
18 uncertainty in energy expenditure values which directly affected the modeled breathing rates.
19 These were important since they were used to classify exposures occurring when children were
20 engaged in moderate or greater exertion. Health effects observed in the controlled human
21 exposure studies generally occurred under these exertion levels for 6 to 8-hour exposures to 63
22 concentrations at or near 0.08 ppm. Reports that describe these analyses (U.S. EPA, 2007a, c;
23 Langstaff, 2007) can be found at: http://www.epa.gov/ttn/naaqs/standards/O3/s_O3_index.html.
24 3.1.2 Overview of risk assessment from last review
25 The human health risk assessment presented in the review completed in March 2008 was
26 designed to estimate population risks in a number of urban areas across the U.S., consistent with
27 the scope of the exposure analysis described above (U.S. EPA, 2007b, c). The risk assessment
28 included risk estimates based on both controlled human exposure studies and epidemiological
29 and field studies. Ch-related risk estimates for lung function decrements were generated using
30 probabilistic exposure-response relationships based on data from controlled human exposure
31 studies, together with probabilistic exposure estimates from the exposure analysis. For several
32 other health endpoints, Os-related risk estimates were generated using concentration-response
33 relationships reported in epidemiological or field studies, together with ambient air quality
34 concentrations, baseline health incidence rates, and population data for the various locations
35 included in the assessment. Health endpoints included in the assessment based on
3-3
-------
1 epidemiological or field studies included hospital admissions for respiratory illness in four urban
2 areas, premature mortality in 12 urban areas, and respiratory symptoms in asthmatic children in 1
3 urban area.
4 In the health risk assessment conducted in the previous review, EPA recognized that there
5 were many sources of uncertainty and variability in the inputs to the assessment and that there
6 was significant uncertainty in the resulting risk estimates. The statistical uncertainty surrounding
7 the estimated 63 coefficients in epidemiology-based concentration-response functions as well as
8 the shape of the exposure-response relationship chosen for the lung function risk assessment
9 were addressed quantitatively. Additional uncertainties were addressed through sensitivity
10 analyses and/or qualitatively. The risk assessment conducted for the previous 63 NAAQS review
11 incorporated some of the variability in key inputs to the assessment by using location-specific
12 inputs (e.g., location-specific concentration-response functions, baseline incidence rates and
13 population data, and air quality data for epidemiological-based endpoints, location specific air
14 quality data and exposure estimates for the lung function risk assessment). In that review, several
15 urban areas were included in the health risk assessment to provide some sense of the variability
16 in the risk estimates across the U.S.
17 Key observations and insights from the Os risk assessment, in addition to important
18 caveats and limitations, were addressed in Section II.B of the Final Rule notice (73 FR 16440 to
19 14 16443, March 27, 2008). In general, estimated risk reductions associated with going from
20 then-current O^ levels to just meeting the then-existing and alternative 8-hour standards showed
21 patterns of decreasing estimated risk associated with just meeting the lower alternative 8-hour
22 standards considered. Furthermore, the estimated percentage reductions in risk were strongly
23 influenced by the baseline air quality year used in the analysis, which was due to significant
24 year-to-year variability in 63 concentrations. There was also noticeable city-to-city variability in
25 the estimated Os-related incidence of morbidity and mortality across the 12 urban areas.
26 Uncertainties associated with estimated policy-relevant background (PRB) concentrations2 were
27 also addressed and revealed differential impacts on the risk estimates depending on the health
28 effect considered as well as the location. EPA also acknowledged that at the time of the previous
29 review there were considerable uncertainties surrounding estimates of 63 C-R coefficients and
30 the shape of concentration-response relationships and whether or not a population threshold or
31 non-linear relationship exists within the range of concentrations examined in the epidemiological
32 studies.
Policy-relevant background (PRB) O3 has been defined in previous reviews as the distribution of O3 concentrations
that would be observed in the U.S. in the absence of anthropogenic (man-made) emissions of O3 precursor
emissions (e.g., VOC, CO, NOx) in the U.S., Canada, and Mexico.
3-4
-------
1 3.2 PLAN FOR THE CURRENT EXPOSURE AND RISK ASSESSMENTS
2 The Scope and Methods Plan, including updates (U.S. EPA, 201 Ib; Wegman, 2012),
3 outlined a planned approach for conducting the current quantitative 63 exposure and risk
4 assessments, including broad design issues as well as more detailed aspects of the analyses. A
5 critical step in designing the quantitative risk and exposure assessments is to clearly identify the
6 goals for the analysis based on the policy-relevant questions identified in Chapter 2. We have
7 identified the following goals for the urban area exposure and risk assessments: (1) to provide
8 estimates of the percent of people in the general population and in sensitive populations with 63
9 exposures above health-based benchmark levels; (2) to provide estimates of the percentage of the
10 general population and in sensitive populations with impaired lung function (defined based on
11 decrements in FEVi) resulting from exposures to 63; (3) to provide estimates of the potential
12 magnitude of premature mortality associated with both short-term and long-term 03 exposures,
13 and selected morbidity health effects associated with short-term 63 exposures; (4) to evaluate the
14 influence of various inputs and assumptions on risk estimates to the extent possible given
15 available methods and data; (5) to gain insights into the spatial and temporal distribution of risks
16 and patterns of risk reduction and uncertainties in those risk estimates. For the exposure and risk
17 analyses, we will estimate exposures and risks for recent ambient levels of Os and for Os
18 concentrations after simulating just meeting the existing 63 standard and potential alternative
19 standards.
20 With regard to selecting alternative levels for the 8-hour 03 standards for evaluation in
21 the quantitative risk assessment, we base the range of levels on the evaluations of the evidence
22 provided in the first draft PA, which received support from the CASAC in their advisory letter
23 on the first draft PA. The first draft PA recommended evaluation of 8-hour maximum
24 concentrations in the range of 60 to 70 ppb, with possible consideration of levels somewhat
25 below 60 ppb. The upper end of this range is supported by the clear evidence from both clinical
26 and epidemiological studies of effects at exposures of 70 ppb reported in the ISA and
27 summarized in the first draft PA. The lower end of this range is based on considerations of
28 evidence from clinical studies that have shown lung function decrements in healthy adult
29 populations at 60 ppb 63 exposures, and that 10 percent of healthy adults exposed to 60 ppb 63
30 experienced lung function decrements that could be adverse to asthmatics. The evidence showing
31 effects in healthy adults at exposures of 60 ppb supports the consideration of risks to sensitive
32 populations at exposure levels below 60 ppb, although specific exposure levels below 60 ppb at
33 which risks may be occurring are not supported by the evidence. An important distinction is that
34 the evidence from controlled human exposure studies is based on exposures, while the standard
5-5
-------
1 addresses ambient concentrations. Typically, exposures are lower than ambient concentrations
2 because people spend a large fraction of their time indoors where 63 concentrations are lower.3
3 Because of the year-to-year variability in Os concentrations that results from temporal
4 variability in meteorology and emissions that drive O^ formation, the exposure and risk
5 assessments evaluate scenarios for meeting the existing and alternative standards based on
6 multiple years of Os data. Os concentrations from 2006-2010 are used in estimating exposure and
7 risk. This range of years captures a high degree of variability in meteorological conditions, as
8 well as reflecting years with higher and lower emissions of 63 precursors.
9 In order to provide greater confidence in the exposure and risk estimates, this REA uses
10 an urban case study approach for assessing both exposure and risk. This approach provides
11 greater confidence in estimates by allowing us to make use of air quality data, population
12 information, health data, and epidemiology results that are well matched, and it does not require
13 extrapolation of results to locations without these data. In addition, the urban case study
14 approach allows us to simulate just meeting existing and alternative Os standards for each urban
15 area, which is not currently feasible for health risk assessment at the national scale.4 Specific
16 selection criteria for case study urban areas included in the exposure and risk assessments are
17 described in the following sections. In order to gain an understanding of how well the urban case
18 study areas represent risks at a national level and to provide context for the urban case study
19 results, we also include two national level analyses, 1) estimation of the national mortality
20 burden associated with recent ambient Os and 2) characterization of how well the risk estimates
21 for the set of urban areas modeled reflect the national distribution of mortality risk.
22 Throughout the exposure and risk analyses, we recognize that there are many sources of
23 variability and uncertainty. Each analysis considers carefully the potential sources and
24 significance of variability and uncertainties and, where data are available, provides quantitative
25 assessment of variability and uncertainties, either through probabilistic analyses or through
26 sensitivity or scenario analyses. In general the analyses follow the WHO guidelines for
27 uncertainty assessment (World Health Organization, 2008), which recommend a tiered approach
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 REA, some areas would see O3 design values decreased below the targeted
standard level due to O3 transport between locations. We were not able to devise an approach that would
just meet the standard in every location simultaneously. Using the urban case study approach, we can,
acknowledging the counterfactual nature of the analysis, assume independence of attainment for each urban
case study area, which allows us to simulate just meeting the standards in each urban case study area.
-------
1 in which progressively more sophisticated methods can be used to evaluate and characterize
2 sources of uncertainty depending on the overall complexity, end use of the assessment, and
3 resources and data available to conduct particular uncertainty characterizations.
4 The planned approaches for conducting the exposure and risk analyses are briefly
5 summarized below. We begin with a general discussion of how uncertainty and variability are
6 addressed in the different elements of the exposure and risk assessment. This is followed by a
7 discussion of the air quality data that will be used in both the exposure and risk assessments and
8 then discussions of each component of the exposure and risk assessments.
9 3.3 CHARACTERIZATION OF UNCERTAINTY AND VARIABILITY IN THE
10 CONTEXT OF THE O3 EXPOSURE AND RISK ASSESSMENT
11 An important component of this population exposure and health risk assessment is the
12 characterization of both uncertainty and variability. Variability refers to the heterogeneity of a
13 variable of interest within a population or across different populations. For example, populations
14 in different regions of the country may have different behavior and activity patterns (e.g., air
15 conditioning use and time spent indoors) that affect their exposure to ambient 63 and thus the
16 population health response. The composition of populations in different regions of the country
17 may vary in ways that can affect the population response to exposure to 63 - e.g., two
18 populations exposed to the same levels of 63 might respond differently if one population is older
19 than the other. Variability is inherent and cannot be reduced through further research.
20 Refinements in the design of a population risk assessment are often focused on more completely
21 characterizing variability in key factors affecting population risk - e.g., factors affecting
22 population exposure or response - in order to produce risk estimates whose distribution
23 adequately characterizes the distribution in the underlying population(s).
24 Uncertainty refers to the lack of knowledge regarding the actual values of inputs to an
25 analysis. Models are typically used in analyses, and there is uncertainty about the true values of
26 the parameters of the model (parameter uncertainty) - e.g., the value of the coefficient for 63 in a
27 C-R function. There is also uncertainty about the extent to which the model is an accurate
28 representation of the underlying physical systems or relationships being modeled (model
29 uncertainty) - e.g., the shapes of C-R functions. In addition, there may be some uncertainty
30 surrounding other inputs to an analysis due to possible measurement error—e.g., the values of
31 daily 63 concentrations in a risk assessment location or the value of the baseline incidence rate
32 for a health effect in a population.5
5 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
3-7
-------
1 In any risk assessment, uncertainty is, ideally, reduced to the maximum extent possible
2 through improved measurement of key variables and ongoing model refinement. However,
3 significant uncertainty often remains, and emphasis is then placed on characterizing the nature of
4 that uncertainty and its impact on risk estimates. The characterization of uncertainty can be both
5 qualitative and, if a sufficient knowledge base is available, quantitative.
6 The characterization of uncertainty associated with risk assessment is ideally addressed in
7 the regulatory context using a tiered approach in which progressively more sophisticated
8 methods are used to evaluate and characterize sources of uncertainty depending on the overall
9 complexity and intended use of the risk assessment (WHO, 2008). Guidance documents
10 developed by EPA for assessing air toxics-related risk and Superfund Site risks as well as recent
11 guidance from the World Health Organization specify multitier approaches for addressing
12 uncertainty.
13 Following the approach used for previous NAAQS risk and exposure assessments (U.S.
14 EPA, 2008c, 2009b, 2010a, b), for the Os risk assessment, we are using a tiered framework
15 developed by WHO to guide the characterization of uncertainty. The WHO guidance presents a
16 four-tiered approach, where the decision to proceed to the next tier is based on the outcome of
17 the previous tier's assessment. The four tiers described in the WHO guidance include:
18 Tier 0: recommended for routine screening assessments, uses default uncertainty factors
19 (rather than developing site-specific uncertainty characterizations);
20 Tier 1: the lowest level of site-specific uncertainty characterization, involves qualitative
21 characterization of sources of uncertainty (e.g., a qualitative assessment of the general magnitude
22 and direction of the effect on risk results);
23 Tier 2: site-specific deterministic quantitative analysis involving sensitivity analysis,
24 interval-based assessment, and possibly probability bounded (high-and low-end) assessment; and
25 Tier 3: uses probabilistic methods to characterize the effects on risk estimates of sources
26 of uncertainty, individually and combined.
27 With this four-tiered approach, the WHO framework provides a means for systematically
28 linking the characterization of uncertainty to the sophistication of the underlying risk assessment.
29 Ultimately, the decision as to which tier of uncertainty characterization to include in a risk
with the effort to accurately characterize variability in key model inputs actually reflecting an effort to
reduce uncertainty.
-------
1 assessment will depend both on the overall sophistication of the risk assessment and the
2 availability of information for characterizing the various sources of uncertainty.
3 This risk and exposure assessment for the Os NAAQS review is relatively complex,
4 possibly warranting consideration of a full probabilistic (WHO Tier 3) uncertainty analysis. For
5 the exposure assessment, we include probabilistic representations of important sources of
6 variability; however, due to lack of information regarding reasonable alternative parameter
7 settings for model input variable distributions, we are not able to include a complete probabilistic
8 analysis incorporating both variability and uncertainty. Instead, we provide sensitivity analyses
9 to explore the impact of specific model assumptions, and we include a comprehensive qualitative
10 discussion of uncertainty regarding the model inputs and outputs.
11 While a full probabilistic uncertainty analysis is not undertaken for the epidemiology-
12 based risk assessment due to limits in available information on distributions of model inputs, we
13 provide a limited assessment using the confidence intervals associated with effects estimates
14 (obtained from epidemiological studies) to incorporate statistical uncertainty associated with
15 sample size considerations in the presentation of risk estimates. Technically, this type of
16 probabilistic simulation represents a Tier 3 uncertainty analysis, although as noted here, it will be
17 limited and only address uncertainty related to the fit of the C-R functions. Incorporation of
18 additional sources of uncertainty related to key elements of C-R functions (e.g., competing lag
19 structures, alternative functional forms, etc.) into a full probabilistic WHO Tier 3 analysis would
20 require that probabilities be assigned to each competing specification of a given model element
21 (with each probability reflecting a subjective assessment of the probability that the given
22 specification is the correct description of reality). However, for most model elements there is
23 insufficient information on which to base these probabilities. One approach that has been taken
24 in such cases is expert elicitation; however, this approach is resource- and time-intensive, and,
25 consequently, it is not feasible to use this technique in support of this O^ risk assessment.
26 For most elements of the quantitative risk assessments, rather than conducting a full
27 probabilistic uncertainty analysis, we include a qualitative discussion of the potential impact of
28 uncertainty on risk results (WHO Tier 2). For some critical elements of the epidemiology-based
29 risk assessment, e.g., the effect-estimate in the C-R function, we include sensitivity analyses to
30 explore the potential impact of our assumptions. This falls under the WHO Tier 2 classification,
31 although we are not able to assign probabilities to the sensitivity analyses. For these sensitivity
32 analyses, we will include only those alternative specifications for input parameters or modeling
33 approaches that are deemed to have scientific support in the literature (and so represent
34 alternative reasonable input parameter values or modeling options). This means that the array of
35 risk estimates presented in this assessment is expected to represent reasonable risk estimates that
5-9
-------
1 can be used to provide some information regarding the potential impacts of uncertainty in the
2 model elements.
3 3.4 AIR QUALITY CHARACTERIZATION
4 Figure 3-1 diagrams the basic information used in developing the air quality inputs for
5 the REA. Air quality inputs to the urban area exposure and risk assessments include (1) recent air
6 quality data developed from Os ambient monitors in each selected urban study area and (2)
7 simulated air quality that reflects changes in the distribution of O^ air quality estimated to occur
8 when the urban area just meets the existing or alternative 63 standards under consideration. In
9 addition, Os air quality surfaces for recent years covering the entire continental U.S. were
10 generated for use in the national-scale assessment. Details of the air quality data used in the REA
11 are discussed in Chapter 4.
12
03 Precursor Emissions
13
14
15
16
17
18
19
20
21
22
23
Ozone Air Quality Data
National Ambient
O3 Spatial Fields:
recent conditions
Model-based O3 Sensitivities
Os Metrics in Urban Case
Study Areas: recent conditions
and after just meeting
existing and alternative standards
Figure 3-1 Conceptual Diagram for Air Quality Characterization in the Health REA
The urban case study area exposure and risk analyses are based on five recent years of air
quality data, 2006-2010. We are including 5 years to reflect the considerable variability in
meteorological conditions and the variation in O^ precursor emissions that have occurred in
recent years. The analyses focus on the 63 season, 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 case study areas are described in more detail in Chapter 4.
In developing the Oj air quality surfaces for the national-scale analysis, a combination of
monitoring data and modeled 03 concentrations are used to provide greater coverage across the
3-10
-------
1 U.S. The procedure for fusing O^ monitor data with modeling results is described further in
2 Chapter 4.
3 Several Os metrics are generated for use in the urban area exposure and risk analyses.
4 The exposure analyses use hourly Os concentrations, while the risk analyses use several different
5 averaging times. The specific metrics used in each analysis are discussed further in following
6 chapters. For the exposure analysis, hourly Os concentrations are interpolated to census tracts
7 using Voronoi neighbor averaging (VNA), a distance weighted interpolation method (Gold,
8 1997; Chen et al., 2004). For the epidemiology-based risk analysis, we developed a composite of
9 all monitors in the urban area for application with the epidemiology studies. We also evaluated
10 several different definitions of the spatial boundaries of the urban areas that determined the
11 monitors included in the spatial average. Some of the epidemiological studies specify a relatively
12 narrow set of counties within an urban area, while others use a broader definition, such as all
13 counties in a core based statistical area (CBSA) as defined by the Census Bureau. For those
14 epidemiological studies that used a relatively narrow set of counties, most were based on
15 counties in the center of the urban area. In most of these areas, the non-attaining Oj, monitors are
16 not located in the center of the urban area, but instead in the surrounding areas, reflecting the
17 transport and atmospheric chemistry governing O^ formation. As a result, using a monitor set
18 that exactly reflects the specific counties used in the epidemiology studies can exclude counties
19 in an urban area that would realize the most risk reduction resulting from just meeting the Os
20 standard. To better represent the changes in risk that could be experienced in the urban areas, the
21 core risk estimates for all endpoints will be based on the CBSA definition. Sensitivity analyses
22 are included to evaluate the effect of using only the counties in each urban area that specifically
23 match the county set used in the epidemiology studies.
24 Simulation of just meeting the existing and alternative 63 standards is accomplished by
25 adjusting hourly O^ concentrations measured over the O^ season using a model-based adjustment
26 methodology that estimates 63 sensitivities to precursor emissions changes.6 These sensitivities,
27 which estimate the response of Os concentrations to reductions in anthropogenic NOx and VOC
28 emissions, are developed using the Higher-order Decoupled Direct Method (FtDDM) capabilities
29 in the Community Multi-scale Air Quality (CMAQ) model. This modeling approach incorporates
30 all known emissions, including sources of natural and anthropogenic emissions in and outside of
31 the U.S. By using the model-based adjustment methodology we are able to more realistically
32 simulate the temporal and spatial patterns of 63 response to precursor emissions. We chose to
6 In the first draft of this REA, we used a statistical quadratic rollback approach to simulate just meeting the existing
O3 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
-------
1 simulate just meeting the existing and alternative standards in the urban cast study areas by
2 decreasing U.S. anthropogenic emissions of NOx and VOC throughout the U.S using equal
3 proportional decreases in emissions throughout the U.S., in order to avoid any suggestion that we
4 are approximating a specific emissions control strategy that a state or urban area might choose to
5 meet a standard. More details on the HDDM-adjustment approach are presented in Chapter 4 of
6 this REA and in Simon et al. (2013).
7 In the previous review, background Os (referred to in that review as policy relevant
8 background, or PRB) was incorporated into the REA by calculating risk only in excess of PRB.
9 CASAC members recommended that EPA move away from using PRB in calculating risks
10 (Henderson, 2007). In addition, comments received from CASAC, based on their review of the
11 first draft Risk and Exposure Assessment on September 11-12, 2012 (Frey and Samet, 2012),
12 agreed with the development of risk estimates with reference to zero Os concentration. Based on
13 these recommendations and comments, the second draft REA includes risks associated with 63
14 from all sources after we have simulated just meeting the existing standard and estimates of total
15 risk remaining after meeting alternative levels of the standards. EPA believes that presenting
16 total risk is most relevant given that individuals and populations are exposed to total 63 from all
17 sources, and risks associated with 03 exposure are due to total 03 exposure and do not vary for
18 63 exposure associated with any specific source. In addition, background 63 is fully represented
19 in estimates of total risk given that the measured and adjusted air quality concentrations being
20 used in the risk and exposure analyses include Os produced from precursor emissions from both
21 anthropogenic and background sources. The evidence and information on background 63 that is
22 assessed in the Integrated Science Assessment (ISA) is considered in the Policy Assessment
23 (PA) in conjunction with the total risk estimates provided in this second draft REA. With regard
24 to background 63 concentrations, the PA will consider available information on ambient 63
25 concentrations resulting from natural sources, anthropogenic sources outside the U.S., and
26 anthropogenic sources outside of North America.
27 In providing a broader national characterization of Os air quality in the U.S., this REA
28 draws upon air quality data analyzed in the Os ISA as well as national 03 databases and
29 modeling of Os using the Community Multiscale Air Quality (CMAQ) model. This information,
30 along with additional analyses, is used to develop a broad characterization of recent air quality
31 across the nation. This characterization includes 63 levels in the urban case study areas for the
32 time periods relevant to the risk analysis and information on the spatial and temporal
33 characterization of Os across the national monitoring network. This information is then used to
34 place the relative comparative attributes of the selected study areas into a broader national
35 comparative context to help judge the overall representativeness of the selected study areas in
36 characterizing Os risk for the nation. In addition, to better characterize the spatial patterns of
3-12
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
responses of the distribution of Os to just meeting existing and alternative O?, standards, we also
provide assessments of the historical patterns of responses of 63 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 EXPOSURE ASSESSMENT
Figure 3-2 diagrams the basic structure of the population exposure assessment. Basic
inputs to the exposure assessment include the following: (1) recent measurements of 63
concentrations from monitors in each selected urban study area; (2) Os concentrations that reflect
changes in the distribution of 63 air quality estimated to occur when an area just meets the
existing or alternative 63 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, life-stage development, etc. Basic outputs include
numbers and percent of persons with Os exposures exceeding health-based benchmark levels and
time-series of Os exposures and ventilation rates for individuals (for use in the lung function risk
analysis). Details of the exposure modeling are discussed in Chapter 5.
a Air Quality for Recent Conditions,
and After Just Meeting Existing
and Alternative Standards
1
Population and
Demographic Information
Time Activity
Pattern Data
Physiological
Data
v
Estimation of O3
Exposure
Concentrations
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
V
19 Figure 3-2 Conceptual Diagram for Population Exposure Assessment
3-13
-------
1
2 The scope of the exposure assessment includes 15 urban case study areas.7 These areas
3 were selected to be generally representative of U.S. populations, geographic areas, climates, and
4 different Os and co-pollutant levels, and they include all of the urban case study areas used in the
5 epidemiology-based risk analysis (see Chapter 7). Three additional cities are included in the
6 exposure modeling beyond those included in the epidemiology-based risk analysis. These cities
7 are included to provide additional information on heterogeneity in exposure but could not be
8 included in the epidemiology-based risk analysis because those analyses require additional
9 information not available in the three additional cities. In addition to providing population
10 exposures for estimation of lung function effects, the exposure modeling provides a
11 characterization of urban air pollution exposure environments and activities resulting in the
12 highest exposures.
13 Population exposure to ambient 63 levels is evaluated using version 4.5 of the APEX
14 model. The model and updated documentation are available at
15 http://www.epa.gov/ttn/fera/apex download.html. Exposures are estimated using recent ambient
16 Oj, concentrations, based on 2006-2010 air quality data, and for 63 concentrations resulting from
17 simulations of just meeting the existing 8-hour Os standard and alternative Os standards, based
18 on adjusting 2006-2010 air quality data. Because the Os standard is based on the 3-year average
19 of the 4th highest daily 8-hour maximum, we simulate just meeting the standard for two periods,
20 2006-2008 and 2008-2010. Exposures are estimated for school-age children (ages 5 to 18),
21 asthmatic school-age children, asthmatic adults (ages 19-95), and older persons (ages 65-95).
22 This choice of population groups includes a strong emphasis on children, asthmatics, and persons
23 > 65 years old and reflects the finding of the last O3 NAAQS review (EPA, 2007a) and the ISA
24 (EPA, 2013, Chapter 8) that these are important at-risk groups.
25 In addition to estimating exposures exceeding health-based exposure benchmarks, the
26 exposure estimates are used as an input to the portion of the health risk assessment that is based
27 on exposure-response relationships derived from controlled human exposure studies. The
28 exposure analysis also provides a characterization of populations with high exposures in terms of
29 exposure environments and activities. In addition, the exposure analysis offers key observations
30 based on the results of the APEX modeling, viewed in the context of factors such as averting
31 behavior and key uncertainties and limitations of the model.
7 These cities 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 for most populations in the Seattle area.
3-14
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
3.6 URBAN-SCALE LUNG FUNCTION RISK ANALYSES BASED ON
APPLICATION OF RESULTS FROM 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) personal exposure to ambient Os derived from the
exposure modeling described in Section 3.2.3., 2) data from controlled human exposure studies,
used to construct exposure-response 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 total population and sub-populations, 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 O^ levels and for O^
levels after just meeting existing and alternative standards.
Exposure Estimates:
Individual and Population
Data from controlled
human exposure studies
Lung Function
Ex p o s u re-Res p o nse
Models
Physiological
parameters
Exercise levels
and duration
)% of population with AFEV^I0,15,20%
for recent Oa and after just 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 risk assessments for O?, 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 exposure-response
relationships which are based on analyses of individual data that describe the relationship
between a measure of personal exposure to 63 and the measure(s) of lung function recorded in
3-15
-------
1 the study. The current quantitative lung function risk assessment presents only a partial picture of
2 the risks to public health associated with short-term 63 exposures, as there are additional
3 controlled human exposure studies that have evaluated cardiovascular and neurological outcomes
4 due to Os exposure. However, these studies do not provide sufficient information with which to
5 generate exposure-response functions and therefore are not suitable for quantitative risk
6 assessment.
7 Modeling of risks of lung function decrements is based on application of results from
8 controlled human exposure studies. These studies involve volunteer subjects who are exposed
9 while engaged in different exercise regimens to specified levels of 03 under controlled
10 conditions for specified amounts of time. The responses measured in such studies have included
11 measures of lung function, such as forced expiratory volume in one second (FEVi), respiratory
12 symptoms, airway hyper-responsiveness, and inflammation. The lung function risk assessment
13 includes lung function decrement risk estimates, using FEVi, for the adult population, school-age
14 children (ages 5-18), and asthmatic school-age children (ages 5-18).
15 In addition to estimating lung function decrements for healthy adults that were the study
16 groups in the controlled human exposure studies, this lung function risk assessment estimates
17 lung function decrements (> 10, > 15, and > 20% changes in FEVi) in children 5 to <18 years
18 old. The lung function estimates for children are based on applying data from young adult
19 subjects (18-35 years old) to children aged 5-18. This is based on findings from other chamber
20 studies and summer camp field studies documented in the 1996 03 Staff Paper (U.S. EPA,
21 1996a) and 1996 O3 Criteria Document (U.S. EPA, 1996b), that lung function changes in healthy
22 children are similar to those observed in healthy young adults exposed to Os under controlled
23 chamber conditions.
24 Risk metrics estimated for lung function risk include the numbers of school-age children
25 and other population groups experiencing one or more occurrences of a lung function decrement
26 > 10, > 15, and > 20% in an 63 season and the total number of occurrences of these lung function
27 decrements in school-age children and active school-age children.
28 The risk assessment includes two different modeling approaches. The first approach
29 employs a model that estimates FEVi responses for individuals associated with short-term
30 exposures to Os (McDonnell et al., 2012). This model is based on the data from controlled
31 human exposure studies included in the prior lung function risk assessment as well as additional
32 data sets for different averaging times and breathing rates. These data were from 23 controlled
33 human Os exposure studies that included exposure of 742 volunteers aged 18-35 years (see
34 McDonnell et al., 2007 and McDonnell et al., 2012, for a description of these data). Outputs from
35 this model include FEVi decrements for each simulated individual for each day, which can be
36 used to calculate the population distribution of FEVi decrements, and the percent of the
3-16
-------
1 population with FEVi decrements > 10, > 15, and > 20% after just meeting existing and
2 alternative standards.
3 In addition, we are applying the approach used in the last review and in the first draft of
4 the REA, which employs a probabilistic population-level exposure-response function derived
5 from the results of a number of controlled human exposure studies.
6 This modeling approach uses a smaller set of controlled human exposure studies and the
7 population distribution of 63 exposures to directly estimate the percent of the population with
8 moderate levels of exertion with lung function decrements > 10, > 15, and > 20%.
9 Controlled human exposure studies, carried out in laboratory settings, are generally not
10 specific to any particular real-world location. A controlled human exposure studies-based risk
11 assessment can therefore appropriately be carried out for any locations for which there are
12 adequate air quality data on which to base the modeling of personal exposures. For this
13 assessment, we have selected 15 urban case study areas (matching the areas used in the exposure
14 analysis), representing a range of geographic areas, population demographics, and Os
15 climatology. These 15 areas also include the 12 urban case study areas evaluated in the risk
16 analyses based on concentration-response relationships developed from epidemiological or field
17 studies.
18 In the controlled human exposure study based risk assessment, there are two broad
19 sources of uncertainty to the risk estimates. One of the important sources of uncertainty is the
20 estimation of the population distribution of individual time series of Os exposures and ventilation
21 rates; these uncertainties are addressed as part of the exposure assessment. The second broad
22 source of uncertainty in the risk calculation results from uncertainties in the lung function risk
23 model. Sensitivity analyses are conducted to inform a qualitative discussion of these
24 uncertainties.
25 3.7 URBAN CASE STUDY AREA EPIDEMIOLOGY-BASED RISK ASSESSMENT
26 The major components of the portion of the urban case study area health risk assessment
27 based on data from epidemiological studies are illustrated in Figure 3-4. Basic inputs to this
28 analysis include 1) measured Os concentrations for recent conditions and adjusted air quality
29 representing 63 concentrations after just meeting existing and alternative standards, 2) C-R
30 functions derived from epidemiological studies evaluating associations between Os
31 concentrations and mortality and morbidity endpoints and 3) population counts and baseline
32 incidence rates for mortality and morbidity endpoints. Basic outputs for each urban area include
33 estimates of Cb-attributable incidence and percent Os-attributable incidence for selected
34 mortality and morbidity endpoints and changes and percent changes in Os-attributable incidence.
3-17
-------
National Long-term Exposure
C-R Functions from
Epidemiological Studies
Oa Air Quality for Recent Conditions,
and After Just Meeting Existing
and Alternative Standards
Location-specific Short-term
Exposure C-R Functions
From Epidemiological Studies
_y
BenMAP estimation of O3
Attributable Incidence of Mortality
And Morbidity
Urban Area Population and
Baseline Health Data
v
C Urban Area Estimates of % \
Os Attributable Incidence and
Change in % Os attributable
;idence of mortality and morbidity /
/ Urban Area Estimates of Os
attributable incidence of mortality
and morbidity and change in Oa
\ attributable incidence
2 Figure 3-4 Conceptual Diagram of Urban Case Study Area Health Risk Assessment Based
3 on Results of Epidemiology Studies
4 Epidemiological and field studies provide estimated concentration-response relationships
5 based on data collected in real-world settings. Ambient 63 concentrations used in these studies
6 are typically spatial averages of monitor-specific measurements, using population-oriented
7 monitors. Population health responses for 63 have included population counts of school
8 absences, emergency room visits, hospital admissions for respiratory and cardiac illness,
9 respiratory symptoms, and premature mortality. Risk assessment based on epidemiological
10 studies typically requires baseline incidence rates and population data for the risk assessment
11 locations. To minimize uncertainties introduced by extrapolation, a risk assessment based on
12 epidemiological studies can be performed for the locations in which the studies were carried out,
13 rather than extrapolating results to urban areas where studies for a particular health endpoint
14 have not been conducted.
15 The set of urban case study areas included in this portion of the risk assessment was
16 chosen in order to provide population coverage and to capture the observed heterogeneity in Os-
17 related risk across selected urban study areas. In addition, locations had to have at least one
18 epidemiological study conducted in order for the location to be included for a specific endpoint.
19 This assessment also evaluates the mortality risk results for the selected urban areas within a
20 broader national context to better characterize the nature, magnitude, extent, variability, and
21 uncertainty of the public health impacts associated with 63 exposures. This national-scale
22 assessment is discussed in the next section.
3-18
-------
1 We selected 2007 and 2009 as analysis years for the urban case study area risk analysis.
2 These two years are the midpoint years in the two three-year periods 2006-2008 and 2008-2010.
3 2007 represents a year with generally higher Os concentrations, and 2009 represents a year with
4 generally lower O^ concentrations. Analyses for these two years will provide a good
5 representation of the effects of baseline 63 concentrations on the risk estimates.
6 This risk assessment is focused on health effect endpoints for which the weight of the
7 evidence as assessed in the 63 ISA supports the causal determination that a likely causal or
8 causal relationship exits between a specific health effect category to be due to exposure to 63..
9 The analysis includes estimates of mortality risk associated with short-term 8-hour maximum or
10 8-hour mean 63 concentrations in all 12 urban case study areas, as well as risk of hospitalization
11 for chronic obstructive pulmonary disease and pneumonia. In addition, the analysis includes
12 analysis of hospitalizations for additional respiratory diseases in Los Angeles, New York City,
13 and Detroit, due to limited availability of epidemiological studies covering these endpoints
14 across the 12 urban areas. The analysis also evaluates risks of respiratory related emergency
15 department visits in Atlanta and New York City and risks of respiratory symptoms in Boston,
16 again based on availability of epidemiological studies in these locations. Table 3-1 summarizes
17 the endpoints evaluated for each of the 12 urban case study areas.
18
3-19
-------
1
2
Table 3-1 Short-term O^ Exposure Health Endpoints Evaluated in Urban Case
Areas
Study
Urban Case Study
Area
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
Mortality
X
X
X
X
X
X
X
X
X
X
X
X
COPD and
Pneumonia
hospitalizations
X
X
X
X
X
X
X
X
X
X
X
X
Other
respiratory
hospitalizations
X
X
X
Respiratory
Related ED
visits
X
X
Respiratory
Symptoms
X
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
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 O^ as likely to be causally related to respiratory effects, including respiratory
mortality and morbidity, indicates causal relationship with. There is one national study of long-
term exposures and respiratory mortality which provides a C-R function for use in the risk
assessment. Several other studies have examined long-term exposures and cardiopulmonary
mortality, but consistent with the 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 Oj, and mortality across urban areas, the same
C-R function is applied in each of the 12 urban case study areas. The available epidemiological
studies evaluating long term O^ 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.
We have identified multiple options for specifying the concentration-response functions
for particular health endpoints. This risk assessment provides an array of reasonable estimates for
3-20
-------
1 each endpoint based on the available epidemiological evidence. This array of results provides a
2 limited degree of information on the variability and uncertainty in risk due to differences in study
3 designs, model specification, and analysis years, amongst other differences.
4 As part of the risk assessment, we address both uncertainty and variability. We provide a
5 limited probabilistic characterization of uncertainty in the national-scale mortality risk estimates
6 using the confidence intervals associated with effects estimates (obtained from epidemiological
7 studies). However, this addresses only one source of uncertainty. For other sources of
8 uncertainty, we include a number of sensitivity analyses to evaluate the impact of alternative
9 approaches to simulating just meeting existing and alternative standards, alternative C-R
10 functions, definitions of 63 seasons to which C-R functions are applied, and definitions of urban
11 areas to which the C-R functions are applied. In addition, we evaluate the impact in a subset of
12 locations of using co-pollutant C-R functions. In the case of variability, we identify key sources
13 of variability associated with 63 risk (for both short-term and long-term exposure-related
14 endpoints included in the risk assessment) and discuss the degree to which these sources of
15 variability are reflected in the design of the risk assessment. Finally, we also include a
16 comprehensive qualitative assessment of uncertainty and variability.
17 We also provide a representativeness analysis (see Chapter 8) designed to support the
18 interpretation of risk estimates generated for the set of urban study areas included in the risk
19 assessment. The representativeness analysis focuses on comparing the urban study areas to
20 national-scale distributions for key Os-risk related attributes (e.g., demographics including
21 socioeconomic status, air-conditioning use, baseline incidence rates and ambient 63 levels). The
22 goal of these comparisons is to assess the degree to which the urban study areas provide
23 coverage for different regions of the country as well as for areas likely to experience elevated 63-
24 related risk due to their specific mix of (Vrisk related attributes.
25 The risk assessment based on application of results of epidemiological studies is
26 implemented using the environmental Benefits Mapping and Analysis Program Community
27 Edition (BenMAP-CE) (U.S. EPA, 2013), EPA's GIS-based computer program for the
28 estimation of health impacts associated with air pollution. BenMAP-CE draws upon a database
29 of population, baseline incidence and effect estimates (regression coefficients) to automate the
30 calculation of health impacts. EPA has traditionally relied upon the BenMAP program to
31 estimate the health impacts avoided and economic benefits associated with adopting new air
32 quality rules. It is also suitable for estimating risks associated with ambient concentrations of 63
33 and changes in risk resulting from just meeting existing and alternative Os standards.
3-21
-------
1 3.8 NATIONAL-SCALE MORTALITY RISK ASSESSMENT
2 The major components of the national-scale mortality risk assessment are shown in
3 Figure 3-5. Basic inputs to this analysis are similar to those for the urban case study area
4 epidemiology—based assessment and include 1) gridded O^ concentrations over the continental
5 U.S. for recent conditions, 2) C-R functions relating long-term and short-term exposures to 63 to
6 mortality, and 3) population and baseline mortality rates. Basic outputs include county and
7 national estimates of incidence and percent of mortality attributable to 03.
8 The national-scale mortality risk assessment serves two primary purposes. First, it serves
9 as part of the representativeness analysis discussed above, providing an assessment of the degree
10 to which the urban study areas included in the risk assessment provide coverage for areas of the
11 country expected to experience elevated mortality rates due to (Vexposure. Second, it provides a
12 broader perspective on the distribution of risks associated with recent Os concentrations
13 throughout the U.S., and provides a more complete understanding of the overall public health
o
14 burden associated with 03. We note that a national-scale assessment such as this was completed
15 for the risk assessment supporting the latest PM NAAQS review (US EPA, 2010) with the results
16 of the analysis being used to support an assessment of the representativeness of the urban study
17 areas assessed in the PM NAAQS risk assessment, as described here for 03.
18
1 In the previous O3 NAAQS review, CAS AC commented that "There is an underestimation of the affected
population when one considers only twelve urban "Metropolitan Statistical Areas" (MS As). The CAS AC
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 O3 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 O3 health endpoint.
3-22
-------
1
2
4
5
6
7
8
9
10
1 1
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
National Long-term Exposure
C-R Functions from
Epidemic logical Studies
Nationwide Set of City
Specific Short-term Exposure
C-R Functions from
Epidemic logical Studies
National ambient Os
12 km2 gridded spatial field
for recent co nd itio ns
Ben MAP estimation of Os
Attributable Incidence of Mortality
Nationwide County Specific
Population and Baseline
Health Data
/ County and national level \
I esti mates o f b urd en o f p remature j
V mortality attributable to Os I
County and national level
estimates of % of total mortality
attributable to Os
Figure 3-5 Conceptual Diagram of National
Results of Epidemiology Studies
Mortality Risk Assessment Based on
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
discussion of the urban case study area analyses, the available epidemiological studies evaluating
long-term 63 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 63
exposures.
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. To
address some other key potential sources of uncertainty in the national 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
3-23
-------
1 have greater confidence in the analysis based on the large urban areas included in the
2 epidemiological studies, but the information from the full analysis of all counties is useful to gain
3 understanding of the potential magnitude of risk in less urbanized areas. In addition, the national-
4 scale mortality risk assessment evaluates the sensitivity of the nationwide estimates to
5 assumptions about the transferability of effect estimates from the cities included in the
6 underlying epidemiological studies to other cities in the U.S. Finally, the assessment includes a
7 sensitivity analysis evaluating the use of regional priors city—rather than using a national prior in
8 developing the city specific Bayesian adjusted effect estimates.9 These sensitivity analyses are
9 described in detail in Chapter 8.
10 The national-scale risk assessment is conducted only for recent 63 conditions. We do not
11 attempt to simulate nationwide Os concentrations that would result from just meeting the existing
12 or alternative O^ standards everywhere in the U.S. Such a simulation would require detailed
13 modeling of attainment strategies in all potential non-attainment areas and would need to take
14 into account the interdependence of Os concentrations across urban areas. This type of analysis is
15 beyond the scope of this risk assessment. Analyses of nationwide attainment are included as part
16 of the Regulatory Impact Analyses that accompany proposed and final rulemaking packages and
17 will likely be included in the rulemaking portion of this review.
18 3.9 PRESENTATION OF EXPOSURE AND RISK ESTIMATES TO INFORM THE O3
19 NAAQS POLICY ASSESSMENT
20 We present exposure estimates in three ways: person-occurrences, number, and percent
21 of persons in different populations (e.g., adults, all school-age children, asthmatic school-age
22 children, outdoor workers) with at least one 8-hour average exposure at or above benchmark
23 levels of 60 ppb, 70 ppb, and 80 ppb. In addition, the same types of results are shown for persons
24 with multiple exposures at or above the benchmark levels. The results are presented in summary
25 tables and graphics, while detailed tables of results are provided in an appendix. The focus in the
26 presentation of results is on exposures occurring after simulating just meeting the existing
27 standard and on the change in number and percent of exposures between meeting the existing
28 standard and meeting alternative standards. Results are presented for the five modeled years, for
29 all 15 urban case study areas.
30 Quantitative risk estimates from the analyses based on application of controlled human
31 exposure studies are presented for the two different risk models. For each model, we provide
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.
3-24
-------
1 estimates of the percent of different populations (adults, all children, children with asthma) with
2 lung function decrements greater than or equal to 10, 15, and 20 percent. As with exposure, the
3 focus in the presentation of results is on risk occurring after simulating just meeting the existing
4 standards and on the change in risk occurring between meeting the existing standard and meeting
5 alternative standards.
6 Results from the epidemiology-based risk assessment are presented in two ways: (1) total
7 (absolute) health effects incidence for recent air quality and simulations of air quality just
8 meeting the existing and alternative standards under consideration and (2) risk reduction
9 estimates, reflecting the change in the distribution of Os between scenarios of just meeting the
10 existing standard and just meeting alternative standards. In addition, risks are presented as the
11 percent of baseline incidence, and risks per 100,000 population, to allow for comparisons
12 between urban areas with very different population sizes. We include risk modeled across the
13 full distribution of 63 concentrations, as well as core risk estimates for 63 concentrations down
14 to 0 ppb.
15 We present an array of risk estimates in order to provide additional context for
16 understanding the potential impact of uncertainty on the risk estimates. For core estimates and
17 sensitivity analyses, we provide the statistical confidence intervals, demonstrating the relative
18 precision of estimates. The graphical presentation of sensitivity analyses focuses on the
19 differences from the core estimates in terms of risk per 100,000 population.
20 The results of the representativeness analysis are presented using cumulative probability
21 plots (for the national-lev el distribution of 63 risk-related parameters) with the locations where
22 the individual urban study areas fall within those distributions noted in the plots using vertical
23 lines. Similar types of plots are used to present the distribution of national-scale mortality
24 estimates based on the national-scale risk assessment, showing the location of the urban case
25 study areas within the overall national distribution.
26 Chapter 9 of this risk and exposure assessment provides a synthesis of the results from
27 the four assessments (urban case study area exposure, urban case study area lung function risk,
28 urban case study area epidemiology-based risk, and national mortality risk). Chapter 9 focuses
29 on comparing patterns of results across locations, years, and alternative standards. Chapter 9 also
30 provides perspective on the overall degree of confidence of the analyses and the
31 representativeness of the set of results in characterizing patterns of exposure and risk and
32 patterns of changes in exposure and risk from just meeting alternative standards relative to just
33 meeting the existing standards.
3-25
-------
1 3.10 REFERENCES
2 Chen, J., R. Zhao and Z. Li. 2004. Voronoi-based k-order neighbor relations for spatial analysis.
3 ISPRS J Photogrammetry Remote Sensing, 59(1-2), 60-72.
4 Folinsbee, L.J.; W.F. McDonnell and D.H. Horstman. 1988. Pulmonary function and symptom
5 responses after 6.6-hour exposure to 0.12 ppm O3 with moderate exercise. JAPCA. 38:
6 28-35.
7 Gold, C. 1997. Voronoi methods in GIS. In: Algorithmic Foundation of Geographic Information
8 Systems (va Kereveld M.; J. Nievergelt; T. Roos; P. Widmayer, eds). Lecture Notes in
9 Computer Science, Vol 1340. Berlin: Springer-Verlag, 21-35.
10 Henderson, R. 2007. Clean Air Scientific Advisory Committee's (CAS AC) Review of the
11 Agency's Final O3 Staff Paper. EPA-C AS AC-07-002. March 26.
12 Horstman, D.H.; LJ. Folinsbee; PJ. Ives; S. Abdul-Salaam and W.F. McDonnell. 1990. O3
13 concentration and pulmonary response relationships for 6.6-hr exposures with five hrs of
14 moderate exercise to 0.08, 0.10, and 0.12 ppm. Am Rev Respir Dis. 142:1158-1163.
15 Kim, Chong S.; N.E. Alexis; A.G Rappold; H. Kehrl; J. Milan; J.C. Hazucha; J.C. Lay; M.T.
16 Schmitt; C. Martin; R.B. Devlin; D.B. Peden and D. Diaz-Sanchez. 2011. "Lung
17 Function and Inflammatory Responses in Healthy Young Adults Exposed to 0.06 ppm O3
18 for 6.6 Hours." American Journal of Respiratory and Critical Care Medicine 183, no. 9
19 (2011): 1215-1221.
20 Langstaff, I.E. 2007. OAQPS Staff Memorandum to O3 NAAQS Review Docket (EPA HQ-
21 OAR-2005-0172). Subject: Analysis of Uncertainty in O3 Population Exposure
22 Modeling. [January 31, 2007].
23 .
24 McDonnell, W.F., et al. 1991. "Respiratory response of humans exposed to low levels of O3 for
25 6.6 hours." American Review of Respiratory Disease, 147:804-810.
26 McDonnell, W.F.; P.W. Stewart and M.V. Smith. 2007. "The temporal dynamics of O3 -induced
27 FEVi changes in humans: an exposure-response model." Inhalation Toxicology, 19:483-
28 494.
29 McDonnell W.F.; P.W. Stewart and M.V. Smith. 2010. "Prediction of O3 -induced lung function
30 responses in humans." Inhalation Toxicology, 22(2): 160-8.
31 McDonnell, W.F.; P.W. Stewart; M.V. Smith; C.S. Kim and E.S. Schelegle. 2012. Prediction of
32 lung function response for populations exposed to a wide range of ozone conditions.
33 Inhalation Toxicology 24: 619-633.
34 Samet, J. 2009. Consultation on EPA's Draft Integrated Review Plan for the National Ambient
3 5 Air Quality Standards for Parti culate Matter. (EPA document number EPA-C AS AC-10-
36 004. Decembers).
37 Samet, J. 2011. Consultation on EPA's O3 National Ambient Air Quality Standards: Scope and
38 Methods Plan for Health Risk and Exposure Assessment (April 2011) and O3 National
39 Ambient Air Quality Standards: Scope and Methods Plan for Welfare Risk and Exposure
40 Assessment (April 2011). EPA-C AS AC-11-008. June 21.
3-26
-------
1 Schelegle, E.S.; C.A. Morales; W.F. Walby; S. Marion and R.P. Allen. 2009. "6.6-Hour
2 Inhalation of Os Concentrations from 60 to 87 Parts per Billion in Healthy Humans."
3 American Journal of Respiratory and Critical Care Medicine 180 (2009): 265-272.
4 Simon H.; K.R. Baker; F. Akhtar; S.L. Napelenok; N. Possiel; B. Wells and B. Timin. 2013. A
5 Direct Sensitivity Approach to Predict Hourly Ozone Resulting from Compliance with
6 the National Ambient Air Quality Standard. Environmental Science and Technology.
7 2013 Feb. 12. [Epub ahead of print]
8 U.S. Environmental Protection Agency. 1996a. "Review of National Ambient Air Quality
9 Standards for Os : Assessment of Scientific and Technical Information - OAQPS Staff
10 Paper." (EPA document number EPA/452/R-96-007). Research Triangle Park, NC:
11 Office of Air Quality Planning and Standards. Available from: NTIS, Springfield, VA;
12 PB96-203435. < http://www.epa.gov/ttn/naaqs/standards/O^ /s O^ pr.html>.
13 U.S. EPA. 1996b. "Air Quality Criteria for O3 and Related Photochemical Oxidants." (EPA
14 document number EPA/600/P-93/004aF-cF). Research Triangle Park, NC: Office of
15 Research and Development, National Center for Environmental Assessment.
16 < http://cfpub.epa.gov/ ncea/cfm/recordisplay.cfm?deid=2831>.
17 U.S. EPA. 2007a. Os Population Exposure Analysis for Selected Urban Areas. Research
18 Triangle Park, NC: Office of Air Quality Planning and Standards. (EPA document
19 number EPA-452/R-07-010, July).
20 U.S. EPA. 2007b. Os Health Risk Assessment for Selected Urban Areas. Research Triangle Park,
21 NC: Office of Air Quality Planning and Standards. (EPA document number EPA 452/R-
22 07-009, July).
23 U.S. EPA. 2007c. "Review of the National Ambient Air Quality Standards for O3: Policy
24 Assessment of Scientific and Technical Information. OAQPS Staff Paper." Research
25 Triangle Park, NC: Office of Air Quality Planning and Standards. (EPA report number
26 EPA-452/R-07-007a. < http://www.epa.gov/ttn/naaqs/standards/OVs O3 cr sp.html >.
27 U.S. EPA. 2008a." National Air Quality: Status and Trends Through 2007." Research Triangle
28 Park, NC: Office of Air Quality Planning and Standards. (EPA document EPA-454/R-
29 08-006, November), .
30 U.S. EPA. 2008b. "EPA's Report on the Environment." Washington, DC. (EPA document
31 number EPA/600/R-07/045F, May 2008).
32 .
33 U.S. EPA. 2008c. "Risk and Exposure Assessment to Support the Review of the NO2 Primary
34 National Ambient Air Quality Standard." (EPA document number EPA-452/R-08-008a,
35 November 2008).
36 .
37 U.S. EPA. 2009a. Integrated Review Plan for the Os National Ambient Air Quality Standards
38 Review: External Review Draft. Research Triangle Park, NC: Environmental Media
39 Assessment Group, National Center for Environmental Assessment and Health and
40 Environmental Impacts Division, Office of Air Quality Planning and Standards. (EPA
41 document number EPA 452/D-09-001, September).
3-27
-------
1 U.S. EPA. 2009b. "Risk and Exposure Assessment to Support the Review of the SO2 Primary
2 National Ambient Air Quality Standard." (EPA document number EPA-452/R-09-007,
3 August 2009).
4
5 U.S. EPA. 2010a. "Quantitative Risk and Exposure Assessment for Carbon Monoxide -
6 Amended." EPA Office of Air Quality Planning and Standards. (EPA document number
7 EPA-452/R-10-009, July). .
9 U.S. EPA. 2010b. "Quantitative Health Risk Assessment for Particulate Matter." Research
10 Triangle Park, NC: Office of Air Quality Planning and Standards. (EPA document
11 number EPA-452/R-10-005).
12 .
13 U.S. EPA. 201 la. Integrated Review Plan for the Os National Ambient Air Quality Standards.
14 Research Triangle Park, NC: National Center for Environmental Assessment, Office of
15 Research and Development and Office of Air Quality Planning and Standards, Office of
16 Air and Radiation. (EPA document number EPA 452/R-l 1-006, April).
17 U.S. EPA. 201 Ib. Os National Ambient Air Quality Standards: Scope and Methods Plan for
18 Health Risk and Exposure Assessment. Research Triangle Park, NC: Office of Air
19 Quality Planning and Standards. (EPA document number EPA-452/P-11-001, April).
20 U.S. EPA. 2011 c. Os National Ambient Air Quality Standards: Scope and Methods Plan for
21 Welfare Risk and Exposure Assessment. Research Triangle Park, NC: Office of Air
22 Quality Planning and Standards. (EPA document number EPA-452/P-11-002, April).
23 U.S. EPA. 2013. "Environmental Benefits Mapping Analysis Program Community Edition
24 (BenMAP-CE vl.O)." Research Triangle Park, NC: Health and Environmental Impacts
25 Division, Office of Air Quality Planning and Standards.
26 .
27 Wegman, L. 2012. Updates to information presented in the Scope and Methods Plans for the Os
28 NAAQS Health and Welfare Risk and Exposure Assessments. Memorandum from Lydia
29 Wegman, Division Director, Health and Environmental Impacts Division, Office of Air
30 Quality Planning and Standards, Office of Air and Radiation, US EPA to Holly
31 Stallworth, Designated Federal Officer, Clean Air Scientific Advisory Committee, US
32 EPA Science Advisory Board Staff Office. May 2, 2012.
33 World Health Organization. 2008. Harmonization Project Document No. 6. Part 1: Guidance
34 Document on Characterizing and Communicating Uncertainty in Exposure Assessment.
35 .
36
37
3-28
-------
1 4 AIR QUALITY CONSIDERATIONS
2 4.1 INTRODUCTION
3 Air quality information is used in Chapters 5-8 to assess risk and exposure resulting from
4 recent O?, concentrations, as well as to estimate the relative change in risk and exposure that
5 could result from just meeting the existing 63 standard of 75 ppb and the potential alternative
6 standard levels of 70 ppb, 65 ppb, and 60 ppb1. The same air quality data are used to examine
7 fifteen2 urban case study areas in the population exposure analyses discussed in Chapter 5 and
8 the lung function risk assessment based on application of results from clinical studies discussed
9 in Chapter 6: Atlanta, GA; Baltimore, MD; Boston, MA; Chicago, IL; Cleveland, OH; Dallas,
10 TX; Denver, CO; Detroit, MI; Houston, TX; Los Angeles, CA; New York, NY; Philadelphia,
11 PA; Sacramento, CA; St. Louis, MO; and Washington, DC. The epidemiology-based risk
12 assessment discussed in Chapter 7 examines twelve3 of the fifteen urban case study areas
13 evaluated in the population exposure analyses. Finally, Chapter 8 includes an assessment of the
14 national-scale Os mortality risk burden associated with recent Os concentrations, and
15 characterizes the representativeness of the 15 urban case study areas compared to the rest of the
16 U.S. This chapter describes the air quality information developed for these analyses, providing
17 an overview of monitoring data and air quality (section 4.2) and an overview of air quality inputs
18 to the risk and exposure assessments (section 4.3).
19
20 4.2 OVERVIEW OF O3 MONITORING AND AIR QUALITY DATA
21 To determine whether or not the NAAQS have been met at an ambient Os monitoring
22 site, a statistic commonly referred to as a "design value" must be calculated based on 3
23 consecutive years of data collected from that site. The form of the existing O?, NAAQS design
24 value statistic is the 3-year average of the annual 4th highest daily maximum 8-hour Oj
25 concentration in parts per billion (ppb), with decimal digits truncated. The existing primary and
26 secondary Oj, NAAQS are met at an ambient monitoring site when the design value is less than
1 For a subset of urban areas and analyses, the REA evaluates a standard level of 55 ppb, consistent with
recommendations from CASAC 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 4-E.
3 These cities 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.
4-1
-------
1 or equal to 75 ppb.4 In counties or other geographic areas with multiple monitors, the area-wide
2 design value is defined as the design value at the highest individual monitoring site, and the area
3 is said to have met the NAAQS if all monitors in the area are meeting the NAAQS.
4 Air quality monitoring data from 1,468 U.S. ambient O^ monitoring sites were retrieved
5 by EPA staff for use in the risk and exposure assessments. The initial dataset consisted of hourly
6 Os concentrations in ppb collected between 1/1/2006 and 12/31/2010 from these monitors. Data
7 for nearly 1,400 of these monitors were extracted from EPA's Air Quality System (AQS)
8 database5, while the remaining data came from EPA's Clean Air Status and Trends Network
9 CASTNET) database which consists of primarily rural monitoring sites. While CASTNET
10 monitors did not begin reporting regulatory data to AQS until 2011, it is generally agreed that
11 data collected from these monitors prior to 2011 is of comparable quality to the data reported to
12 AQS.
13 These data were split into two design value periods, 2006-2008 and 2008-2010, and all
14 subsequent analyses based on these data were conducted independently for these two periods.
15 Observations flagged in AQS as having been affected by exceptional events were included the
16 initial dataset, but were not used in design value calculations in accordance with EPA's
17 exceptional events policy. Missing data intervals of 1 or 2 hours in the initial dataset were filled
18 in using linear interpolation. These short gaps often occur at regular intervals in the ambient data
19 due to an EPA requirement for monitoring agencies to perform routine quality control checks on
20 their O^ monitors. Quality control checks are typically performed between midnight and 6:00
21 AM when Oi concentrations are low. Missing data intervals of 3 hours or more were not
22 replaced. Interpolated data values were not used in design value calculations.
23 Figures 4-1 and 4-2 show the design values for the existing Oj, NAAQS for all regulatory
24 monitoring sites in the U.S. for the 2006-2008 and 2008-2010 periods, respectively. In general,
25 O3 design values were lower in 2008-2010 than in 2006-2008, especially in the Eastern U.S.
26 There were 518 Os monitors in the U.S. with design values above the existing standard in 2006-
27 2008, compared to only 179 in 2008-2010.
28
4 For more details on the data handling procedures used to calculate design values for the current O3 NAAQS, see 40
CFR Part 50, Appendix P.
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
-------
1
2
3
i O ^ ' gL»*T«
o • « o °£o%
6^ !:•
8-Hour Ozone Design Values, 2006-200
• 33-60 ppb (49 Sites)
O 61-65 ppb (65 Sites)
O 66-70 ppb (140 Sites)
• 71-75 ppb (279 Sites)
• 76-120 ppb (518 Sites)
Figure 4-1 Map of Monitored 8-hour O3 Design Values for the 2006-2008 Period
4-3
-------
1
2
8-Hour Ozone Design Values, 2008-2010
• 33-60 ppb (79 Sites)
• 61-65 ppb (165 Sites)
O 66-70 ppb (305 Sites)
O 71-75 ppb (300 Sites)
• 76-120 ppb (179 Sites)
Figure 4-2 Map of Monitored 8-hour O3 Design Values for the 2008-2010 Period
4 4.3 OVERVIEW OF URBAN-SCALE AIR QUALITY INPUTS TO RISK AND
5 EXPOSURE ASSESSMENTS
6 The air quality information input into the urban-scale risk and exposure assessments
7 includes both recent air quality data from the years 2006-2010, as well as air quality data
8 adjusted to reflect just meeting the existing and potential alternative standard levels. In this
9 section, we summarize these air quality inputs and discuss the methodology used to adjust air
10 quality to meet the existing and potential alterative standards.
11 Figure 4-3 presents a flowchart of air quality data processing steps for the urban-scale
12 analyses. The rest of section 4.3.1 will provide more details on each step depicted in the flow
13 diagram. Additional information is provided in Appendices 4-A, 4-B and 4-D.
4-4
-------
Recent 1
Monitored O3: I
|Hourly 2006-2010 measurements f
at individual monitors J
Community Multiscale
AirQuality model
i nstru mented with the
Higher Order Decoupled Direct
Method (CMAQ-HDDM)
O3 sensitivity
coefficients
Dai iyO3 Metrics at
co m pos ite rn on itor8 fo r
recent conditions in urban
case study areas.
Metrics include:
MDA8
8-hr average
1-hr max
Adjustment of hourly ozone
values for all potential
emissions reduction levels
Daily O3 metrics at
composite monitor8 in urban case
study areas after Just meeting
existing and alternative standards
Metrics include:
MDA8
8-hr average
1-hr max
Hourly ozone concentrations after
Just meeting existing and
a He r nat i ve sta n d a rds
Hourly9 census-tract level
VNA surfaces for each urban case
study area for recent conditions
Air Pollution Exposure
Model (APEX)
Hourly9 census-tract level
VNA surfaces for each urban case
study area after Just meet! ng exist! ng j
And alternative standards
Epi-based Risk Assessment for
Urban Case Study areas.
Endpoints:
Deaths, hospital admissions etc
Clinical-based riskassessment
Endpoints:
Number of individuals experiences
FEV1 decrements > 10%
1
3
4
5
6
Exposure assessment
Endpoints:
N umber of Individ uals exposures to
A single hourly ozone concentration
above 60, 70, and 80 ppb
Benchmarklevels
7
8
9
10
1 1
12
13
Figure 4-3 Flowchart of Air Quality Data Processing for Different Parts of the Urban-
scale Risk and Exposure Assessments
4.3.1 Urban Case Study Areas
4.3.1.1 Exposure Modeling and Controlled Human Study Based Lung Function Risk
Assessment
The 15 urban case 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 case study areas in the exposure assessment,
including the number of ambient monitoring sites, the required C»3 monitoring season, and the
2006-2008 and 2008-2010 design values for each study area. All 15 of the urban case study areas
had 8-hour 63 design values above the existing standard in 2006-2008, while 13 urban areas had
6 Composite monitors do not always include the highest design value monitor in every urban area.
7 4800 VNA surfaces were created for each urban area/alternative standard level pair: 24 hrs x 365 days x 5 years.
4-5
-------
1
2
O
4
5
8-hour 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.
6Table 4-1 Monitor and Area Information for the 15 Urban Case Study Areas in the Exposure
7 Modeling and Clinical Study Based Risk Assessment
#of #ofO3 Population Required O3 2006-2008 2008-2010
Area Name Counties Monitors (2010) Monitoring Season DV (ppb) DV (ppb)
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
33
7
10
16
8
11
13
9
10
5
27
15
7
17
26
13
7
14
26
13
20
26
12
22
54
31
19
26
17
22
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
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
95
91
83
78
82
89
86
81
91
119
90
92
102
85
87
80
89
77
74
77
86
77
75
85
112
84
83
102
77
81
9
10
11
12
13
14
15
16
17
Since O3 design values are based on the annual 4* highest 8-hour daily maximum O3
concentrations from 3-consecutive years, it is useful to look at inter-annual variability. In
general, the annual 4th highest 8-hour O3 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 4* 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.
4-6
-------
4
5
6
7
2006
2006
2006
Baltimore
Boston
NewYork
Philadelphia
Washington
2007
2008
2009
2010
Atlanta
Chicago
Cleveland
Detroit
SaintLouis
2007
2008
2009
2010
Dallas
Denver
Houston
LosAngeles
Sacramento
2007
2008
2009
2010
Figure 4-4 Trends in Annual 4th Highest 8-hour Daily Maximum Os
Concentrations in ppb for the 15 Urban Case Study Areas for 2006-
2010. Urban areas are grouped into 3 regions: Eastern (top), Central
(middle), and Western (bottom).
4-7
-------
4.3.1.2 Epidemiology Based Risk Assessment
1 Table 4-2 gives some basic information on the 12 urban case study areas in the
2 epidemiology-based risk assessment for each set of area boundaries. The spatial extent of each
3 urban case study area was based on the respective Core Based Statistical Area (CBSA)8. The
4 CBS As were generally smaller than the study areas used in the exposure modeling and clinical
5 study based risk assessments, except for Baltimore and Houston, where the two study areas were
6 identical. The rationales for the definitions of the spatial areas used in each type of analysis are
7 provided in the corresponding chapters. The final two columns in Table 4-2 show the annual 4*
8 highest daily maximum 8-hour Oi concentration in ppb for the monitors within each urban case
9 study area in 2007 and 2009.
10 It should be noted that the CBSA boundaries used for the urban case study areas in this
11 assessment are different than those used in the 1st draft of the REA, where the study areas were
12 derived from the Zanobetti and Schwartz (2008) study. The change to the CBSA boundaries was
13 intended to capture a larger portion of the urban area populations by including some surrounding
14 suburban counties, rather than focusing strictly on the urban population centers. Two sensitivity
15 analyses were conducted to determine the effect of changing the spatial extent of the urban case
16 study areas on the epidemiology-based risk estimates. These sensitivity analyses are presented in
17 Chapter 7, and a summary of the two alternative sets of boundaries for the 12 urban case study
18 areas are provided in Appendix 4-A.
19 Since 63 is not directly emitted but is formed through photochemical reactions, precursor
20 emissions may continue to react and form O^ downwind of emissions sources, thus the highest
21 Oj, concentrations are often found downwind of the highest concentrations of precursor
22 emissions near the urban population center. There were some instances where the highest
23 monitor occurred outside of the CBSA, but within the exposure area, which was designed to
24 always include the monitor associated with the area-wide design value. For example, in Los
25 Angeles, the CBSA includes Los Angeles and Orange counties, but the highest Os concentrations
26 are typically measured further downwind in Riverside and San Bernardino counties. Thus, the
27 values reported in Table 4-2 may not match the values shown in Figure 4-4.
28
29
30
31
1 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 deliniations from 2008. For more information see:
h1tp://www.whitehouse.gov/sites/default/files/omb/assets^ulletins^lO-02.pdf
4-8
-------
1 Table 4-2
2
Monitor and Area Information for the 12 Urban Case Study Areas in the
Epidemiology Based Risk Assessment
#ofO3 Population 2007 4th high 4th high
Area Name # of Counties Monitors (2010) (ppb) (ppb)
Atlanta
Baltimore
Boston
Cleveland
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
28
7
7
5
10
6
10
2
23
11
4
16
13
7
11
10
16
8
22
21
22
15
17
17
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
102
92
89
83
97
93
90
105
94
102
93
94
77
83
75
72
79
73
91
108
81
74
96
74
3
4
5
6
7
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 4-A.
4.3.2.1 Exposure Modeling and Controlled Human Exposure Study Based Risk
Assessment
10 As discussed in more detail in Chapter 5, the REA uses the Air Pollutants Exposure
11 (APEX) model (U.S. EPA, 2012a, b) to simulate exposure and to estimate lung function
12 decrements based on application of results of controlled human exposure studies to populations
13 in the 15 urban case study areas. The APEX model uses spatial fields of hourly O^
14 concentrations at each census tract within an urban area to simulate exposure. In the first draft
15 REA, these hourly spatial fields were generated for four urban areas using the concentrations
16 from the nearest neighboring Os monitor. In this draft, we use Voronoi Neighbor Averaging
17 (VNA) (Gold, 1997; Chen et al, 2004) to estimate hourly 63 concentrations at each census tract
18 in all 15 urban case study areas, for recent measured air quality, air quality meeting the existing
19 standard of 75 ppb, and air quality meeting potential alternative standards. The VNA fields were
4-9
-------
1 estimated using ambient hourly O^ concentrations from monitors in each urban area, as well as
2 monitors within a 50 km buffer region around the boundaries of each area. Additional details on
3 the procedure used to generate the VNA fields, and a technical justification for the change from
4 nearest neighbor fields to VNA fields are included in Appendix 4-A.
5
6 Figure 4-5 shows county-level maps of the 15 urban case study areas. Counties colored
7 pink indicate the study area boundaries used in the Zanobetti & Schwartz (2008) and/or Smith et
8 al (2009b) studies9, where applicable. Counties colored gray indicate additional counties within
9 the CBS A boundaries, and counties colored peach indicate any additional counties included in
10 the exposure and lung function risk assessments. The X's indicate locations of the 63 monitors
11 used in the risk and exposure assessments, including those within the 50 km buffer region used
12 to create the VNA fields.
9 The Zanobetti and Schwartz (2008) and Smith et al (2009) study area boundaries were identical for 6 of the 12
urban case study areas, and had at least one county in common for all 12 urban case study areas. The
'Epidemiology Study Area' labels in figures 4-5 refer to counties included in either of these two studies.
4-10
-------
Baltimore
Philadelphia
1
2
3
New York
U 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 Case Study Areas Including
Locations
Monitor
4-11
-------
Atlanta
Detroit
Cleveland
U Epidemiology
Study Area
D Additional Counties
in CBSA
D Additional Counties
in Exposure Area
x Ozone Monitor
1
2 Figure 4-5b Maps of the 5 Central U.S. Urban Case Study Areas Including Os Monitor
3 Locations
4-12
-------
Dallas
Los Angeles
1
2
3
U 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 Case Study Areas Including
Locations
Monitor
4-13
-------
4.3.2.2 Epidemiology Based Risk Assessment
1 We input Os air quality concentration data for the epidemiology-based risk analyses into
2 the environmental Benefits Mapping and Analysis Program Community Edition (BenMAP-CE)
3 (U.S. EPA, 2013) for assessment. We used BenMAP to analyze four different daily Oi metrics in
4 12 of the 15 urban case study areas, which were the basis for concentration-response
5 relationships derived in various epidemiology studies:
6 1. Daily maximum 1-hour concentration
7 2. Daily maximum 8-hour concentration
8 3. Daytime 8-hour average concentration (10:OOAM to 6:OOPM)
9 4. Daily 24-hour average concentration
10 The air quality monitoring data used in BenMAP were daily time-series of "composite
11 monitor" values for each of the 12 urban areas for years 2007 and 2009, which were chosen to
12 represent years with high and low Os concentrations, respectively. The composite monitor values
13 were calculated by first averaging the hourly Oj, concentrations for all monitors within the area-
14 of-interest (resulting in a single hourly time-series for each urban area), then calculating the four
15 daily metrics listed above. More details on the composite monitor value calculations and a
16 presentation of the resulting concentrations can be found in Appendices 4-A and 4-D,
17 respectively.
18
19 4.3.3 Air Quality Adjustments for "Just Meeting" Existing and Potential Alternative Os
20 Standards
21 The focus of the risk and exposure assessments is the evaluation of risks and exposures
22 after just meeting existing and alternative standards, and the change in risk between just meeting
23 existing standards and just meeting alternative standards. These evaluations require estimation of
24 the change in hourly O^ concentrations that may occur in each urban area when "just meeting"
25 the existing and potential alternative 63 standards.
26 The first draft REA and the previous Os NAAQS review used the "quadratic rollback"
27 method to adjust ambient 63 concentrations to simulate just meeting existing and alternative
28 standards (U.S. EPA, 2007; Wells et al., 2012).Although the quadratic rollback method replicates
29 historical patterns of air quality changes better than some alternative methods (e.g. simply
30 shaving peak concentrations off at the NAAQS level and the proportional rollback technique), its
31 implementation relies on a statistical relationship instead of on a mechanistic characterization of
32 physical and chemical processes in the atmosphere. Because of its construct as a statistical fit to
33 measured 63 values, the quadratic rollback technique cannot capture spatial and temporal
34 heterogeneity in Os response and also cannot account for nonlinear atmospheric chemistry that
4-14
-------
1 causes increases in O?, during some hours and in some locations as a result of emissions
2 reductions under some circumstances.
3 Photochemical grid models are better able to simulate these phenomena and therefore the
4 first draft REA proposed to replace quadratic rollback with a model-based O^ adjustment
5 methodology and presented a test case for Atlanta and Detroit using modeling for July/ August
6 2005 (Simon et al., 2012). The section below summarizes the methodology applied in this
7 second draft REA to adjust air quality for attainment of existing and alternative standards. This
8 new methodology applies Higher-Order Decoupled Direct Method (HDDM) capabilities in the
9 Community Multi-scale Air Quality (CMAQ) model to simulate the response of O^
10 concentrations to reductions in US anthropogenic NOX and VOC emissions. The model
1 1 incorporates anthropogenic U.S., Canadian, Mexican and other international emissions, as well
12 as emissions from non-anthropogenic sources. Since sources of background O?, are incorporated
13 explicitly in the modeling, specifying U.S. background concentrations is unnecessary.
14 Application of this approach also addresses the recommendation by the National Research
15 Council of the National Academies (NRC, 2008) to explore how emissions reductions might
16 effect temporal and spatial variations in 63 concentrations, and to include information on how
17 NOX versus VOC control strategies might affect risk and exposure.
18
4.3.3.1 Methods
19 The EPA has developed an HDDM-adjustment methodology to estimate hourly 63
20 concentrations that could occur at each monitor location if urban case study areas were to meet
21 the existing and various alternative levels of the 63 standard. An early version of this
22 methodology was proposed in the first draft REA (Simon et al., 2012). The methodology was
23 later improved and published in a peer-reviewed journal (Simon et al., 2013). The methodology
24 and its application to hourly 63 concentrations in the urban case study areas is summarized
25 below and described in more detail in Appendix 4-D.
26 The HDDM-adjustment methodology uses the CMAQ photochemical model to determine
27 monitoring site-specific response of hourly 63 concentrations to reductions in US anthropogenic
28 NOx and VOC emissions. These responses are then applied to ambient data to create a 5-year
29 time-series of hourly Os concentrations at each monitor location which is consistent with
30 meeting various potential levels of the 63 NAAQS for the two three-year attainment periods
3 1 2006-2008 and 2008-2010. The steps are outlined in Figure 4-6 and summarized below:
32 • Step 1: Run CMAQ simulation with HDDM to determine hourly Os sensitivities to NO
33 emissions and VOC emissions for the grid cells containing monitoring sites in an urban
34 area.
4-15
X
-------
1 • Inputs: Model-ready emissions and meteorology data
2 • Outputs: Os concentrations and sensitivities at locations of monitoring sites for
3 each hour in January and April-October, 2007
4 • Step 2: For each monitoring site, season, and hour of the day use linear regression to
5 relate first order sensitivities of NOX and VOC (SVc* and Svoc) to modeled Os and
6 second order sensitivities to NOx and VOC (S2^Ox and S2yoc) to the first order
7 sensitivities.
8 • Inputs: Step 1 outputs
9 • Outputs: Functions to calculate typical sensitivities based on monitor location,
10 Os concentration, season, and hour of the day
11 • Step 3: For each measured hourly Os value between 2006 and 2010, calculate the first
12 and second order sensitivities based on monitoring site-, season-, and hour-specific
13 functions derived in Step 2.
14 • Inputs: Step 2 outputs and hourly ambient data for 2006-2010.
15 • Outputs: Hourly Os observations paired with modeled sensitivities for all hours
16 in 2006-2010 at all monitor locations
17 • Step 4: Adjust measured hourly Os concentrations for incrementally increasing levels of
18 emissions reductions using assigned sensitivities and then recalculate design values until
19 an emissions reduction level is reached at which all monitors in an urban area are below
20 the existing and potential alternative levels of the standard.
21 • Inputs: Step 3 outputs
22 • Outputs: Adjusted hourly Os values for 2006-2010 at monitor locations to show
23 compliance with the existing and potential alternative standard levels based on the
24 three year average of the 4* highest 8-hour daily max Os value. For each
25 standard, two sets of data are created: 2006-2008 and 2008-2010. Because the
26 emissions reductions used to attain standards in the two time periods might be
27 different, adjusted 2008 Os values are different for the two sets of data.
28
4-16
-------
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Anthropogenic
). . „ Canada and
1 ua 1 Mexico
1
Outside North America
|_
O3 and O3 Precursor Emissions
"i 1
• T n
1 1 Meteorology
' 1 I J
i
r^ ^
i
| | Initial and Boundary conditions
i 1 ^ J
Other Model Inputs
\ '
Step 1a:
CMAQ
HDDM Modeling
(Jan. Apr-Oct 2007)
/ Gridded hourly O3 \
\°°™"y
Step 1 b:
Extract Output
Unique Linear
Relationships Between
Sensitivities and Hourly
Ozone at Each Monitor
Location for Each
Season and
Hour-of-the-Day
E
Step 2:
Create Regressions
Step 4:
Adjust Hourly Ozone
to Meet Alternate
Standards
Ozone Concentrations
& Sensitivities at
Locations of
Monitoring Sites for
Each Modeled Hour
(Jan, Apr-Oct 2007)
Adjusted Hourly Ozone"\
Values for 2006-201 Oat \
Each Monitor Location to
Show Attainment with /
Alternate Standards y
Figure 4-6 Flowchart of HDDM adjustment methodology to inform risk and exposure
assessments.
We chose to adjust air quality for just meeting the existing and alternative standards by
decreasing U.S. anthropogenic emissions of NOx and 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 4-B. 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 equal proportional decreases in emissions
throughout the U.S., we were able to estimate how Os would respond to changes in ambient NOx
and VOC concentrations without simulating a specific control strategy. The model was set up to
track response in hourly Oj concentrations to these across-the-board changes in US
anthropogenic NOx and VOC emissions. 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
4-17
-------
1 across the domain, we allow for the possibility of contribution from both regional and local
2 emissions sources to nonattainment and to the overall distribution of 63 concentrations in urban
3 areas. The modeling included sources which contribute to background Os such as biogenic
4 emissions, wildfire emissions, and transport of Os and its precursors from international source
5 regions. In addition, the HDDM tool was set-up to specifically calculate the changes in 63 that
6 would occur from changes in US anthropogenic emissions alone, yet to account for the effects of
7 background sources on this response. Consequently, it is not necessary to set a "floor"
8 background 63 concentration as was done for quadratic rollback because background sources are
9 explicitly accounted for in the model estimates of Os response to US anthropogenic emissions.
10 As described in more detail in Appendix 4-D, the HDDM adjustment methodology
11 estimates hourly Os concentrations that would be associated with attaining a targeted level of the
12 standard either though reductions in US anthropogenic NOx emissions alone or through
13 reductions of both US anthropogenic NOx and VOC emissions in equal percentages. Because
14 the combined NOx/VOC cuts are constrained to equal percentage cuts of both precursors, this is
15 not an optimized NOx/VOC control scenario but rather a sensitivity analysis to characterize the
16 range of results that could be obtained with alternate assumptions. In most of the urban areas,
17 although the NOx/VOC scenario affected 03 response on some days, it did not affect 03
18 response at the highest design value (or controlling) monitor in such a way to reduce the total
19 required emissions cuts. However, for the two cities of Chicago and Denver, the NOx/VOC
20 scenarios allowed for lower percentage emissions cuts (applied to both NOx and VOC) to reach
21 targeted standard levels than the NOx only scenario. Because of this, the core analyses presented
22 in Chapters 5, 6, and 7 were based on the NOx only assumption for all cities except for Chicago
23 and Denver which used the NOx/VOC equal percentage reduction assumption. Sensitivity
24 analyses were performed to compare the NOx only and the NOx/VOC cases in 9 cities: Denver,
25 Detroit, Houston, Los Angeles, New York, Philadelphia, and Sacramento. The effects of these
26 sensitivity analyses on air quality and on the epidemiology-based risk assessment are discussed
27 in more detail in Appendix 4-D and Chapter 7, respectively.
28 For New York and Los Angeles it should also be noted that a somewhat different
29 approach was used for the HDDM-adjustment application. The HDDM adjustment methodology
30 produces estimates of hourly Os concentrations with standard error bounds for every potential
31 emission reduction scenario. Uncertainties in the application of the methodology to very large
32 emissions perturbations along with the fact that the mean estimate does not capture the
33 variability in modeled responses on similar days resulted in the inability of this methodology to
34 estimate Os distributions in these two cities which would meet lower alternative standard levels
35 (65 ppb for New York, 60 ppb for Los Angeles). This does not indicate that these two areas
36 would not be able to meet these lower standard levels in reality, but simply reveals the
4-18
-------
1 limitations of this adjustment methodology. Consequently for these two cities, we used the 95*
2 percent confidence interval lower bound estimate of hourly Os concentrations to capture a
3 scenario in which these cities could meet lower standard levels based on the range of responses
4 in Os concentrations to emissions reduction predicted by the model for each city (See Appendix
5 4-D for more details). Estimates of risk for these two cities for these alternative standards will be
6 significantly more uncertain, reflecting the use of the lower bound Os predictions.
7
4.3.3.2 Resulting Air Quality
8 The HDDM adjustment technique tended to have several effects on the distribution of air
9 quality values. First, adjusted hourly Os concentrations at night and during the morning rush-
10 hour tended to be higher than the recent observed concentrations (additional details are provided
11 in Appendix 4-D). The CMAQ model predicts that, in general, these times have NOx titration
12 conditions meaning that a reduction in NOx causes an increase in Os concentrations. The NOx
13 titration effect was most pronounced in urban core areas which have higher volume of mobile
14 source NOx emissions from vehicles than do the surrounding areas. Response of daytime
15 concentrations was more varied. In general, Os tended to increase on low days and decrease on
16 high days. However, specific monitors that were either always heavily VOC limited or always
17 heavily NOx limited showed consistent increases and decreases respectively regardless of
18 whether Os concentrations were high or low on a particular day. It should be noted that locations
19 which were heavily VOC limited tended to have much lower observed Os concentrations than
20 downwind areas. The tendency of the model to predict Os increases on lower concentration days
21 and decreases on higher concentration days also leads to more compressed Os distributions in the
22 HDDM adjustment cases. The variability in predicted daily Os concentrations decreased when
23 meeting lower standard levels. The following paragraphs summarize a comparison of Os
24 distributions from application of the quadratic rollback and HDDM adjustment approach for a
25 case where the existing standard is estimated to be met, characterize the distribution of
26 composite monitor Os values at different standard levels, and provide a discussion of the spatial
27 distribution of Os changes in several cities. More details and figures for other case-study areas
28 are provided in Appendix 4-D.
29 Figures 4-7 and 4-8 show a comparison of April-October composite monitor Os
30 distributions for recent conditions (2006-2008) and for meeting the existing standard using the
31 quadratic rollback technique versus the HDDM adjustment methodology. The composite monitor
32 values in these plots are based on the monitors included in the composite monitor from the
33 Zanobetti and Schwartz (2008) study which was used in the 1st draft REA and do not include all
34 monitors in the CBSA as used in the main Chapter 7 analysis. In general, the Os distribution in
4-19
-------
1 the HDDM adjustment case is shifted upward compared to the quadratic rollback case. The
2 upward shift is more pronounced in the lower parts of the 63 distribution. In all cities displayed
3 in Figure 4-7, the 25* percentile, median, and mean of the 8-hour daily maximum Os
4 concentrations are higher in the HDDM adjustment case than the quadratic rollback. In some
5 cities (Sacramento and St. Louis) the 75th percentile values appear approximately equivalent in
6 the two cases while in other cities the 75* percentile values are slightly higher in the HDDM
7 adjustment case. In Houston, the very highest portion of the 63 distribution is lower in the
8 HDDM adjustment case than in the quadratic rollback case but in many cities the upper parts of
9 the distributions for these two cases are roughly equivalent. Similar results are seen in the 2008-
10 2010 time period; however there are more cases during this time period where HDDM
11 adjustment and quadratic rollback have similar values in the upper half of the Os distribution. A
12 comparison of Figure 4-7 and 4-8 shows that there is some seasonality to this effect. The two
13 techniques appear to give very similar 8-hour daily maximum 63 composite monitor
14 distributions during the summer months (June-August) and most of the situations with higher Os
15 levels with the HDDM adjustment come from cooler, lower Os time periods (April, May,
16 September, and October). Although here we discuss composite monitor distributions based on
17 April-October, the risk analyses in Chapter 7 are based on the required O?, monitoring season,
18 which is longer than April - October for some cities. We expect that the Os increases shown for
19 spring and fall months here are also representative of the type of response in other "cool season"
20 months. The exceptions to this occur in Denver, Houston, New York and Los Angeles which
21 have higher composite monitor Oj values from the HDDM adjustment compared to quadratic
22 rollback even in the summer time period.
4-20
-------
Atlanta: Z & S, April-October, 2006-200B
Baltimore: Z & S, April-October, Z006-ZOOB
Boston: Z & S. April-October. Z006-Z008
Cleveland: Z S S, April-October, Z006-ZOOB
Denver: Z & S, April-October, Z006-Z008
Detroit: Z & S, April-October, Z006-ZOOB
1
5
i
ba
3
o
^ @
1
r T
SB 7
^^ quadratic rollback
^^ ddm adjustment
O
1
r-j-i
L2LJ
T
a
a —
a
OD
. .
fa -
Si -
D
D —
^^» quadratic rollback
^^~ ddm adjustment
S
^r s
i
base 75
Houston: Z & S, April-October, Z006-Z008
LosAngeles: Z S S, April-October, Z006-ZOOB
NewVork: Z & S. April-October. Z006-Z008
Philadelphia: Z & S, April-October, Z006-Z008
Sacramento: Z S S, April-October, Z006-ZOOB
SaintLouis: Z & S, April-October, Z006-ZOOB
2 Figure 4-7 Distributions of composite monitor 8-hour daily maximum Os concentrations from
3 ambient measurements (black), quadratic rollback (blue), and the HDDM
4 adjustment methodology (red) for meeting the existing standard. Values are based
5 on the Zanobetti & Schwartz study areas for April-October of 2006-2008.
4-21
-------
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. 2006-2008
Denver: Z & S, June-August, 2006-2008
Detroit: Z & S, June-August, 2006-2008
Houston: Z & S, June-August. 2006-2008
LosAngeles: Z & S, June-August, 2006-2008
NewYork: Z & S, June-August 2006-2008
« quadratic (oltoaek
• Odm adlustment
75
Philadelpnia: Z & S, June-August, 2006-2008
Sacramento: Z & S, June-August, 2006-2008
SaintLouis: Z & S, June-August, 2006-2008
base
2 Figure 4-8 Distributions of composite monitor 8-hour daily maximum Os concentrations
3 from ambient measurements (black), quadratic rollback (blue), and the
4 HDDM adjustment methodology (red) for meeting the existing standard.
5 Values are based on the Zanobetti & Schwartz study areas for June-August of
6 2006-2008.
4-22
-------
1
2 Figures 4-9 and 4-10 show "box-and-whisker" plots of the April-October composite
3 monitor daily maximum 8-hour Os concentration distributions for the 12 urban case study areas
4 evaluated in the epidemiology-based risk assessment; for recent air quality, and air quality
5 adjusted to meet the existing and potential alternative standards. Figure 4-9 shows values from
6 2007, while figure 4-10 shows values from 2009. Appendix 4-D contains additional plots
7 comparing the changes in the distribution of composite monitor values in each urban area due to
8 the air quality adjustments across varying spatial extents, season lengths, and years. In general,
9 the range of the composite monitor distributions decreased (i.e. the minimum value increased,
10 while the maximum value decreased) in all 12 urban case study areas as the air quality data were
11 adjusted to meet lower standard levels. However, the changes within the inter-quartile range of
12 these distributions (represented by the "boxes") varied in response to the model-based air quality
13 adjustments across the 12 urban areas. Three different types of responses are highlighted in the
14 boxplots for Atlanta, New York, and Houston.
15 The Atlanta boxplots provide an example of an urban area in which all but the lowest
16 composite monitor values decreased as the air quality data was adjusted to simulate compliance
17 with progressively lower levels of the standard. The upper tail of the distribution (represented by
18 the top whisker in each boxplot) decreased more quickly than the remainder of the distribution,
19 resulting in less total variability in the composite monitor values with each progressively lower
20 standard level. This type of response was also seen Sacramento and St. Louis, and to a lesser
21 extent in Baltimore, Denver, and Philadelphia.
22 In New York, the boxplots showed an initial increase in the 25* percentile and median
23 composite monitor values when the observed 63 concentrations were adjusted to meet the
24 existing standard. However, the median composite monitor value decreased relative to the
25 existing standard as O^ concentrations were adjusted to meet the 70 ppb standard, and both the
26 median and 25th percentile values decreased when air quality were further adjusted to meet the
27 65 ppb standard. When the air quality were adjusted to meet 65 ppb, the median and mean
28 composite monitor values were lower than under observed conditions. This type of response was
29 also observed in Cleveland, Detroit, and Los Angeles.
30 In Houston, the median composite monitor value also increased between observed air
31 quality and air quality adjusted to meet the existing standards. However, the pattern in Houston
32 differed from New York and other cities as air quality was further adjusted to reflect meeting the
33 potential alternative standards. The median value remained relatively constant relative to the
34 existing standard, while the 25th percentile values continued to increase. Thus, in Houston, the
35 air quality adjustments always resulted in a median composite monitor value higher than what
4-23
-------
1 was seen in the observed data. The composite monitor distributions in Boston also exhibited this
2 type of behavior.
4-24
-------
1
2
Atlanta
base
75 70 65
Baltimore
base 75 70 65
Boston
base 75 70 65
Cleveland
base 75
70
65
Denver
bill
base 75 70 65
Detroit
60
base 75 70 65
Houston
HI: i
base 75 70 65
LosAngeles
HI: i
base 75
70
65
NewYork
HLI
base 75 70 65
Philadelphia
HO
HO
HO
base 75
70
65
HO
base 75 70 65 60
Sacramento
base 75 70 65 60
SaintLouis
HO
4
5
6
1
8
9
10
Figure 4-9 Distributions of composite monitor 8-hour daily maximum values for the 12 urban
case study areas in the epidemiology-based risk assessment. Plots depict values
based on ambient measurements (base), and values obtained with the HDDM
adjustment methodology showing attainment of 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 show attainment of a 60
ppb standard in New York, so no boxplot is shown for that case.
4-25
-------
Atlanta
Denver
NewYork
base
75 70 65
Baltimore
60
base
75 70
Detroit
60
base 75 70 65
Philadelphia
60
base 75 70 65
Boston
CO
base 75 70 65
Houston
60
base 75 70 65 60
Sacramento
base
75 70 65
Cleveland
CO
base 75 70 65 60
LosAngeles
base 75 70 65
Saintl nins
base 75
70
CO
base 75
70
60
base 75
70
65
60
1
2 Figure 4-10 Distributions of composite monitor 8-hour daily maximum values for the 12
3 urban case study areas in the epidemiology-based risk assessment. Plots
4 depict values based on ambient measurements (base), and values obtained
5 with the HDDM adjustment methodology showing attainment of 75, 70, 65
6 and 60 ppb standards. Values shown are based on CBSAs for April-October
7 of 2009. Note that Detroit air quality was meeting 75 ppb in 2008-2010, and
8 the HDDM adjustment technique was not able to adjust air quality to show
9 attainment of a 60 ppb standard in New York, so no boxplots are shown for
10 those cases.
4-26
-------
1
2 The exposure modeling and the clinical-based risk assessments used spatially varying
3 surfaces of hourly Os concentrations estimated at the centroid of each census tract within the 15
4 urban case study areas. The maps in Figures 4-11, 4-12, and 4-13 depict the spatial distributions
5 of the 2006-2008 average 4th highest (top) and May - September mean (bottom) daily maximum
6 8-hour (MDA8) Os concentrations for 3 of the 15 urban case study areas; for observed air quality
7 (left), air quality adjusted to meet the existing standard (center), and air quality adjusted to meet
8 the 65 ppb alternative standard (right). Appendix 4-A contains additional maps of the observed
9 4* highest MDA8 and May - September mean MDA8 concentrations in all 15 urban case study
10 areas for 2006-2008 and 2008-2010. Appendix 4-D contains maps and related figures showing
11 the changes in air quality that resulted from the HDDM adjustments for just meeting the existing
12 standard, and just meeting the potential alternative standard of 65 ppb.
13 These maps portray the general pattern seen in all 15 urban case study areas for the 4th
14 highest concentrations, which decreased when observed air quality were adjusted to meet the
15 existing standard, and continued to decrease as the air quality were further adjusted to meet the
16 various alternative standards. The May-September average values also generally decreased in
17 suburban and rural areas surrounding the urban population center in all 15 areas. However, three
18 different types of general behavior which were seen in the seasonal average values near the
19 urban population centers, which are exemplified in Figures 4-11 (Atlanta), 4-12 (New York), and
20 4-13 (Houston).
21 In Atlanta, the observed May - September average were nearly constant across the entire
22 study area. The observed values decreased nearly uniformly across the entire study area when
23 observed air quality was adjusted to meet the existing standard, and continued to do so when air
24 quality was further adjusted to meet the alternative standard of 65 ppb. The magnitudes of these
25 decreases were slightly larger in suburban and rural areas than near the urban population center.
26 This type of behavior was also seen in Sacramento and Washington, D.C.
27 In New York, the observed May - September average values were lower near the urban
28 population center than in the surrounding suburban areas. When the observed air quality was
29 adjusted to meet the existing standard, the seasonal average values increased near the urban
30 population center and decreased in the suburban areas, so that the spatial pattern was reversed.
31 When air quality was further adjusted to meet the 65 ppb alternative standard, large area-wide
32 decreases in the seasonal average values were seen relative to the existing standard. While New
33 York represents one of the most extreme examples, similar behavior was observed in 7 other
34 urban areas: Baltimore, Cleveland, Dallas, Detroit, Los Angeles, Philadelphia, and St. Louis.
35 Houston started out in a similar fashion as New York. The observed May - September
36 average concentrations were lower near the urban population center than in the surrounding
4-27
-------
1
2
3
4
5
6
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 IV1DA8 - Base
93 7C R
4th Highest MDAS-75 ppb
4th Highest IV1DA8 - 65 ppb/
7
8
9
10
11
12
13
14
15
May - Sep mean IV1DA8 - 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-hour Os concentrations in Atlanta 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
-------
New York, 2006-2008
4th Highest MDAS - Base
4th Highest MDAS -75 pptr
sa TC *>
";4th Highest MDAS - 65 ppb~
70 0
1
2
3
4
5
May - Sep mean MDAS - Base
May - Sep mean MDAS - 75 ppb
May - Sep mean MDAS - 65 ppb
39 35
Figure 4-12 Maps showing the 4th highest (top) and May-September average (bottom)
daily maximum 8-hour 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.
4-29
-------
Houston, 2006 - 2008
V,4th Highest MDAS - Base } ^.4th Highest IVIDA8 - 75 ppb \ 4th Highest MDAS - 65 ppb
1
2
3
4
5
9
10
11
12
13
14
15
16
17
18
May - Sep mean MDAS - Base
May - Sep mean MDAS - 75 ppb
May - Sep mean MDAS - 65 ppb
Figure 4-13 Maps of 4th highest (top) and May-September average (bottom) daily
maximum 8-hour 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.
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 63 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
4-30
-------
1 model simulation, run for a 12 km domain covering the contiguous U.S. In the first draft of the
2 REA, the spatial surfaces were created using the enhanced Voronoi Neighbor Averaging (eVNA)
3 technique (Timin et al, 2010), using the EPA's Model Attainment Test Software (MATS; Abt
4 Associates, 201 Ob). In this draft, the spatial surfaces are created using EPA's Downscaler
5 software (Berrocal et al, 2012). More details on the ambient measurements, the 2007 CMAQ
6 model simulation, the Downscaler fusion technique, and a technical justification for changing
7 from eVNA to Downscaler can be found in Appendix 4-C.
8 Three national "fused" spatial surfaces were created for:
9 1) the May-September average of the 8-hour daily maximum O^ concentrations
10 (consistent with the metric used by Smith et al. 2009);
11 2) the June-August average of the daily 10am-6pm mean Os concentrations (consistent
12 with the metric used by Zanobetti and Schwartz 2008); and
13 3) the April-September average of the 1-hour daily maximum 63 concentrations
14 (consistent with the metric used by Jerrett et al 2009).
15 Figures 4-14 to 4-16 show the geographic distributions of these spatial surfaces. The
16 spatial distributions of these three surfaces are very similar, with the highest levels occurring in
17 Southern California for all three surfaces.
18
4-31
-------
1 20 30 40 50 60 70 80
2 Figure 4-14 May-September average 8-hour daily maximum Os concentrations in ppb,
3 based on a Downscaler fusion of 2006-2008 average monitored values with a
4 12km 2007 CMAQ model simulation.
4-32
-------
1 20 30 40 50 60 70 80
2 Figure 4-15 June-August average 8-hour daily 10am-6pm mean Os concentrations in
3 ppb, based on a Downscaler fusion of 2006-2008 average monitored values
4 with a 12km 2007 CMAQ model simulation.
4-33
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
20 30 40 50 60 70 80 90
Figure 4-16 April-September average 1-hour daily maximum 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 O^ 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 4-3, and correlation coefficients between the three metrics are given in Table 4-4.
The May-September average 8-hour daily maximum 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 average 1-hour
daily maximum concentrations were about 5 ppb higher on average than the May-September
average 8-hour daily maximum 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).
4-34
-------
1
2
3
4
5
6
7
8
Q)
3
IT
Q)
0.4
0.3
0.2
0.1
0
100%
I May-September average 8-hr daily maximum
I June-August average 8-hr daily mean 10am-6pm
I April-September average 1-hr daily maximum
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
0%
0
Concentration (ppb)
Figure 4-17 Frequency and Cumulative Distributions of the Three Fused Seasonal
Average Os Surfaces Based on all CMAQ 12 km Grid Cells.
4-35
-------
1
2
Table 4-3 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 average
8 -hour daily maximum
concentration (ppb)
21.8
43.6
43.2
54.3
76.1
June-August average daily
10am-6pm mean
concentration (ppb)
14.9
41.7
40.9
54.8
80.1
April-September average
1-hour daily maximum
concentration (ppb)
26.2
48.8
48.2
59.0
84.2
3
4
5
6
Table 4-4 Correlation Coefficients Between the Three Fused Seasonal Average
Surfaces Based on all CMAQ 12 km Grid Cells
Seasonal metrics compared
May-September average 8-hour daily maximum vs.
June-August average daily 10am-6pm mean
May-September average 8-hour daily maximum vs.
April-September average 1-hour daily maximum
June-August average daily 10am-6pm mean vs.
April-September average 1-hour daily maximum
Correlation coefficient
0.974
0.995
0.972
9
10
11
12
13
14
15
16
17
18
19
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 4-5 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 average 1-hour daily maximum was the most strongly correlated with the
design values (R = 0.75), followed by the May-September average 8-hour daily maximum (R =
0.71), and then the June-August average daily 10am-6pm mean (R = 0.69).
4-36
-------
1
2
3
4
5
6
80
70
-Q
Q.
a.
c
o
+J
ro
60
g
O
u
-------
1
2
Table 4-5 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 average
8 -hour daily maximum
0.71
1.1
1.3
1.5
1.6
2.0
2.4
June-August average
daily 10am-6pm mean
0.69
1.1
1.3
1.5
1.6
2.2
3.0
April-September average
1-hour daily maximum
0.75
1.0
1.2
1.4
1.4
1.6
1.9
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
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 4-6 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 O^ 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 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 63 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 63 response over the entire range of emissions levels. The accuracy
4-38
-------
1 of the HDDM estimates can be quantified at distinct emissions levels by re-running the model
2 with modified emissions inputs and comparing the results. This method was applied to quantify
3 the accuracy of 3-step HDDM Os estimates for 50% and 90% NOx cut conditions for each urban
4 case study areas (as shown in Appendix 4-D). At 50% NOx cut conditions, HDDM using
5 information from these multiple simulations predicted hourly Os concentrations with a mean bias
6 and a mean error less than +/- 1 ppb in all case study areas compared to brute force model
7 simulations. At 90% NOx cut conditions, HDDM using information from these multiple
8 simulations predicted hourly Oj concentrations with a mean bias less than +/- 3ppb and a mean
9 error less than +/- 4 ppb in all case study areas. These small bias and error estimates show that
10 uncertainty due to the HDDM approximation method is small up to 90% emissions cuts.
11 In order to apply modeled Os response to ambient measurements, regressions were
12 developed which relate O?, response to emissions perturbations with ambient Os concentrations
13 for every season, hour-of-the-day, and monitor location. Applying Oj responses based on this
14 relationship adds uncertainty. Preliminary work showed that the relationships developed with
15 these regressions were generally statistically significant for most season, hour-of-the-day, and
16 monitor location combinations for 2005 modeling in Detroit and Charlotte (Simon et al, 2012).
17 Statistical significance was not evaluated for each regression in this analysis since there were
18 over 460,000 regressions created (322 monitors x 5 sensitivity coefficients x 3 emissions levels
19 x 4 seasons x 24 hours = 463,680 regressions). Statistics can quantify the goodness of fit for the
20 modeled relationships and can quantify the uncertainty in response at any given O^ concentration
21 based on variability in model results at that portion of the distribution for each regression. The
22 regression model provided both a central tendency and a standard error value for Os response at
23 each measured hourly Oj, concentration. The base analysis in all case study areas except New
24 York and Los Angeles used the central tendency which will inherently dampen some of the
25 variability in 03 response. The standard error of each sensitivity coefficient was propagated
26 through the calculation of predicted Os concentrations at various standard levels. These standard
27 errors reflect the amount of variability that is lost due to the use of a central tendency. Since
28 emissions reductions increased for lower standard levels the standard errors were larger for
29 adjustments to lower standards. Mean (95th percentile) standard errors for the 75 ppb adjustment
30 case ranged from 0.13 (0.26) to 1.18 (2.87) ppb in the 15 case study areas. Mean (95* percentile)
31 standard errors for the 65 ppb adjustment case ranged from 0.54 (1.07) to 1.39 (2.98) ppb. The
32 largest standard errors occurred in Los Angeles and New York due to the large emissions
33 reductions applied in these cases. In cases where the use of the central tendency of response
34 reduced the total estimated emissions reductions required to achieve a given standard level, in
35 general we expect that the benefits of reducing high Os concentrations and the disbenefits of
36 increasing low O^ would both be underestimated. For the exposure assessment which estimates
4-39
-------
1 health outcomes that occur at O?, concentrations above 60, this would lead to an underestimation
2 of risks. For the epidemiology-based risk assessment which is effected by the entire range of 63
3 concentrations, the impact is undetermined since changes at both ends of the Os distribution in
4 opposite directions would affect the results. The opposite would be true in cases where the use of
5 the central tendency of response increased the total estimated emissions reductions required to
6 achieve a given standard. However, given the small standard error values even in the case study
7 areas with the greatest uncertainty (i.e. less than 1.5 ppb mean standard error), this source of
8 uncertainty is not expected to substantially impact results.
9 Relationships between O^ response and hourly O^ concentration were developed based on
10 8 months of modeling: January and April-October 2007. These relationships were applied to
11 ambient data from 2006-2010. Some locations monitor for months not included in this modeling
12 (i.e., February, March, November, and December) while others do not. Seasonal relationships
13 were developed between 63 response to emissions reductions and 63 concentration. Summer was
14 the only season for which modeling data was created for all months (June, July, August). The
15 winter relationships were developed based on January modeling, the spring relationships were
16 developed based on April/May modeling, and the autumn relationships were developed based on
17 September/October modeling. The reduction in data points (31 or 61 instead of-90) increases
18 uncertainty in the statistical fit for these seasons. In addition, the modeling generally showed
19 more OT, disbenefits to NOx decreases in cooler months. So applying April/May relationships to
20 March and September/October relationships to November could potentially underestimate 03
21 increases that would happen in those two months in the five case study areas which measure 63
22 during March and/or November: Dallas, Denver, Houston, Los Angeles, and Sacramento. The
23 eight months that were modeled capture a variety of meteorological conditions. In cases where
24 other years have more frequent occurances of certain types of meteorological conditions, the
25 regressions should be able to account for this. For instance, if a monitor only had 2-3 high 03
26 days associated with sunny, high pressure conditions in the 2007 modeling but had 30-40 of
27 those days in another year, the regression may be more uncertain at those high Os values but
28 should still be able to capture the central tendency which can be applied to the more frequent
29 occurances in other years. If, on the other hand, the meteorology/ 63 conditions in another year
30 were completely outside the range of conditions captured in the model, then the regression based
31 on modeled conditions might not be able to capture those conditions. Finally, if emissions
32 change drastically between the modeled period and the time of the ambient data measurements
33 this could also change the relationship between 03 response and 03 concentrations. The
34 regressions derived from the 2007 modeling period are only applied to measurements made
35 within 3 years of the modeled time period. Although some emissions changes did occur over this
4-40
-------
1 time period, we believe it is still reasonable to apply 2007 modeling to this relatively small
2 window of measurements which occurs before and after the modeling.
3 Os response is modeled for across-the-board reductions in U.S. anthropogenic NOx (and
4 VOC). These across-the-board cuts do not reflect actual emissions control strategies. The form,
5 locations, and timing of emissions reductions that would be undertaken to meet various levels of
6 the Os standard are unknown. The across-the-board emissions reductions bring levels down
7 uniformly across time and space to show how 63 would respond to changes in ambient levels of
8 precursor species but do not reflect spatial and temporal heterogeneity that may occur in local
9 and regional emissions reductions. In cases where VOC reductions were modeled, equal
10 percentage NOx and VOC reductions were applied in the adjustment methodology. Regional
11 NOx reductions are likely to be the primary means used to reduce high OT, concentrations at DV
12 monitors. In limited cases, VOC emissions reductions may also help lower high O?,
13 concentrations at these locations. In actual control strategies, NOx and VOC reductions may be
14 applied in combination but are unlikely to be applied in equal percentages. The available
15 modeling constrained the NOx/VOC case to this type of control scenario. The across-the-board
16 cuts and the equal percentage NOx and VOC reductions scenario does not optimize the lowest
17 cost or least total emissions combinations as state and local agencies will likely attempt to
18 achieve.
4-41
-------
1 Table 4-6 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. 03
measurements
63 concentrations
measured by ambient
monitoring instruments
have inherent
uncertainties associated
with them. Additional
uncertainties due to other
factors may include:
- monitoring
network locations
- OT, monitoring
seasons
- monitor
malfunctions
- wildfire and
smoke impacts
- interpolation of
missing data
Both
Low
Low
KB: 63 measurements are assumed to be
accurate to within 1A 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. The CASTNET monitoring
data were subject to their own set of QA
requirements, and these data are generally
believed to be of comparable quality to the data
stored in AQS.
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
63 instruments. Measurements collected by 63
analyzers were reported to be biased high by 5.1-
6.6 ppb per 100 |ig/m3 of PM2.5 from wildfire
smoke ,EPA, 2007). However, smoke
concentrations high enough to cause significant
interferences are infrequent and the overall impact
is believed to be minimal.
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)
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 63 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 63 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 63 concentrations in
that area, which may cause some bias toward
higher measured concentrations.
INF: Each state has a required OT,
monitoring season which varies in length from
May - September to year-round. Some states turn
their 03 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 QI monitoring is required.
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)
B. Veronoi
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
63 concentrations to provide estimates of 63
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 63
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 03 monitors.
C.CMAQ
modeling
Model predictions from
CMAQ, like all
deterministic
photochemical models,
have both parametric and
structural uncertainty
associated with them
Both
Low-
Medium
Low-
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
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)
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 4-B 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 occur
on low 63 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 63
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
Both
Low-
Medium
Low-
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 63 response over the entire
range of emissions levels. The replication of brute
force hourly O^ concentration model results by the
HDDM approximation was quantified for 50%
and 90% NOx cut conditions for each urban case
study areas (as shown in Appendix 4-D). At 50%
NOx cut conditions, HDDM using information
from these multiple simulations predicted hourly
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)
especially under nonlinear
chemistry conditions.
Os concentrations with a mean bias and a mean
error less than +/- 1 ppb in all urban case 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 case study areas.
E. Application of
HDDM
sensitivities to
ambient data
In order to apply
modeled sensitivities to
ambient measurements,
regressions were
developed which relate 63
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 x 5 sensitivity coefficients x 3 emissions
levels x 4 seasons x 24 hours = 463,680
regressions). Statistics can quantify the goodness
of fit for the modeled relationships and can
quantify the uncertainty in response at any given
Os concentration based on variability in model
results at that portion of the distribution for each
regression. 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 and a standard error value for Os
response at each measured hourly Os
4-46
-------
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)
concentration. The base analysis used the central
tendency which will inherently dampen some of
the variability in 63 response. The standard error
of each sensitivity coefficient was propagated
through the calculation of predicted Os
concentrations at various standard levels. These
standard errors reflect the amount of variability
that is lost due to the use of a central tendency.
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 in the 15 case 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. The largest standard errors occurred in
Los Angeles and New York.
INF: The NOx emissions reductions
resulted in both increases and decreases in 03
depending on the time and location. In cases
where the use of the central tendency of response
reduced the total estimated emissions reductions
required to achieve a given standard level, in
general we expect that the benefits of reducing
high 63 concentrations and the disbenefits of
increasing low 63 would be underestimated. For
the exposure assessment which estimates health
outcomes that occur at OT, concentrations above
4-47
-------
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)
60, this would lead to an underestimation of risks.
For the epidemiology-based risk assessment which
is effected by the entire range of 63
concentrations, the impact is undetermined since
changes at both ends of the Os distribution in
opposite directions would affect the results. The
opposite would be true in cases where the use of
the central tendency of response increased the total
estimated emissions reductions required to achieve
a given standard.
F. Applying
modeled
sensitivities to un-
modeled time
periods
Relationships
between Os response and
hourly 63 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
Low-
Medium
Low-
Medium
KB: The eight months that were modeled
capture a variety of meteorological conditions. In
cases where other years have more frequent
occurances of certain types of conditions, the
regressions should be able to account for this. For
instance, if a monitor only had 2-3 high O3 days
associated with sunny, high pressure conditions in
the 2007 modeling but had 30-40 of those days in
another year, the regression may be more
uncertain at those high O3 values but should still be
able to capture the central tendency which can be
applied to the more frequent occurances in other
years. If, on the other hand, the meteorology/O3
conditions in another year were completely
outside the range of conditions captured in the
model, then the regression based on modeled
conditions might not be able to capture those
conditions.
KB: If emissions change drastically
4-48
-------
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)
between the modeled period and the time of the
ambient data measurements this could also change
the relationship between O3 response and O3
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 O3 response to emissions
reductions and O3 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 O3
disbenefits to NOx decreases in cooler months. So
applying April/May relationships to March and
September/October relationships to November
could potentially underestimate O3 increases that
would happen in those two months in the five
4-49
-------
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)
urban case study areas which measure O3 during
March and/or November: Dallas, Denver,
Houston, Los Angeles, and Sacramento.
G. Assumptions
of across-the-
board emissions
reductions
63 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.
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 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 Oj,
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
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-
4-50
-------
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)
surface. Uncertainties may
occur in sparsely
monitored regions, or in
urban areas with dense
monitoring networks and
large spatial gradients.
validation analysis in Appendix 4-A 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 4-
A 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 umonitored
areas.
1 * Refers to the degree of uncertainty associated with our understanding of the phenomenon, in the context of assessing and
2 characterizing its uncertainty. Sources classified as having a "low" impact would not be expected to impact the interpretation of risk
3 estimates in the context of the Os NAAQS review; sources classified as having a "medium" impact have the potential to change the
4 interpretation; and sources classified as "high" are likely to influence the interpretation of risk in the context of the 63 NAAQS
5 review.
4-51
-------
1 4.6 REFERENCES
2 Abt Associates, Inc. 2010a. "Environmental Benefits and Mapping Program (Version 4.0)."
3 Bethesda, MD. Prepared for U.S. Environmental Protection Agency Office of Air Quality
4 Planning and Standards. Research Triangle Park, NC. Available on the Internet at
5 .
6 Abt Associates, Inc. 2010b. "Model Attainment Test Software (Version 2)." Bethesda, MD.
7 Prepared for the U.S. Environmental Protection Agency Office of Air Quality Planning
8 and Standards. Research Triangle Park, NC. Available on the Internet at:
9 http://www.epa.gov/scram001/modelingapps.mats.htm.
10 Bell, M.L.; A. McDermott; S.L. Zeger; J.M. Samet and F. Dominici. 2004. "Ozone and Short-
11 term Mortality in 95 U.S. Urban Communities," 1987-2000. JAMA, 292:2372-2378.
12 Berrocal, V.J.; A. E. Gelfand and D.M. Holland. 2012. "Space-Time Data Fusion Under Error in
13 Computer Model Output: An Application to Modeling Air Quality." Biometrics, 68(3),
14 837-848.
15 Chen, J. R.; Zhao; Z. Li. 2004. "Voronoi-based k-order Neighbor Relations for Spatial Analysis."
16 ISPRS JPhotogrammetry Remote Sensing, 59(1-2), 60-72.
17 Duff, M.; R. L. Horst; T.R. Johnson, 1998. "Quadratic Rollback: A Technique to Model
18 Ambient Concentrations Due to Undefined Emission Controls." San Diego, CA:
19 Presented at the Air and Waste Management Annual Meeting, June 14-18, 1998.
20 Fann, N.; A.D. Lamson; S.C. Anenberg; K. Wesson; D. Risley; B.J. Hubbell. 2012. "Estimating
21 the National Public Health Burden Associated with Exposure to Ambient PM2.5 and
22 Ozone." Risk Analysis, 32:81-95.
23 Gold, C. 1997. "Voronoi Methods in GIS," Vol. 1340. In Algorithmic Foundation of Geographic
24 Information Systems (Kereveld M., J. Nievergelt, T. Roos, P. Widmayer eds). Lecture
25 notes in Computer Science, Berlin: Springer-Verlag, 21-35.
26 Hall, E.; A. Eyth; S. Phillips. 2012. "Hierarchical Bayesian Model (HBM)-Derived Estimates of
27 Air Quality for 2007: Annual Report." (EPA document number EPA/600/R-12/538).
28
-------
1 Jerrett, M.; R.T. Burnett; C.A. Pope, III; K. Ito; G. Thurston; D. Krewski; Y. Shi, E. Calle and
2 M. Thun. 2009. "Long-term O3 Exposure and Mortality." N. Eng. J. Med., 360:1085-
3 1095.
4 Johnson, T. 2002. "A Guide to Selected Algorithms, Distributions, and Databases Used in
5 Exposure Models Developed by the Office of Air Quality Planning and Standards,"
6 prepared by TRJ Environmental, Inc. for the U.S. Environmental Protection Agency.
7 Research Triangle Park, NC: Office of Research and Development.
8 National Research Council of the National Academies. 2008. "Estimating Mortality Risk
9 Reduction and Economic Benefits from Controlling Ozone Air Pollution." Washington,
10 DC: The National Academies Press.
11 Rizzo, M. 2005. "A Comparison of Different Rollback Methodologies Applied to Ozone
12 Concentrations," posted on November 7, 2005,
13 http ://www. epa.gov/ttn/naaq s/standards/ozone/s_O^_cr_td.html.
14 Rizzo, M. 2006. "A Distributional Comparison between Different Rollback Methodologies
15 Applied to Ambient Ozone Concentrations," posted on May 31, 2006,
16 http ://www. epa.gov/ttn/naaq s/standards/ozone/s_O^_cr_td.html.
17 Simon, H.; K. Baker; N. Possiel; F. Akhtar; S. Napelenok; B. Timin; B. Wells. 2012. "Model-
18 based Rollback Using the Higher Order Direct Decoupled Method (HDDM)," posted at
19 < http://www.epa.gOv/ttn/naaqs/standards/ozone/s O^td.html>.
20 Simon, H.; K. R. Baker; F. Akhtar; S.L. Napelenok; N. Possiel; B. Wells and B. Timin. 2013. "A
21 Direct Sensitivity Approach to Predict Hourly Ozone Resulting from Compliance with
22 the National Ambient Air Quality Standard" Environmental Science and Technology,
23 Vol. 47, 2304-2313.
24 Smith, R.L., B. Xu, P. Switzer. 2009b. "Reassessing the Relationship Between Ozone and Short-
25 term Mortality in U.S. Urban Communities," Inhale Toxicol, Vol. 21: 37-61.
26 Timin, B.; K. Wesson and J. Thurman. 2010. "Application of Model and Ambient Data Fusion
27 Techniques to Predict Current and Future Year PM2.5 Concentrations in Unmonitored
28 Areas, " in D.G. Steyn and St Rao (eds), Air Pollution Modeling and Its Application XX,
29 Netherlands: Springer, pp. 175-179.
4-53
-------
1 U.S. Environmental Protection Agency. 2007. Review of the National Ambient Air Quality
2 Standards for Ozone: Policy Assessment of Scientific and Technical Information OAQPS
3 Staff Paper. Washington, DC: EPA Office of Air and Radiation. (EPA document number
4 U. S. EPA. 2012a. Integrated Science Assessment for Ozone and Related Photochemical
5 Oxidants: Third External Review Draft. Research Triangle Park, NC: EPA Office of Air
6 Quality Planning and Standards. (EPA document number EPA-452/R-07-007;
7 EPA/600/R-10/076C).
8 U.S. EPA. 2012b. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
9 Documentation (TRIM.Expo / APEX, Version 4.4) Volume I: User's Guide. Research
10 Triangle Park, NC: Office of Air Quality Planning and Standards. (EPA document
11 number EPA-452/B-12-001a). .
12 U.S. EPA. 2012c. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
13 Documentation (TRIM.Expo / APEX, Version 4.4) Volume II: Technical Support
14 Document. Research Triangle Park, NC: Office of Air Quality Planning and Standards.
15 (EPA document number EPA-452/B-12-001b).
16 .
17 Wells, B.; K. Wesson and S. Jenkins. 2012. "Analysis of Recent U.S. Ozone Air Quality Data to
18 Support the O3 NAAQS Review and Quadratic Rollback Simulations to Support the First
19 Draft of the Risk and Exposure Assessment."
20 .
21 Zhang, L.; D.J. Jacob; N.V. Smith-Downey; D.A. Wood; D. Blewitt; C.C. Carouge; A. van
22 Donkelaar; D.B. A. Jones; L.T. Murray and Y. Wang. 2011. "Improved Estimate of the
23 Policy-relevant Background Ozone in the United States Using the GEOS-Chem Global
24 Model with l/2°x2/3° Horizontal Resolution Over North America." Atmos Environ, Vol.
25 45, pp. 6769-6776.
26 Zanobetti, A. and J. Schwartz. 2008. "Mortality Displacement in the Association of Ozone with
27 Mortality: An Analysis of 48 Cities in the United States." American Journal of
28 Respiratory and Critical Care Medicine, 177:184-189.
29
4-54
-------
1
2 5 CHARACTERIZATION OF HUMAN EXPOSURE TO OZONE
3 5.0 OVERVIEW
4 As part of the previous 2007 O^ NAAQS review, EPA staff conducted exposure analyses
5 for the general population, all school-age children (ages 5-18), all active school-age children,1
6 and asthmatic school-age children (U.S. EPA, 2007a,b). Exposure estimates were generated for
r\
1 12 urban study areas for recent years of air quality and for just meeting the existing 8-hr
8 standard and several alternative 8-hr standards. EPA also conducted a health risk assessment that
9 produced risk estimates for the number and percent of all school-age children experiencing
10 impaired lung function and other respiratory symptoms associated with the exposures estimated
11 for these same 12 study areas.
12 The exposure analysis conducted for this current NAAQS review builds upon the
13 methodology and lessons learned from the exposure analyses conducted in previous 63 reviews
14 (U.S. EPA, 1996a, 2007a,b) and information provided in the final ISA (U.S. EPA, 2013). Here,
15 we estimate exposures for people residing in 15 urban study areas in the U.S.3 The population
16 exposures to ambient 63 concentrations were modeled using EPA's Air Pollutants Exposure
17 (APEX) (US EPA, 2012a,b). Exposures were calculated considering Os concentrations in recent
18 years, using 2006 to 2010 spatially interpolated ambient monitoring data. Exposures were also
19 estimated considering alternative air quality scenarios, that is, where 63 concentrations just meet
20 the existing 8-hr 03 NAAQS and at several other standard levels considering the same indicator,
21 form, and averaging time, based on adjusting data as described in Chapter 4. Exposures were
22 modeled for 1) all school-age children (ages 5-18), 2) asthmatic school-age children (ages 5-18),
23 3) asthmatic adults (ages 19-95), and 4) all older adults (ages 65-95), each while at moderate or
24 greater exertion level at the time of exposure.4 The strong emphasis on children, asthmatics, and
25 older adults reflects the finding of the last O3 NAAQS review (U.S. EPA, 2007a) and the ISA
26 (U.S. EPA, 2013, Chapter 8) that these are important at-risk groups. Exposure model output of
27 interest for this chapter are the percent (and number) of persons exposed at or above 8-hr average
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 (PAT) 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 DC (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 DC) in this current assessment. Inclusion
of Seattle, WA was considered but not included due to a lack of appropriate monitoring data.
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.
5-1
-------
1 Os concentrations of concern, all while at moderate or greater exertion levels, based on adverse
2 effects observed in human clinical exposure studies. Further, the complete time series of
3 individual exposures estimated by APEX serves as input to a module that estimates human health
4 risk (Chapter 6).
5 This chapter first provides a brief overview of human exposure and exposure modeling
6 using APEX (section 5.1), the scope of this Os exposure assessment and key inputs used to
7 model exposure in the 15 U.S. study areas selected (section 5.2), and followed by the main body
8 exposure results (section 5.3). Then, section 5.4 presents an assemblage of targeted analyses
9 designed to provide additional insight to the main body of exposure results by focusing on
10 important data inputs, additional at-risk populations, lifestages, or scenarios, influential attributes
11 in estimating exposures, and performance evaluations. The results of these and other exposure
12 model targeted analyses are integrated in an uncertainty characterization section (section 5.5)
13 along with a final section summarizing the key observations for this chapter (section 5.6).
14 5.1 SYNOPSIS OF O3 EXPOSURE AND EXPOSURE MODELING
15 5.1.1 Human Exposure
16 Human exposure to a contaminant is defined as "contact at a boundary between a human
17 and the environment at a specific contaminant concentration for a specific interval of time," and
18 has units of concentration times duration (National Research Council, 1991). For air pollutants
19 the contact boundary is nasal and oral openings in the body, and personal exposure of any
20 individual to a chemical in the air for a discrete time period is fundamentally quantified as (Lioy,
21 1990; National Research Council, 1991):
22 (5-1)
23 £[/,,/,] =f'C(0<*
24 J'i
25 where E[^/2] is the personal exposure or exposure concentration during the time period
26 from t\ to t2, and C(t) is the concentration at time t in the breathing zone. The breathing rate at
27 the time of exposure will influence the dose received by the individual. While we do not directly
28 estimate dose in this assessment, intake is the total 63 inhaled (i.e., exposure concentration,
29 duration, and ventilation combined).5
30
31
' 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.
5-2
-------
1 5.1.2 Estimating Os Exposure
2 Exposure to 63 can be directly estimated by monitoring the concentration of 63 in a
3 person's breathing zone (close to the nose/mouth) using a personal exposure monitor. Studies
4 employing this measurement approach have been reviewed in the ISA and EPA 03 Air Quality
5 Criteria Documents (U.S. EPA, 1986, 1996b, 2006, 2013). Personal exposure measurements
6 from these studies are useful in describing a general range of exposure concentrations (among
7 other reported measurement data) and in identifying factors that may influence varying exposure
8 levels. However, these measurement studies are largely limited by the disparity between sample
9 measurement duration and exposure concentration averaging-times of interest and in
10 appropriately capturing variability in population exposure occurring over large geographic areas ,
11 particularly when considering both concentration (e.g., spatial variability) and population (e.g.,
12 age, sex) attributes that influence exposure.
13 Oj, exposure for individuals, small groups of individuals or large populations can be
14 calculated indirectly (or modeled) using Equation 5-1. When employing such an approach in a
15 population exposure assessment, two basic types of input data are needed; a time-series of 03
16 concentrations that appropriately represents spatial heterogeneity in 63 concentrations and a
17 corresponding time-series of locations visited by the persons exposed. When considering air
18 pollutant concentrations, population exposure models are commonly driven by ambient
19 concentrations. These ambient concentrations may be provided by monitoring data, by air quality
20 model estimates, or perhaps by a combination of these two data sources. Then, an understanding
21 of the relationships between ambient pollutants and the locations people occupy is needed. This
22 is because human exposure, regardless of the pollutant or whether one is interested in individual
23 or population exposure, depends on where an individual is located, how long they occupy that
24 location, and what the pollutant concentration at the point of contact is. Furthermore, if interested
25 in air pollutant intake rate or dose, one needs to know what activity the person is performing
26 while exposed.
27 Thus, the types of measurement and modeling studies that provide information for more
28 realistically estimating exposure to O?, can be augmented from the above list to include studies
29 of: 1) 63 formation, deposition, and decay, 2) people's locations visited and activities performed,
30 3) human physiology, and 4) local scale meteorological measurements and/or modeling. Useful
31 data derived from these varied studies are 03 concentrations (i.e., fixed site, personal exposure,
32 indoor and outdoor locations), built environment physical factors (i.e., air exchange rates
33 (AERs), infiltration rates, decay and deposition rates), human time-location-activity patterns
34 (minute-by-minute, hourly, daily, and longer-term), time-averaged or activity-specific breathing
35 rates among varying sexes and/or lifestages, and hourly ambient temperatures.
5-3
-------
1 When integrating these varied data (among others such as population demographics and
2 disease prevalence) and understanding factors affecting exposure, exposure models can extend
3 beyond the limited information given by measurement data alone. For example, an exposure
4 model can reasonably estimate exposures for any perceivable at-risk population (e.g., asthmatics
5 living in a large urban area) and considering any number of hypothetical air quality conditions
6 (e.g., just meeting a daily maximum 8-hr average concentration of 70 ppb). Exposure models that
7 account for variability in human physiology can also realistically estimate pollutant intake by
8 using activity-specific ventilation rates. These types of measurements cannot realistically be
9 performed for a study group or population of interest, particularly when considering time, cost,
10 and other constraints. The following section provides an overview of how such exposure
11 modeling can be done using APEX, the model developed by EPA to perform such calculations
12 and used to estimate O^ exposures in this REA.
13 5.1.3 Modeling O3 Exposure Using APEX
14 EPA has developed the APEX model for estimating human population exposure to
15 criteria and air toxic pollutants, used most recently in estimating exposures for the Os (U.S. EPA,
16 2007b), nitrogen dioxide (U.S. EPA, 2008), sulfur dioxide (U.S. EPA, 2009a), and carbon
17 monoxide (U.S. EPA, 2010) NAAQS reviews. APEX is a probabilistic model designed to
18 account for the numerous sources of variability that affect people's exposures. An overview of
19 the approaches used by APEX to estimate exposure concentrations is found in Appendix 5 A with
20 details provided in U.S. EPA (2012a,b).
21 Briefly, APEX simulates the movement of individuals through time and space and
22 estimates their exposure to a given pollutant while occupying indoor, outdoor, and in-vehicle
23 locations. The model stochastically generates simulated individuals in selected study areas using
24 census-derived probability distributions for demographic characteristics. Population
25 demographics are drawn from the 2000 Census data6 at a tract level, and a national commuting
26 database based on 2000 Census data provides home-to-work commuting flows between tracts.7
27 Any number of individuals can be simulated, and collectively they approximate a random
28 sampling of people residing in a particular study area.
29 Daily activity patterns for individuals in a study area, an input to APEX, are obtained
30 from detailed daily time-location-activity pattern survey data that are compiled in the
31 Consolidated Human Activity Database (CHAD) (McCurdy et al., 2000; U.S. EPA, 2002). These
32 daily diaries are used to construct a sequence of locations visited and activities performed for
33 APEX simulated individuals consistent with their demographic characteristics, day-type (e.g.,
' Due to resource limitations and data availability, the 2010 Census data have not yet been processed to include in
this 2nd draft REA.
There are approximately 65,400 census tracts in the -3,200 counties in the U.S.
5-4
-------
1 weekend or weekday), and season of the year, as defined by ambient temperature regimes
2 (Graham and McCurdy, 2004). The time-location-activity data input to APEX are linked with
3 personal attributes of the surveyed individuals' such as age, sex, employment status, day-of-
4 week surveyed, and daily maximum and daily mean temperature. These specific personal
5 attribute data are then used by APEX to best match the daily diary with the simulated persons of
6 interest, using the same variables as first-order diary selection characteristics. The approach is
7 designed to capture the important attributes contributing to an individuals' time-location-activity
8 pattern, and of particular relevance here, time spent outdoors (Graham and McCurdy, 2004). In
9 using a diverse collection of time-location-activity diaries that capture the duration and
10 frequency of occurrence of visitations/activities performed, APEX can simulate expected
11 variability in human behavior, both within and between individuals. This, combined with
12 exposure concentrations, allows for the reasonable estimation of the magnitude, frequency,
13 pattern, and duration of exposures an individual experiences.
14 A key concept in modeling exposure using APEX is the microenvironment, a term that
15 refers to the immediate surroundings of an individual at a particular time. APEX has a flexible
16 approach for modeling micro-environmental concentrations whereas the model user defines the
17 type, number and characteristics of the microenvironments to be modeled. Typical
18 microenvironments include indoors at home, indoors at school, near roadways, inside cars, and
19 outside home. In this exposure assessment, all microenvironmental Os concentrations are
20 derived from ambient Os concentrations input to APEX and are estimated using either a mass-
21 balance or transfer factors approach, selected by the user. The mass balance approach assumes
22 that the air in an enclosed microenvironment is well-mixed and that the air concentration is
23 spatially uniform at a given time within the microenvironment. The approach employs indoor-to-
24 outdoor AERs (i.e., number of complete air exchanges per hour) and considers removal
25 mechanisms such as deposition to building surfaces and chemical decay rates. The transfer
26 factors model is simpler than the mass balance model, and employs two variables, a proximity
27 factor, used to account for proximity of the microenvironment to sources or sinks of pollution, or
28 other systematic differences between concentrations just outside the microenvironment and the
29 ambient concentrations, and a penetration factor, which quantifies the degree to which the
30 outdoor air penetrates into the microenvironment.
31 Activity-specific simulated breathing rates of individuals are used in APEX to
32 characterize intake received from an exposure. This is done because controlled human exposure
33 studies have shown adverse health outcomes are associated with both elevated concentrations
34 and study participant exertion levels. The breathing rates calculated by APEX are derived from
35 the energy expenditure associated with each simulated persons' activity performed, adjusted for
36 age- and sex-specific physiological parameters (Graham and McCurdy, 2005). The energy
5-5
-------
1 expenditure estimates themselves are derived from distributions of METS (or metabolic
2 equivalents of work) associated with every activity performed (McCurdy et al., 2000, using
3 Ainsworth et al., 1993).
4 An important feature of APEX is the ability to account for variability in exposure by
5 representing input variables as statistical distributions along with dependent conditional
6 variables, where appropriate. For example, the distribution of AERs in a home, office, or motor
7 vehicle can depend on the type of heating and air conditioning present, which are also stochastic
8 inputs to the model, as well as the ambient temperature on a given day. The user can choose to
9 keep the value of a stochastic parameter constant for the entire simulation (appropriate for the
10 volume of a house), or can specify that a new value shall be drawn hourly, daily, or seasonally
11 from specified distributions.
12 Finally, APEX calculates a unique time-series of exposure concentrations on the order of
13 minutes or smallest diary event duration that each simulated person may experience during the
14 modeled time period, based in that individual's estimated microenvironmental concentrations
15 and the time spent in each of sequence of microenvironments visited according to the time-
16 location-activity diary of each individual. Then, hourly average exposures of each simulated
17 individual are estimated using time-weighted averages of the within-hour exposures. From
18 hourly exposures, APEX calculates any other time averaged exposure of interest (e.g., 8-hr or
19 daily average) that a simulated individual experiences during the modeled period.
20 5.2 SCOPE OF THE EXPOSURE ASSESSMENT
21 This section broadly presents the scope of the exposure assessment including descriptions
22 of the modeling domains, ambient concentrations used, time periods and populations modeled, as
23 well as identifying key approaches, inputs and outputs used by APEX in estimating population
24 63 exposures. Detailed descriptions regarding APEX modeling, model inputs and other
25 supporting information are provided in Appendix 5A-5E and the APEX user's guide and
26 technical support documents (U.S. EPA 2012a,b). Figure 5-1 illustrates the general conceptual
27 framework for generating our population exposure concentrations, including the time series of
28 exposure and ventilation rate output generated as input to population risk calculations in Chapter
29 6.
30 5.2.1 Urban Areas Selected
31 The selection of urban areas to include in the exposure assessment considered the
32 location of 63 epidemiological studies, the availability of ambient 63 monitoring data, and the
! 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
-------
1 desire to represent a range of geographic areas, encompassing variability in climate and
2 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
O3 NAAQS review.
• The locations should be focused on areas that do not meet or are close to not meeting the
existing 8-hr O3 NAAQS and should include areas having O3 non-attainment
designations.
• There must be sufficient O3 ambient air quality data for the recent 2006-2010 period.
• The study areas should include the 12 cities modeled in the epidemiologic-based risk
assessment (Chapter 7).
3 Based on these criteria, we chose the 15 study areas listed in Table 5-1 to develop our
4 population exposure estimates. We then defined an air quality domain for each study area,
5 broadly bounding the ambient concentration field where exposures were to be estimated. To do
6 this, we evaluated 1) counties modeled in the previous 2007 O3 NAAQS review common to
7 current study areas, 2) political/statistical county aggregations (e.g., whether in a metropolitan
8 statistical areas or MS As), and 3) if the study area was designated as a non-attainment area
9 (NAA), the counties that were part of the NAA list. We identified a final list of 215 counties9 to
10 comprise the air quality domain for the 15 study areas, the names of which are provided in
11 Appendix 5B.
12 5.2.2 Time Periods Simulated
13 The exposure periods modeled are the O3 seasons for which routine hourly O3 monitoring
14 data were available for years 2006 to 2010 (Table 5-1), and defined by 40 CFR part 58,
15 Appendix D, Table D-3. These periods are designed to reasonably capture year-to-year
16 variability in ambient concentrations and meteorology and include most of the high
17 concentration events occurring in each area. Having this wide range of air quality data across
18 multiple years allows us to more realistically estimate a range of exposures, rather than using a
19 single year of air quality data. While the number of available O3 monitors may vary slightly from
20 year to year, we assumed constant representation by the available monitors and associated
21 statistically interpolated data for each year over the simulation period (see section 5.2.3).
9 Of the 215 counties defining the air quality domain, 207 remained in the exposure model domain.
5-7
-------
Population Information !
US Census Tract
Population
Distributions
V
f A
US Census Tract i
Home-to-Work
Commuting |
r
US Census Tract
i Asthma Prevalence
V /
US Census Tract
Employment
Distributions
V x
i
Individual Information I
Daily Human Time
Location Activity
Patterns (CHAD)
Metabolic Equivalents
of Work (METS)
Distributions
^ Distributions of
Physiological
Attributes & Ventilation
V Equations
•,
i
i
' i
V
1
1
i
(Chapter 4)
( Census Tract Hourly
03 Concentrations:
recent conditions and
adjusted to just meet
existing and
v alternative standards j
ivieieuiuiuyy
1 • !
• i Urban Area Hourly
1 j Temperatures (ISH)
i - I )
i
i
i
APEX |
Calculate 03 Exposure
Individuals Comprising ,
Population
I
/ Number and percent of\
/ persons with 8-hour \
f average exposures at or \
above selected
1 benchmark level per 1
V year while at moderate /
X^ or greater exertion y
I
•
. !
Microenvironmental Information
^~ In-Vehicle and Near- "^j
Road Proximity and
Penetration Factor
•^ Distributions )
f
r '
Urban Area Vehicle j
! Miles Traveled
(US DOT)
i V S
f \ i
Indoor Air Exchange
Rate (AER) and 03
Decay Distributions
J
T
f ~i
Urban Are a Air
Conditioning j
1 Prevalence (AHS)
f' Number of persons with i\
/ multiple 8-hour average \
/ exposures at or above i
| selected benchmark
1 level per year while at 1
\ moderate or greater /
\ exertion / t,
Hnffis:
/" ~^
/ Time-series of 03 exposure
{ concentrations and ventilation rate for
\ individuals
X^_ _^-
AHS: American Housing Survey
x CHAD: US EPA Consolidated Human
\ Activity Database
J ISH: National Climatic Data Center
Integrated Surface Hourly Database
US DOT: US Department of Transportation
Federal Highway Administration
Output to FEV1 Modeling
Figure 5-1 Conceptual Framework Used for Estimating Study Area Population Os Exposure 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
Apr1-Sep30
Apr1-Oct31
Apr1-Oct31
Mar1-Oct31
Mar 1-Sep30
Apr1-Sep30
Jan 1 -Dec 31
Jan 1 -Dec 31
Apr1-Oct31
Apr1-Oct31
Jan 1 -Dec 31
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
Persons
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
4 : Each study area's O3 monitoring season is defined by 40 CFR part 58, Appendix D, Table D-3.
5 5.2.3 Ambient Concentrations Used
6 We used the available hourly ambient monitor concentration data within and around each
7 study area along with a statistical interpolation technique (Chapter 4) to estimate hourly census
8 tract concentrations within the counties comprising each study area. These concentrations served
9 as the 'base' air quality input for each study area year. Ambient concentrations were also
10 adjusted to just meet the existing standard (75 ppb, 4* highest 8-hr average, averaged over a 3-
11 year period) and alternative standard levels (70, 65, 60, and 55 ppb) using an air quality model
12 and the statistical interpolation technique (Chapter 4).
13 These estimated hourly census tract O?, concentrations served as the APEX air districts,
14 the basic ambient concentrations from which each simulated persons microenvironmental
15 concentrations are estimated. Having these temporally and spatially resolved air districts in each
5-9
-------
1 study area allows for better utilization of APEX spatial and temporal capabilities in estimating
2 exposure. Because APEX simulates where individuals are located at specific times of the day,
3 more realistic exposure estimates are obtained in simulating the contact of individuals with these
4 spatially and temporally diverse concentrations.
5 Even though we estimated 63 ambient concentrations at all census tracts in each county-
6 level study area, the study area exposure modeling domain was defined as a subset of these
7 census tracts by using the ambient monitoring sites within the urban core of each study area's air
8 quality domain and a 30 km radius of influence. This zone of influence is consistent with what
9 was done in the 1st draft Os REA, though in that exposure assessment, only the ambient
10 monitoring data sites themselves were used to represent the APEX air districts, hence
11 concentrations measured at a particular monitoring site would be directly extrapolated outwards
12 to all census tracts within 30 km of that site. In contrast, by incorporating the VNA estimated
13 concentrations and retaining the same 30 km radius of influence, we are stressing the
14 significance of the monitor information in defining the urban core air quality while also
15 reasonably estimating concentration gradients (where such gradients exist) with increasing
16 distance from monitoring locations.
17 Thus, all air districts10 and census tracts that fall within the 30 km radius of each ambient
18 monitor were used to estimate the exposures, defining the final exposure modeling domain in
19 each study area (Table 5-1). The monitor IDs used to select the census tracts to be modeled are
20 provided in Appendix 5B, while the complete list of census tract IDs where exposures are
21 modeled are within the APEX control files for each study area (and are the same for each
22 simulation year).
23 5.2.4 Meteorological Data Used
24 APEX uses study area temperature data to select representative diaries for a particular
25 day and in selecting an appropriate air exchange rate used to calculate indoor residential
26 microenvironmental concentrations. APEX uses the data from the closest weather station to each
27 Census tract. To ensure reasonable coverage for each study area, a few to several meteorological
28 stations recording hourly surface temperature measurements were identified using data obtained
29 from the National Weather Service ISH data files.u Details regarding the meteorological stations
30 selected and data processing are given in Appendix 5B. Briefly, APEX requires the temperature
31 input data to be 100% complete. In general, any missing values were filled using a linear
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
5-10
-------
1 interpolation or regression approach that employs information from proximal meteorological
2 stations.
3 5.2.5 Populations Simulated
4 Exposure was estimated for four at-risk study groups residing in each study area: all
5 school-age children (ages 5-18), asthmatic school-age children, asthmatic adults (ages 19-95),
6 and all older adults (ages 65-95). Due to the increased amount of time spent outdoors engaged in
7 relatively high levels of physical activity (which increases intake), school-age children as a group
8 are particularly at risk for experiencing (Vrelated health effects (U.S. EPA, 2013, Chapter 8).
9 We report results for all school-age children down to age five, recognizing an increasing trend
10 for younger children to attend school. Some U.S. states allow 4-year-olds to attend kindergarten,
11 and most states have preschool programs for children younger than five. In 2000, six percent of
12 U.S. children ages 3 to 19 who attend school were younger than five years old (2000 Census
13 Summary File 3, Table QT-P19: School Enrollment). Currently we do not estimate exposure for
14 these younger children due to a lack of information that would let us confidently characterize
15 these younger aged children. While EPA guidance recommends, for certain instances, an upper
16 age group of children ages 16 through 21 (U.S. EPA, 2005), we restricted our upper age
17 classification of children through age 18. In considering the expected variability in activity
18 patterns over the span of ages 16 through 21 (e.g., time spent outdoors, time in school, each in
19 contrast to time spent working) and the relatively small difference in respiratory physiology over
20 that same age span compared with that of adults (e.g., Figure 5-17), factors critical for high O^
21 exposure and dose, we assumed simulated persons age 19 to 21 would be best included in our
22 adult study group. The number of persons represented in each of the 15 study areas is given in
23 Table 5-1 and, considering all study areas together, captures approximately 32.8 % of all
24 children ages 5 to!8 and 32.0 % of the total U.S. population ages 5 to 95.
25 The number of asthmatic school-age children and asthmatic adults in each study area was
26 estimated using asthma prevalence from the Center for Disease Control (CDC) and Prevention's
27 National Health Interview Survey (NHIS).12 Briefly, years 2006-2010 NHIS survey data were
28 combined to calculate asthma prevalence, defined as the probability of a "Yes" response to the
29 question: "do you still have asthma?" among those that responded "Yes" to the question "has a
30 doctor ever diagnosed you with asthma?". The asthma prevalence was first stratified by NHIS
31 defined regions (Midwest, Northeast, South, and West), sex, age (single years for ages 0-17) or
32 age groups (ages > 18), and by a family income/poverty ratio.13 These new asthma prevalence
33 estimates were then linked to U.S. census tract level poverty ratio probabilities (U.S. Census
2 See http://www.cdc.gov/nchs/nhis.htm (accessed October 4, 2011).
3 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.
5-11
-------
1
2
3
4
5
6
7
9
10
11
12
13
14
15
16
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 calculated for each study area simulation is provided here in Table 5-2.
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 DC
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 Persons (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
All simulated persons (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 (<1 to
94). The database is geographically diverse, containing diaries 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.
5-12
-------
1 Table 5-3 summarizes the studies and number of diaries in CHAD used in this
2 assessment, noting that the total CHAD diaries used by APEX is restricted to just over 41,000
3 given our simulation age range (5-95) and additionally selected usability requirements.14
4 Additional context regarding the representativeness of the CHAD data in estimating exposure is
5 provided in section 5.3.1 and Appendix 5G.
6 APEX creates a sequence of daily diaries across the entire Os season for each simulated
7 individual using a method designed to capture the tendency of individuals to repeat activities,
8 based on reproducing realistic variation in a key diary variable (Glen et al., 2008). For this 63
9 analysis, the key variable selected is the amount of time an individual spends outdoors each day,
10 one of the most important determinants of exposure to high levels of 63 (see section 5.3.2). The
11 longitudinal method targets two statistics, a population diversity statistic (D) and a within-person
12 autocorrelation statistic (^4). Values of 0.2 for D and 0.2 for^4 were initially developed based on
13 analyses by Geyh et al. (2000) and Xue et al. (2004), with both studies evaluating groups of
14 children ages 7 to 12 in a single study area. We adjusted values for D upwards to 0.5 to reflect a
15 broader range of ages and to better estimate repeated activities.15 Further details regarding the
16 development of the longitudinal methodology can be found in U.S. EPA (2012a, b).
17
18
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.
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.
5-13
-------
1 Table 5-3 Consolidated Human Activity Database (CHAD) Study Information and
2 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 Os Exposure: Elementary
School (LAE)
LA Os 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 63 Averting Behavior (OAB)1
RTP Panel (RTP)1
Seattle (SEA)1
Study of Use of Products and
Exposure Related Behavior (SUP)
1,2
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
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
1 Study data added after 2007 O3 NAAQS review.
2 Study data added after 2012 1st Draft O3 REA.
5-14
-------
1
2 5.2.6 Key Physiological Processes And Personal Attributes Modeled
3 The modeling of physiological processes relevant to the 63 exposure and intake is
4 complex, particularly when representing inter- and intra-personal variability in energy
5 expenditure (EE) and ventilation rates (VE). APEX has a module capable of estimating several
6 variables associated with every activity performed by simulated individuals. Briefly, the module
7 links the diary indicated activities to specific energy expended, the rate of oxygen consumed
8 (VO2) and the associated ventilation rate, all considering the unique sequence of events
9 individuals go through each simulated day. The activity-specific time-series of VE estimates
10 ultimately serve as an important variable used in estimating 63 intake as well as in identifying
11 when simulated individuals performing activities at moderate or greater exertion. In addition,
12 age, sex, and body mass related physiological differences are specifically taken into account by
13 the ventilation algorithm, derived using ventilation data obtained from several human studies
14 (see Graham and McCurdy, 2005):
15 \n(VE/BM)=b0+bl\n(V02/BM) + b2\n(l + age) + b3sex+eb+ew (5-2)
16 where,
17
18 In = natural logarithm of variable
•
19 VE! BM = activity specific ventilation rate, body mass normalized (liter air/kg)
20 bt = see Table 5-4
•
21 V'oilBM = activity specific oxygen consumption rate, body mass normalized
22 (liter/O2/kg)
23 age = age of the individual (years)
24 sex = sex (-1 for males, +1 for females)
25 6b = randomly sampled error term for between persons N{0, se}, (liter air/kg)
26 ew = randomly sampled error term for within persons N{0, se}, (liter air/kg)
27
28 As indicated by Equation 5-2, the random error (e) is allocated to two variance
29 components used to estimate the between-person (inter-individual variability) residuals
30 distribution (e£) and within-person (intra-individual variability) residuals distribution (ew). The
31 regression parameters bo, bi, 62, and 6j are assumed constant over time for all simulated persons,
32 eb is sampled once per person by APEX, while whereas ew varies from event to event. Point
33 estimates of the regression coefficients and standard errors of the residuals distributions are
34 given in Table 5-4. See Appendix 5A, Isaacs et al. (2008), and Chapter 7 of the APEX TSD (US
5-15
-------
1
2
3
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.
4 Table 5-4 Ventilation equation coefficient estimates (bj) and residuals distributions (e\)
5
6
7
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
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
eb
0.0955
0.1217
0.1260
0.1064
ew
0.1117
0.1296
0.1152
0.0676
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
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-2'21*1 EM0'6*21 (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.
The importance of modeling indoor microenvironments (e.g., homes, offices, schools) is
underscored by research indicating that personal exposure measurements of 63 may not be well-
' 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.
5-16
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
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 63 concentrations were
reduced by atmospheric reactions (e.g., scavenging by NOx) or other processes, and vehicular
microenvironments considered both the outdoor concentration attenuation and
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
16
17
18
19
20
21
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 63 air pollution, we
selected the daily maximum 8-hr average 03 exposure17 for every simulated individual and
It is important to stress here that only the maximum 8-hr exposure concentration is retained for each day
simulated, per person. While every day could contain twenty-four unique 8-hr averages and that it is entirely
5-17
-------
1 stratified these exposures by exertion level at the time of exposure. This indicator was selected
2 based on controlled human exposure studies where reported adverse health responses were
1 &
3 associated with exposure to Os and while the study subject was exercising. Factors important in
4 calculating this indicator includes the magnitude, duration, frequency of exposures, and the
5 breathing rate of individuals at the time of exposure. As a reminder, the calculated daily
6 maximum 8-hr average exposure concentrations are distinct from that of daily maximum 8-hr
7 average ambient concentrations by accounting for simulated individual's time-location-activity
8 patterns and 63 concentration decay/variation occurring within the occupied microenvironments.
9 Benchmark levels used in this assessment include 8-hr average O?, exposure
10 concentrations of 60, 70 and 80 ppb; the same benchmark levels used for the 2007 Os exposure
11 assessment (U.S. EPA, 2007b). Estimating exposures to ambient Os concentrations at and above
12 these benchmark levels is intended to provide perspective on the public health impacts of 03-
13 related health effects observed in human clinical and toxicological studies, but that cannot
14 currently be evaluated in quantitative risk assessments (e.g., lung inflammation, increased airway
15 responsiveness, and decreased resistance to infection). The 80 ppb-8hr benchmark level
16 represents an exposure level where there is substantial clinical evidence demonstrating a range of
17 Os-related effects including lung inflammation and airway responsiveness in healthy individuals.
18 The 70 ppb-8hr benchmark level reflects evidence that asthmatics have larger and more serious
19 effects than healthy people as well as a substantial epidemiological evidence of adverse effects
20 associated with 03 levels that extend below 80 ppb-8hr. The 60 ppb-8hr benchmark level
21 represents the lowest exposure level at which (Vrelated effects have been observed in clinical
22 studies of healthy individuals. See ISA section 6.2.1 for further discussions regarding the body of
23 evidence supporting the selection of these benchmark levels.
24 The level of exertion of individuals engaged in particular activities is approximated by an
25 equivalent ventilation rate (EVR), that is, ventilation normalized by body surface area (BSA, in
26 m2) and is calculated as VE/BSA, where VE is the ventilation rate in liters/minute. For
27 identifying moderate or greater exertion occurring during any 8-hr average exposure period in
r\
28 this assessment, we used the lower bound EVR value of 13 (liters/min-m ) based on a range of
29 EVRs used by Whitfield et al. (1996) to categorize persons engaged in moderate exertion
30 activities for an 8-hr period. Whitfield et al. (1996) developed this range from EVR data reported
31 in a 6.6-hr controlled human exposure study conducted by McDonnell et al. (1991).
possible multiple benchmark exceedances could occur for an individual on certain high O3 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 number of benchmark exceedances is underestimated because of
the lesser likelihood of an 8-hr exposure above the same threshold due to the longer averaging time.
5-18
-------
1 APEX then calculates two general types of exposure estimates for the population of
2 interest: the estimated number of people exposed to a specified 63 concentration level and, the
3 number of days per Os season that they are so exposed; the latter metric is expressed in terms of
4 person-days. The former highlights the number of individuals exposed one or more times per Os
5 season at or above a selected benchmark level. The person-days measure estimates the number of
6 times per season the simulated individuals are exposed at or above a selected benchmark level
7 and summed across individuals comprising the population. We note that a person-days metric
8 conflates people and occurrences: one occurrence for each of 10 people would be counted the
9 same as 10 occurrences for one person (i.e., 10 person-days at or above benchmark level). In this
10 assessment we are more interested in reporting multiday exposures rather than total person-days,
11 that is, the number of times an individual experiences multiple exposures at or above a
12 benchmark level during an Os season. Given the complexities of the exposure modeling, the four
13 study groups considered, the 15 study areas, the 5 years of ambient air quality, the multiple air
14 quality scenarios simulated, and ultimately the output data generated, including both single and
15 multiday exposures for simulated individuals, the consolidation of the results and the related
16 graphic depictions used in this assessment requires additional discussion.
17 To begin, a simple example of summary results is the estimated percent of asthmatic
18 school-age children experiencing exposures at or above a single 8-hr benchmark level when
19 considering base air quality stratified by year (e.g., Figure 5-2, left panel). This presentation
20 largely depicts the variability in Os exposure across the 15 study areas within years, along with
21 an illustration of broad year-to-year temporal variability. A general finding regarding temporal
22 variability extracted from this graph would be that fewer asthmatic school-age children exceed
23 daily maximum 8-hr average exposures of 60 ppb considering 2009 base air quality when
24 compared with other simulation years. An observation regarding the spatial variability could
25 include the range of exposures within years (i.e., the study area variability) spans between 15 to
26 35 percentage points, dependent on the particular simulation year (Figure 5-2, left panel). One
27 could also stratify the same exposure results by study area (e.g., Figure 5-2, right panel), thus
28 depicting variability in estimated exposures across years within each study area, along with
29 having broad study area comparisons. A general finding regarding temporal variability in this
30 type of presentation would be that the range of exposures within study areas spans about 20
31 percentage points, though some study areas have a generally small range (<5 percentage points)
32 for most simulated years. An observation regarding spatial variability could be that Chicago
33 largely has the fewest asthmatic school-age children at or above benchmark levels, having a
34 mean about 15%, while Los Angeles consistently has the most asthmatic school-age children,
35 having a mean about 35%, at or above benchmark levels while at moderate or greater exertion.
36
5-19
-------
2006 2007 2008 2009 2010
Shiulatai year
ATL BAL BOS CHI CLE DAL DEN DET HOU LA NY PHI SAC STL WAS
Study Area Abbreviation
1 Figure 5-2 Percent of asthmatic school-age children in all study areas with at least one Os
2 exposure at or above 60 ppb-8hr while at moderate or greater exertion using base air
3 quality (2006-2010), stratified by year (top left panel) or by study area (bottom left panel).
4
5 While these boxplots are an efficient tool that summarize potentially complex data sets
6 by illustrating important statistical aspects of data analysis results (e.g., means, ranges,
7 occasional upper percentile data values), at times important features of the data may be masked
8 (e.g., trends or patterns within consolidated variables) and the presentation of other aspects of the
9 exposure results would require the generation of additional graphs (e.g., results for additional
10 benchmark levels). A tabular format could be one way to present all possible data, though given
11 the number of APEX simulations performed (i.e., > 1,000) and aforementioned dimensions of
12 the assessment, linking the trends and patterns across all study areas, years, and benchmark
13 levels from the numerous output tables would be visually challenging.
14 This discussion regarding properly representing temporal and spatial variability in the
15 exposure results can be further extended to include the added dimension of the five air quality
16 scenarios (base, existing standard, and three alternative standard levels). Mindful of these
17 complexities, we elected to use a multi-panel graphing approach to succinctly summarize the
18 exposure output data, while also retaining as much information as possible in a single page
19 format to allow for visual analysis of trends and patterns. As an example, Figure 5-3 (top panels)
5-20
-------
1 illustrates boxplots for Atlanta similar to those presented above, though the exposure results are
2 for the three exposure benchmark levels of interest, with each stratified by the particular adjusted
3 air quality scenario. As expected with increasing stringency of the 8-hr standard level, fewer
4 asthmatic school-age children are exposed at or above a given benchmark level. Also expected is
5 the fewer percent of asthmatic school-age children exposed to higher benchmark levels when
6 compared with lower benchmark levels. While these three graphs can provide a clear depiction
7 of the exposure results for a single study area, the six years encompassing the two averaging
8 periods 2006-2008 and 2008-2010 are combined in the graphic and difficulty would remain in
9 simultaneously exhibiting all 15 areas.
10 To overcome these limitations, Figure 5-3 (lower panel) exhibits all of the dimensions of
11 the exposure results mentioned above (i.e., year, benchmark level, study area) along with
12 distinguishing between the two standard averaging periods for each the existing (75 ppb-8hr) and
13 alternative standard levels (60, 65, 70 ppb-8hr). The nomenclature above each subgraph
14 indicating the particular air quality scenario requires defining. For example, a panel heading of
15 "75" contains the exposures estimated when air quality was adjusted to just meet the existing
16 standard level of 75 ppb-8hr (4th highest daily maximum 8-hr average 63 concentration averaged
17 over a three year period) either using years 2006, 2007, and 2008 ambient air quality data or for a
18 second averaging period that extended from 2008 through 2010 (with results for each given by
19 two separate lines on the same plot). Exposure results are readily observed for any air quality
20 scenario, year, or benchmark level of interest. For example, when considering the 75ppb
21 standard 2006-2008 averaging time scenario, 20% of asthmatic school-age children in Atlanta
22 experience at least one daily maximum 8-hr average exposure of 60 ppb occurs when
23 considering year 2006 air quality, while only about 5% experience exposures at or above the
24 same benchmark level considering 2008 air quality (though when considering the 2008-2010
25 averaging period, approximately 20% of asthmatic children are estimated experience at least one
26 exposure at or 60 ppb-8hr). Fewer than 5% of asthmatic school-age children in Atlanta
27 experience at least one benchmark exposure of 70 ppb-8hr considering any year and any air
28 quality scenario, including just meeting the existing O^ standard.
29 Because APEX simulates the complete time series of exposure for every simulated
30 individual, also output is the number of times an individual experiences a benchmark exceedance
31 over the duration of the simulation (i.e., the entire 63 season simulated in each study area). These
32 data can also be summarized in a similar multi-panel format, though differ slightly in
33 composition from that of Figure 5-3. Instead of displaying the percent of persons with at least
34 one exceedance of each of the three benchmarks, presented are the percent of persons with
35 multiple exposures at or above a single benchmark within an Os season. For example, Figure 5-4
36 illustrates the percent of asthmatic school-age children in Atlanta having multiple days where
5-21
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
exposures (>2, >4, and >6 per O^ season) were at or above 60 ppb-8hr 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% 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 63
season.
60 ppb benchmark
70 ppb benchmark
80 ppb benchmark
ii\2)
^25
1 20
^ 15
IT
V
* 0
— TT
A]S
c S
i
6
j;
•6
VO 0
....i
65 60 66 70 75 56 60
?Qs
| 1S
1 1-°
I as
•s
., n
n
T
•56 70 75 55 BL> 65 70 75
9i Standard Lwd (ppM ah Samlsrd Laval [ppb) ai Standard L£vel Jiptj
IP •
(y
-------
2
O
4
5
6
7
8
9
10
1 1
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
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
persons experiencing a single exceedance of 70 ppb-8hr and multiple exceedances (e.g., four) of
60 ppb-8hr also driven by the overall ambient concentration distribution. However, given that
very few persons experience these types of 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 the REA.
Figure 5-4 Percent of asthmatic school-age children in Atlanta with multiple Os exposures
at or above 60 ppb-8hr while at moderate or greater exertion, years 2006-2010 air quality
adjusted to just meet the existing and alternative Os standard levels.
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., persons experiencing daily maximum 8-hr average Oi
exposure concentrations at or above 60 ppb, 70 ppb, 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 63 exposures associated with
the ambient air quality adjusted to just meet the existing and potential alternative O^ standards.
19 Thus, the year 2008 will have two sets of estimated exposures, one from each of the two sets of design values. In
Figure 5-2, the greater temporal variability observed for 2008 is driven in part by differences in some study areas
resulting from the air quality adjustment period. Exposure results for both 2008 averaging periods are provided
when presenting data by year. 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.
5-23
-------
1 Thus, most of the exposure results presented and discussed are for where ambient air quality was
2 adjusted to just meet these particular scenarios. While understanding exposures and health risks
3 associated with historical and existing air quality is important, the primary goal of this and any
4 REA is to evaluate to what extent the existing NAAQS, and its associated air quality, protect
5 health and to what extent alternative NAAQS protect health. Exposure results associated with
6 recent (base) air quality are briefly discussed here first, though largely reported in Appendix 5F.
7 5.3.2 Exposure Modeling Results for Base Air Quality
8 The exposure results for the base air quality are distinguished from the other air quality
9 scenario results primarily due to the wide ranging variability in estimated exposures across the
10 study areas and years. The variability in exposures are the result of the wide ranging variability
11 in ambient concentration levels, with perhaps some years in some study areas exhibiting air
12 quality at or near that just meeting the current 8-hr standard, while other study areas and years
13 exhibiting air quality levels much higher than the existing 8-hr standard. These exposures are
14 informative in describing the existing or recent health risks associated with a unique air quality
15 scenario, but because they variably diverge from a set concentration level of interest (such as the
16 existing 8-hr standard), they are of limited relevance in evaluating the adequacy of either the
17 existing NAAQS as well as potential alternative air quality standards. That said, detailed tabular
18 and graphic presentations of exposure results associated with the base air quality (years 2006-
19 2010) are provided in Appendix 5F, with only key findings summarized in the following
20 discussion.
21 Consistent with the previously discussed observations regarding year-to-year variability
22 in ambient concentrations (Chapter 4), most study areas have the greatest percent of all school -
23 age children experiencing concentrations at or above the three benchmark levels during 2006 or
24 2007 along with having the lowest percent of all school-age children exposed during 2009. In
25 general, between 20-40% of all school-age children experience at least one Os exposure at or
26 above 60 ppb-8hr, 10-20% experience at least one 63 exposure at or above 70 ppb-8hr, and 0-
27 10% experience at least one 63 exposure at or above 80 ppb-8hr, all while at moderate or greater
r\
28 exertion (i.e., an 8-hr EVR > 13 L/min-m ) and considering the base air quality (2006-2010).
29 Year-to-year variability observed for asthmatic school-age children and the percent of asthmatic
30 school-age children were similar to exposure results for all school-age children, largely a
31 function of having both simulated study groups using an identical time-location-activity diary
32 pool to construct each simulated individual's time series of activities performed and locations
33 visited.
34 The overall year-to-year pattern of exposure for asthmatic adults is similar to that
35 observed for all school-age children, though the percent of the asthmatic adult study group
5-24
-------
1 exposed is lower by a factor of about three or more. Having a lower percent of asthmatic adults
2 exposed is expected given that outdoor time expenditure is an important determinant of 63
3 exposure (section 5.4.2) and that adults spend less time outdoors than children (section 5.4.1), as
4 well as adults having a lower outdoor participation rate. The percent of all older adults
5 experiencing exposures at or above the selected benchmark levels is lower by a fewer percentage
6 points when compared with the results for asthmatic adults. Again, older adults, on average,
7 would tend to spend less time outdoors and do so with less frequency when compared with both
8 adults and children (section 5.4.1), in addition to fewer older adults performing activities at
9 moderate or greater exertion for extended periods of time, thus leading to fewer persons exposed
10 to 63 concentrations of concern.
11 The year-to-year patterns of the single and multiple exposure occurrences considering
12 base air quality (2006-2010) were similar among the four exposure study groups, therefore only
13 results for all school-age children will be summarized here. Depending on the year and study
14 area, about 10-25% of all school-age children could experience at least two exposures above the
15 60 ppb-8hr benchmark during the 63 season, while about 5-10% school-age children could
16 experience at least four. Most study areas and years are estimated to have fewer than 5% of all
17 school-age children experience six or more exposures above 60 ppb-8hr considering the base air
18 quality. When considering the multi-day exposures for all school-age children at or above the 70
19 ppb-8hr benchmark, about 2-10% of all school-age children could experience at least two
20 exposures during the O^ season, while four or more exposures were generally limited to fewer
21 than 4% of all school-age children. Almost half of the study area-year combinations had no
22 school-age children experiencing two or more exposures at or above the 80 ppb-8hr benchmark,
23 with the other half estimated to have about 1% of all school-age children experiencing two or
24 more exposures at or above the 80 ppb-8hr benchmark. School-age children having four or more
25 80 ppb-8hr benchmark exceedances were limited to only a few study area years and, where a
26 non-zero value was estimated, were limited to < 0.5% of the study group.
27 5.3.3 Exposure Modeling Results for Simulations of Just Meeting Existing and
28 Alternative O3 Standards
29 In this section, we present the exposures estimated when considering the air quality
30 adjusted to just meeting the existing 03 NAAQS standard, as well as when considering potential
31 alternative standard levels (55, 60, 65, 70 ppb 8-hr) of the existing standard. Comprehensive
32 multi-panel displays of exposure results are presented for each of the study groups of interest,
33 i.e., all school-age children (5-18), asthmatic school-age children, asthmatic adults (19-95), and
34 all older adults (ages 65-95; Figure 5-5 to Figure 5-8, respectively). Included in each display are
35 the three benchmark levels (60, 70, and 80 ppb-8hr), the five years of air quality (2006-2010), for
36 the 15 study areas. A single multi-panel display is used to present the results for each of the four
5-25
-------
1 study groups, beginning with the estimated percent of persons exposed at least one time at or
2 above the selected benchmark levels. Modeled exposures in the 15 study areas and considering
3 each benchmark level are presented on the same scale to allow for direct comparisons across the
4 multi-panel display. The most notable patterns in the exposure results are described here using
5 one study group (i.e., all school-age children), as there is a general consistency in the year-to-
6 year variability within each study area across all four study groups. Any deviation from the
7 observed pattern will be discussed for the subsequent study group.
8 We note that after adjusting to just meet a potential 8-hr ambient standard level of 55
9 ppb, there were nearly no persons exposed at or above any of the selected benchmark levels, thus
10 these data, while modeled, are not presented in detail here. In addition, in one study area
11 (Chicago), OT, ambient monitor design values were below that of the existing standard during the
12 2008-2010, therefore APEX simulations could not be performed for meeting the existing
13 standard for that 3-year period. And finally, we were not able to simulate just meeting a standard
14 level of 60 ppb-8hr or below in the New York study area (see Chapter 4 for details), thus APEX
15 simulations for these air quality scenarios could not be performed in New York.
16 Figure 5-5 illustrates the exposures estimated for all school-age children in each study
17 area with general observations as follows. After adjusting air quality to just meet the existing and
18 alternative standards, there are virtually no school-age children exposed at or above 80 ppb-8hr,
19 with very few school-age children exposed at or above the 70 ppb-8hr benchmark. For example,
20 out of 87 possible study area and year combinations considering air quality adjusted to just meet
21 the existing standard (the least stringent standard level considered here), only 29 resulted in >
22 0.1% estimated percent of all school-age children exposed at least once at or above the 80 ppb-
23 8hr benchmark with the maximum percent of all school-age children exposed estimated for St.
24 Louis (1.1%). Ninety-four percent of study area and year combinations had fewer than 5% of all
25 school-age children experiencing at least one daily maximum 8-hr average exposure > 70 ppb
26 considering ambient air quality adjusted to just meeting the existing standard, again with a
27 maximum of 8.1% occurring in St. Louis. When considering air quality adjusted to just meet an
28 8-hr ambient standard level of 70 ppb, < 0.2% of all school-age children experience at least one
29 80 ppb-8hr exposure benchmark exceedance for all study area and year combinations, while for
30 76 or 90 study area and year combinations, < 1% of all school-age children experience a 70 ppb-
31 8-hr exposure benchmark exceedance. This pattern of having very few school-age children
32 experiencing exposures at or above 70 and 80 ppb-8hr is as expected given the nature of the air
33 quality adjustment procedure that limits 8-hr ambient concentrations at or above the selected
34 potential alternative standard level.
35 In contrast, approximately 10-20% percent of all school-age children are estimated to be
36 exposed to at least one 60 ppb-8hr concentration when considering air quality just meeting the
5-26
-------
1 existing standard (Figure 5-5). And similar to that mentioned above regarding exposures
2 associated with the base air quality, a general year-to-year exposure pattern emerges with respect
3 to study area and year. For the Northeastern (Boston, New York), Mid-Atlantic (Philadelphia,
4 Washington DC, Cleveland) and Mid-Western (Chicago, Detroit, and St. Louis) study areas, the
5 maximum percent of all school-age children exposed generally occurs during year 2007. For the
6 Southern (Atlanta, Dallas, Houston) and Western (Denver, Los Angeles, Sacramento) study
7 areas, the maximum exposure occurs during year 2006. Deviations from this temporal exposure
8 pattern appear mostly as a result of the standard averaging period, with the 2008-2010 period
9 producing equal or greater maximum exposures during either 2008, 2010, or both years and most
10 prevalent in the Northeastern and Mid-Atlantic study areas (Baltimore, Boston, New York,
11 Philadelphia, Washington DC; note also a trend in Atlanta, Denver, St. Louis).
12 These 60 ppb-8hr exposure patterns remain consistent when considering air quality
13 adjusted to just meet a 70 ppb-8hr ambient standard, though the percent of all school-age
14 children exposed is less than that observed when considering the air quality adjusted to just meet
15 existing standard. Further, 75 of 90 study area and year combinations are estimated to have <
16 10% of all school-age children experience a 60 ppb-8hr or greater exposure, though between 10-
17 20% of all school-age children were estimated to be exposed for a few study area and year
18 combinations (e.g., Atlanta-2006, Chicago-2007 and -2010, and Houston-2009). When
19 considering air quality adjusted to just meet a 65 ppb standard level, the percent of all school-age
20 children experiencing an exposure at or above 60 ppb-8hr diminishes to 5% or less for most
21 study areas and years (i.e., 81 of 90 study area year combinations).
22 All of what has been described regarding the estimated exposures to school-age children
23 (i.e., the year-to-year and benchmark level patterns, and the percent of the study group exposed)
24 also applies to the exposures estimated for asthmatic school-age children (Figure 5-6). Different
25 however would be the relative number of asthmatic school-age children exposed in each study
26 area if compared with all school-age children, as the asthma prevalence rates vary by study area
27 (Table 5-2), though on average are about 10% of the population of children.
28 The percent of asthmatic adults (Figure 5-7) experiencing daily maximum 8-hr average
29 exposures above the selected benchmark levels is sharply lower than that estimated for all
30 school-age children. For example, only three of a possible 84 study area and year combinations
31 (Chicago-2007, Houston-2009, and St. Louis-2007) were estimated have > 0.1% of asthmatic
32 adults experience a daily maximum 8-hr average exposure > 80 ppb, and only six of a possible
33 84 study area and year combinations were estimated have >1% of asthmatic adults experience an
34 daily maximum 8-hr average exposure > 70 ppb, all occurring when considering air quality just
35 meeting the existing standard. No study area or year combination has more than 10% of
36 asthmatic adults estimated to experience an exposure at or above 60 ppb-8hr when considering
5-27
-------
1 air quality just meeting the existing standard, with 67 of 84 study area and year combinations
2 estimated to have 5% or less asthmatic adults experiencing such exposures.
3 When considering air quality adjusted to just meeting a standard level of 70 ppb-8hr, no
4 asthmatic adults experience an exposure at or above 80 ppb-8hr and < 0.6% experience a daily
5 maximum 8-hr average exposure > 70 ppb for any study area or year combination. Less than 5%
6 of asthmatic adults could experience an exposure at or above 60 ppb-8hr when considering air
7 quality adjusted to just meet a standard level of 70 ppb-8hr for 88 or 90 possible study area year
8 combinations, with the maximum percent of adult asthmatics exposed outside this range
9 occurring in Denver (6.8%-2008) and St. Louis (5.5%-2007).
10 Older adults are estimated to have the fewest exposures above the two highest benchmark
11 levels when considering the adjusted air quality. For example, only two of a possible 84 study
12 area and year combinations (St. Louis-2007 and Washington DC-2008) were estimated have >
13 0.1% of asthmatic adults experience a daily maximum 8-hr average exposure > 80 ppb, and only
14 six of a possible 84 study area and year combinations were estimated have > 1% of asthmatic
15 adults experience a daily maximum 8-hr average exposure > 70 ppb, all occurring when
16 considering air quality just meeting the existing standard (Figure 5-8). Also, exceeding the 60
17 ppb-8hr exposure benchmark appears to be limited to fewer than 5% of all older adults when
18 considering air quality adjusted to just meet the existing standard and a standard level of 70 ppb-
19 8hr, and occurs in < 2% of all older adults when considering a standard level of 65 ppb-8hr.
20 An example of multi-day exposure results associated with adjusted air quality is provided
21 in Figure 5-9. The percent of all school-age children estimated to experience multi-day exposures
22 above benchmark levels during each study area's Os season is largely limited to two air quality
23 scenarios: the existing standard and air quality adjusted to just meeting a standard level of 70
24 ppb-8hr. This is because of the small percent of school-age children experiencing even a single
25 exposure above the lowest benchmark level when considering standard levels at or below 65
26 ppb-8hr. In addition, when experiencing multiple exposures, most school-age children appear to
27 have at most two days above benchmark levels per Os season, even when considering the lowest
28 benchmark level of 60 ppb-8hr. For example, 81 of 87 possible study area and year combinations
29 have < 10% of all school-age children experiencing two or more exposures > 60 ppb-8hr when
30 considering an ambient standard level of 75 ppb-8hr, while 83 of 90 possible study area and year
31 combinations have < 5% of all school-age children experiencing two or more exposures > 60
32 ppb-8hr when considering an ambient standard level of 70 ppb-8hr. With increasing stringency
33 in the standard level to 65 ppb-8hr, 81 of 90 possible study area and year combinations have <
34 1% of all school-age children experiencing two or more exposures > 60 ppb-8hr.
35 Multi-day exposure to the higher exposure benchmarks (either the 70 or 80 ppb-8hr) is a
36 rare occurrence, even when considering the air quality adjusted to the existing O^ standard. For
5-28
-------
1 example, there were no school-age children experiencing two or more exposures above 80 ppb-
2 8hr in all but one study area year combination and, and when considering that one study year
3 having a non-zero value (St. Louis-2007), the estimated percent of all school-age children at or
4 above the exposure benchmark was only 0.1%. Further, 83 of 87 possible study area and year
5 combinations have < 1% of all school-age children experiencing two or more exposures > 70
6 ppb-8hr, also when considering an ambient standard level of 75 ppb-8hr.
7
5-29
-------
0>
2
3
4
5
< .t
£
C 0)
0) -a
s.i
o ro
0.2
fl
Is
o «j
= «
< <6
o|
c °
0) o
i2m
0)
Q.
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 while at moderate or greater
exertion, years 2006-2010, air quality adjusted to just meet the existing and potential
alternative standards.
5-30
-------
2
3
4
5
O £
±? re
< 3
4-> CD
O 0)
o >
*- O
0) ja
25 -
20 -
15 -
10 -
5 -
0 -
25
20 -
15 -
10 -
5 -
0 -
25 -
20 -
15 -
10 -
S-
0 -
25 -
20 -
15 -
10-
5-
0 -
25 -
20 -
15 -
10 -
5 -
0 -
25 -
20 -
15 -
10 -
5 -
0 -
25 -
20
15 -
10 -
5 -
0 -
25 -
20 -
15-
10 -
5-
0-
25 -
20 -
15 -
10 -
5 -
0 -
25 -
20 -
15 -
10 -
5 -
0 -
25 -
20 -
15 -
10 -
5 -
0 -
25 -
20 -
15-
10 -
5-
0 -
25 -
20 -
15 -
10 -
5-
0 -
25 -
20
15 -
10 -
5 -
a -
25 -
20 -
15 -
10 -
5 -
0-
"•o • » — e B
o o o_^
0—
0—
° -C e O
— e & — o 0
£=*».
I
HO
8 O
O « O 00
•e o o o <
e o • • n o
o o—=8=
-------
2
3
4
5
Figure 5-7 Percent of all asthmatic adults with at least one daily maximum 8-hr
average Os exposure at or above 60, 70, and 80 ppb-8hr while at moderate or
greater exertion, years 2006-2010, air quality adjusted to just meet the existing and
potential alternative standards.
5-32
-------
2
3
4
5
o- o m o o—D o e » o—o
I
9 o—a a o
00*
Figure 5-8 Percent of all older adults with at least one daily maximum 8-hr average
Os exposure at or above 60, 70, and 80 ppb-8hr while at moderate or greater
exertion, years 2006-2010, air quality adjusted to just meet the existing and potential
alternative standards.
5-33
-------
1
2
3
8" D 0
^ 2--
a— =*
o o o o o
o o o o o
-e- e e- o
9—9
i
Figure 5-9 Percent of all school-age children with multiple daily maximum 8-hr
average Os exposures at or above 60 ppb while at moderate or greater exertion,
years 2006-2010, air quality adjusted to just meet the existing and potential
alternative standards.
5-34
-------
1 5.4 TARGETED EVALUATION OF EXPOSURE MODEL INPUT AND OUTPUT
2 DATA
3 This section summarizes the results of several targeted evaluations intended to provide
4 additional insights to APEX input data or approaches used to estimate exposures (CHAD data
5 attributes and activity pattern evaluations, comparison of CHAD outdoor time data with ATUS,
6 comparisons of asthmatic outdoor time expenditure and exertion levels to that of non-
7 asthmatics), exposure results for additional exposure populations of interest (outdoor workers,
8 school-age children during summers, impact of averting), and model performance evaluations
9 (personal exposure measurements and independent ventilation rate estimates compared with
10 APEX estimates). Detailed analysis results are provided in Appendix 5G.
11 5.4.1 ANALYSIS OF TIME-LOCATON-ACTIVITY DATA
12 While CHAD is the most comprehensive and relevant source of time-location-activity
13 data available for use in our exposure modeling, there are a few limitations to the survey data
14 contained therein, many of which are founded in the individual studies from which activity
15 patterns were derived (Graham and McCurdy, 2004). CHAD is a collection of related survey
16 data, though individual study attributes can range widely (e.g., survey participant ages, region or
17 city of residence, time-of-year data collected). We note that many of the assumptions about use
18 of these activity patterns in exposure modeling are strengthened by the manner in which they are
19 used by APEX. This is done by focusing on selecting the most important individual attributes
20 that contribute to variability in human behavior (e.g., age, sex, day-of-week, ambient
21 temperature) and linking these attributes of simulated individuals to the population demographics
22 of each census tract (see section 5.2.5) and the study area temperatures (section 5.2.4). Further,
23 one key lifestyle attribute is also accounted for in generating longitudinal diary profiles by
24 simulating both the intra- and interpersonal variability in time spent outdoors (section 5.2.5; Glen
25 et al., 2008).
26 A few questions may arise as to the representativeness of the CHAD diaries to the
27 simulated population. For example, the year of a particular survey study may differ from our
28 simulated exposure population by as much as 30 years (i.e., some activity pattern data were
29 generated in the 1980s). In addition, there are other personal attributes (e.g., ethnicity, income
30 level, lifestyle factors20), health conditions (e.g., asthma, cardiovascular disease), and situational
31 factors (e.g., availability of parks and recreation areas) that are not used in creating the simulated
32 persons that could be influential in estimating exposures. Considering this, a number of
20 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.
5-35
-------
1 evaluations were performed to answer questions regarding important personal attributes used in
2 generating simulated individuals and the general representativeness of the CHAD time-location-
3 activity data. First though, we summarize the newly acquired activity pattern data now included
4 in CHAD compared with data available and used in the 1st draft O?, REA.
5 5.4.1.1 General Evaluation of CHAD Study Data: Historical and Recently A cquired Data
6 The number of diary days having complete information and used by APEX in the 2nd
7 draft 63 REA is 41,474 (Table 5-3). This is an increase of about 8,700 diaries currently used by
8 APEX compared with what was used by APEX in the 1st Draft Os REA. Further, there have been
9 eight new study data sets incorporated into CHAD and used in our current exposure assessment
10 since the previous Os NAAQS review conducted in 2007, most of which were from recently
11 conducted activity pattern studies (see Appendix 5B, Section 5B-4 for more information
12 regarding these studies). The diary data included from these new studies have more than doubled
13 the total activity pattern data used for 2007 Os exposure modeling and has increased the number
14 of children's diaries by about a factor of five. Currently, the majority of diaries (54%) from
15 CHAD are taken from surveys conducted in the past decade, while the pre-1990s diaries
16 represent less than 15% of the total diaries available by APEX.
17 5.4.1.2 Exposure-Relevant Personal A ttributes Included in CHAD and APEX Simulated
18 Individuals
19 The survey participants whose diary data are within CHAD were asked a number of
20 questions regarding their personal attributes. The number and type of attributes present for
21 diaries in CHAD is driven largely by the original intent of the individual study. In our exposure
22 assessment, we have strict requirements to simulate individuals using several personal attributes,
23 namely age, sex, temperature (as a surrogate for seasonal variation in activity patterns), and day-
24 of-week. These attributes are considered as important drivers influencing daily activity patterns
25 (Graham and McCurdy, 2004) and when diaries do not have these particular attributes for a
26 particular day, the diary day will not be used by APEX. We compared the representation of these
27 and other attributes in the current CHAD used by APEX with that in the 1st draft O3 REA and
28 found strong similarities in the attribute distributions between both databases, suggesting little
29 change in the overall composition of the database regarding these influential attributes.
30 While there may be other personal or situational attributes that affect daily time
31 expenditure (e.g., socioeconomic status, occupation of an employed person), these attributes are
32 typically not included in our assessment to generate simulated individuals simply because the
33 response to the attribute is missing for most of the study participants/CHAD diary days. For
34 example, income level is missing for about two-thirds of the CHAD diaries because either the
35 original study did not have an income/occupation related survey question or perhaps the
36 participant refused to answer the question if it were posed. If one were to select this personal
5-36
-------
1 attribute in developing a simulated individual's activity pattern (among using any other attribute
2 having missing responses), the pool of diaries available to simulate individuals may be extremely
3 limited, likely leading to repetition of diaries used for individuals or groups of similar individuals
4 and artificially reducing both intra- and inter-personal variability in time expenditure, or perhaps
5 resulting in model simulation failure altogether. This is why personal attributes are carefully
6 selected and prioritized according to both their prevalence in CHAD and whether the attribute
7 has a known significant influence on activity patterns.
8 5.4.1.3 Evaluation of Afternoon Time Spent Outdoors for CHAD and Survey Participants
9 There have been questions raised regarding the representativeness of the diaries from
10 studies conducted in the 1980s and whether there are any recognizable patterns in time
11 expenditure in the CHAD diaries across the time period when data were collected. Because time
12 spent outdoors is a significant factor influencing daily maximum 8-hr average 63 exposures, we
13 evaluated the current collection of CHAD diaries used by APEX for two metrics and considering
14 two dimensions: outdoor participation rate (i.e., the percent of people who spent some time
15 outdoors during their survey day) and the mean time spent outdoors for where the persons spent
16 at least one minute outdoors or at least 2 hours outdoors. Because time spent outdoors is an
17 important determinant for highly exposed individuals, we summarize the results here for the
18 diaries having at least 2 hours of outdoor time here, while all other results are provided in
19 Appendix 5G. CHAD diaries were stratified by five age groups (4-18, 19-34, 35-50, 51-64, 65+)
20 and three decades (1980s, 1990s, and 2000s) using the year the particular activity pattern study
21 was conducted. We note that CHAD is composed of primarily cross-sectional data (single diary
22 days per person), thus the trend evaluated over the three decades is changes (if any) in
23 participation rate and the time spent outdoors by the composite study population, not within
24 individuals.
25 Regardless of decade and duration of time spent outdoors, children tended to have the
26 highest outdoor participation rate when compared with the other age groups, while the oldest
27 adults (aged 65 or greater) tend to have the lowest participation rate. The CHAD diaries from the
28 1980's studies for children ages 4-18 have the highest outdoor participation rate (50%) compared
29 to other decades (35-40%) and all other age groups and decade of collection. When considering
30 the pool of diaries available for this age group, these 1980's studies contribute to approximately
31 19% of diaries having two or more hours of time spent outdoors during the afternoon. This
32 translates to a small effect on the overall outdoor participation rate for diary pools that would
33 include these earlier studies (39% participation rate) compared to the participation rate excluding
34 these studies (36% participation rate). In general, these outdoor participation rates are similar to
35 the finding reported recently by Marino et al. (2012) of 37.5%, though estimated for pre-school
36 age children. Thus, when considering participation in outdoor activities and the
5-37
-------
1 representativeness of the CHAD study data from the 1980s, it is unlikely that use of these oldest
2 diaries would strongly influence exposure model estimates.
3 There is variability in the amount of outdoor time evaluated over the three decades, with
4 diaries from the 2000's studies exhibiting perhaps the lowest range of mean outdoor time (190-
5 220 min/day) compared with the 1980's (210-240 min/day) and 1990's (212-258 min/day)
6 studies, a trend perhaps most notable when considering the children's diaries (a decrease in time
7 spent outdoors of about 30 minutes over the three decades). However, the coefficient of variation
8 (COV) for each of the age groups and across all decades for the cross-sectional data was
9 consistently about 40%, supporting a general conclusion of no appreciable differences in the
10 mean time spent outdoors over the three decades of data collection. Thus, when considering all
11 diaries having at least 2 hours of afternoon outdoors time and the representativeness of the
12 CHAD study data from the 1980s, inclusion of these earlier diaries is also unlikely to have a
13 strong adverse influence on exposure modeling outcomes. Though combined with the higher
14 participation rate for these earlier diaries, exposures estimated using these diaries may be higher
15 than when estimated when excluding these diaries from CHAD.
16 5.4.1.4 Evaluation of Afternoon Time Spent Outdoors for A TUS Survey Participants
17 We evaluated recent year (2002-2011) time expenditure data from the American Time
18 Use Survey (ATUS) (US BLS, 2012). As was done with the CHAD data set, the purpose was to
19 evaluate trends (if any) in outdoor time over the period of time data were collected. A few
20 strengths of the ATUS data are (1) its recent and ongoing data collection efforts, (2) large sample
21 size (totaling over 120,000 diary days), (3) national representativeness, and (4) that varying diary
22 approaches would not be an influential or confounding factor in evaluating trends over time.
23 ATUS does however have a few noteworthy limitations when compared with the CHAD
24 data: (1) there are no survey participants under 15 years of age, (2) time spent at home locations
25 is neither distinguished as indoors or outdoors, and (3) missing or unknown location data can
26 comprise a significant portion of a persons' day (on average, about 40% (George and McCurdy,
27 2009)). To overcome the limitation afforded by the ambiguous home location, we identified
28 particular activity codes most likely to occur outdoors (e.g., participation in a sport) to better
29 approximate each ATUS individual's outdoor time expenditure. Missing time was circumvented
30 by our focused analysis: about 85% of missing time information occurs outside of the hours of
31 interest here (i.e., before 12:00 PM and after 8:00 PM). Data were stratified by the same five age
32 groups as was done for the CHAD data, though here the time trends were assessed over
33 individual survey years.
34 When considering person-days having at least 2 hours of time spent outdoors, there were
35 no clear trends over the nine year ATUS study period regarding either the participation rate or
36 the mean time spent outdoors for any of the age groups. Consistent with CHAD, the participation
5-38
-------
1 rate of children was greater than that of the other age groups. The range in ATUS diary outdoor
2 participation rate (10-20%) for all age groups is lower than that observed for the CHAD data
3 (generally between 20-40%), while the range in mean time spent outdoors (190-240 minutes per
4 day) was similar to that of the CHAD data. The lower participation rate for ATUS participants is
5 not surprising given the lack of distinction regarding time indoors and outdoors while at home
6 for ATUS participants and possibly influenced in part by not having any activity patterns for
7 children under 15 years old. Overall, results of the ATUS data analysis generally support the
8 representativeness of the CHAD data, and while participation in outdoor activities calculated
9 using ATUS diaries was less than CHAD diaries, ATUS survey methods obfuscate the strength
10 of this finding.
11 5.4.1.5 Evaluation of Outdoor Time and Exertion Level for Asthmatics and Non-Asthmatics in
12 CHAD
13 Due to limited number of CHAD diaries with survey requested health information, all
14 CHAD diaries are assumed appropriate for any APEX simulated individual (i.e., whether
15 asthmatic, non-asthmatic, or no compromising health condition was indicated), provided they
16 concur with age, sex, temperature, and day-of-week selection criteria. In general, the assumption
17 of modeling asthmatics similarly to healthy individuals (i.e., using the same time-location-
18 activity profiles) is supported by the activity analyses reported by van Gent et al. (2007) and
19 Santuz et al. (1997), though other researchers, for example, Ford et al., (2003), have shown
20 significantly lower leisure time activity levels in asthmatics when compared with persons who
21 have never had asthma. To provide additional support to the assumption that any CHAD diary
22 day can be used to represent the asthmatic population regardless of the study participants'
23 characterization of having asthma or not, we first compared participation in afternoon outdoor
24 activities at elevated exertion levels among asthmatic, non-asthmatic, and unknown health status
25 using the CHAD diaries. We then compared compatible CHAD diary days with literature
26 reported outdoor time participation at varying activity levels.
27 In the first comparison, participation in afternoon outdoor activities for non-asthmatic
28 children and adults in CHAD were found similar when compared with their respective asthmatic
29 cohorts (both about 40-50%). Outdoor participation rate for persons having unknown asthma
30 status, a smaller fraction of the total diaries, varied ±10% from that having known asthma status
31 (children were higher, adults were lower). The amount of time spent outdoors by the persons that
32 did so varied little across the two populations and three asthma categories. On average, CHAD
33 diaries from children indicate approximately 2Vi hours of afternoon time is spent outdoors, 80%
34 of which is at a moderate or greater exertion level, again regardless of their asthma status, known
35 or unknown. Slightly less afternoon time is spent outdoors by adults when compared with
36 children, and while their participation in moderate or greater exertion level activities is much less
5-39
-------
1 (about 63%), there was little difference between asthmatic adults and non-asthmatic adults
2 considering outdoor time or percent at moderate or greater exertion.
3 For the second comparison, the percentage of waking hours outdoors at varying activity
4 levels for asthmatics reported in three independent asthma activity pattern studies (Shamoo et al.,
5 1994; EPRI, 1988; EPRI1992) were compared to CHAD diary days having similar personal
6 attributes and stratified by asthma status. The range in the percent of waking hours outside at
7 moderate activity level for CHAD diaries was similar to that estimated using the three
8 independent literature sources (2-10%), however the range in percent of outdoor time associated
9 with strenuous activities using the CHAD asthmatic diaries extends beyond that of asthmatic
10 persons from the three independent studies by about a factor of two higher. At this time, the
11 reason for this difference is unknown. Overall, given the above mentioned similarities in outdoor
12 time, participation, and activity levels, use of a CHAD diary regardless of a persons' asthma
13 condition is reasonably justified based on the available data analyzed.
14 5.4.2 Characterization of Factors Influencing High Exposures
15 We investigated the factors that influence persons experiencing the highest daily
16 maximum 8-hr average exposures. These exposure results in six selected study areas, Atlanta,
17 Boston, Denver, Houston, Philadelphia, and Sacramento, considering base air quality and air
18 quality just meeting the existing standard were combined with each simulated individual's
19 microenvironmental time expenditure during the afternoon hours (12:00 PM through 8:00 PM),
20 times of day commonly when daily peak high O^ concentrations occur. We first evaluated the
21 relative contribution seven variables21 had on the total explained variability in daily maximum 8-
22 hr average exposures. We then evaluated the distribution of identified influential variables for
23 simulated individuals with the highest exposures. And finally, we identified the
24 microenvironmental locations highly exposed persons occupied and the activities performed
25 within them, given that within an 8-hr time frame most persons would likely visit multiple
26 locations and perform different activities.
27 When considering only person days having the highest daily maximum 8-hr average Os
28 exposures at any of the six study areas and either air quality scenario and age groupings,
29 collectively the main effects of ambient concentrations and outdoor time combined with their
30 interaction similarly contribute to approximately 80% of the total explained variance results,
31 suggesting that for highly exposed persons, the most important influential factors are time spent
32 outdoors corresponding with high daily maximum 8-hr average ambient Oi concentrations.
21 The seven variables include the main effects of (1) daily maximum 8-hr ambient O3, (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 ambient concentration and (7) PAI by afternoon
time outdoors.
5-40
-------
1 The distributions of afternoon outdoor time and ambient concentration for highly exposed
2 individuals were evaluated considering base air quality and air quality adjusted to just meeting
3 the existing standard. As an example, exposure results in Boston indicated that for about half of
4 the days, simulated school-age children experiencing high exposures spend about 240 minutes
5 outdoors during the afternoon hours along with experiencing daily maximum 8-hr average
6 ambient Os concentrations > 75 ppb. In contrast when adjusting ambient concentrations to just
7 meeting the existing standard, for about half of the days, simulated school-age children
8 experiencing similar high exposures need to spend about 280 minutes outdoors during the
9 afternoon hours along with experiencing daily maximum 8-hr average ambient Os concentrations
10 > 60 ppb. Simply put, under conditions of lower ambient concentrations, persons need to spend a
11 significantly greater amount of time outdoors to experience similar exposures observed at higher
12 ambient concentration conditions.
13 When considering these highly exposed children, on average about half of children's total
14 afternoon time is spent outdoors on high exposure days, 40% is spent indoors, while only 10% of
15 time is spent near-roads or inside motor vehicles. In general, greater than half of the time highly
16 exposed children spent outdoors specifically involves performing a moderate or greater exertion
17 level activity, such as a sporting activity. While apportionment of afternoon microenvironmental
18 time was similar for highly exposed adults in other age groups considered (e.g., 19-35),
19 important high exertion activities performed outdoors also included those associated with paid
20 work and performing chores.
21 5.4.3 Exposure Results for Additional At-Risk Populations and Lifestages, Exposure
22 Scenarios, and Air Quality Input Data Used
23 5.4.3.1 Exposures Estimated for All School-age Children During Summer Months, Neither
24 Attending School or Performing Paid Work
25 As mentioned earlier in describing the longitudinal approach used in the main body of the
26 exposure assessment, the sequence of activity diaries for all simulated individuals is determined
27 by a user-selected profile variable of interest. In this assessment our longitudinal diary approach
28 uses time spent outdoors to link together CHAD diary days, an attempt to appropriately balance
29 intra- and inter-personal variability in that variable. For the primary exposure results, all
30 available diaries were used in developing any one sample pool without restriction outside of the
31 particular characteristics on interest in developing the pool (i.e., age, sex, day-of-week,
32 temperature, time spent outdoors). In this targeted simulation in Detroit during three summer
33 months of 2007 (June, July, and August), we restricted the diary pool of all school-age children
34 to include only those diary days that did not have any time spent inside a school nor had time
35 spent performing paid work during any day of the week. The results of this targeted simulation
36 were compared to an identical simulation, only differing in that all CHAD diary days were used
5-41
-------
1 i.e., including any diary day for persons having school time or paid work, and as was done for
2 the main body of this exposure assessment.
3 Figure 5-10 indicates that when restricting the CHAD diary pool to include only those
4 diaries having no time spent at school or performing paid work activities, there is about 1/3 or
5 33% increase in the number of all school-age children at or above the 60 ppb-8hr benchmark, a
6 relationship also consistent across the alternative standards and when considering multiple
7 exposures. A similar relationship was found for the other benchmarks (not shown, see Appendix
8 5-G). Clearly, based on the analysis results reported in section 5.4.2 regarding factors influencing
9 those highly exposed, using only activity pattern data that do not include school or work-related
10 events (which would likely occur more so indoors than outdoors) and sampling from a pool of
11 diaries consistent with summer temperatures would increase the likelihood simulated individuals
12 spend time outdoors and be exposed to concentrations at or above the selected benchmarks.
>= 1 Exposure-All CHAD Diaries
>= 1 Exposure-NoSchool/Work Diaries
>= 2 Exposures-All CHAD Diaries
>= 2 Exposures-NoSchool/Work Diaries
>= 3 Exposures-All CHAD Diaries
>= 3 Exposures-NoSchool/Work Diaries
0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20%
Percent of Children with 8-hr Daily Max Exposure > 60 ppb
13
14
15
16
17
standard level (ppb)
60
165
70
75
Figure 5-10 Comparison of the percent of all school-age children having daily maximum 8-
hr average 63 concentration at or above 60 ppb during June, July, and August in Detroit
2007: using any available CHAD diary ("All CHAD Diaries") or using CHAD diaries
having no time spent in school or performing paid work ("No School/Work Diaries").
5-42
-------
1 5.4.3.2 Exposures Estimated for Outdoor Workers During Summer Months
2 A targeted APEX simulation was performed for the Atlanta study area to simulate
3 summertime exposures for two hypothetical outdoor worker study groups, persons between the
4 age 19-35 and 36-55, using 2006 air quality just meeting the existing standard. To do this, both
5 the daily and longitudinal activity patterns used by APEX were adjusted to best reflect patterns
6 expected for outdoor workers (e.g., a standardized work schedule during weekdays) while also
7 maintaining variability in those patterns across various occupation types. Briefly, the distribution
8 of all employed persons' occupations was estimated using data provided by the U.S. Bureau of
9 Labor and Statistics (US BLS, 2012b)22 and linked with 144 occupation titles from the
10 Occupational Information Network (O*NET)23 identified as having one or more days per week
11 where paid work was performed outdoors. These data were then aggregated to twelve broadly
12 defined BLS occupation groups, generating a data set containing the number of days per week
13 work time would be performed outdoors by that occupation group and properly weighted to
14 reflect the population distribution of persons employed in each outdoor work group. Then,
15 existing CHAD diary days reflecting outdoor paid work were identified, isolated and replicated
16 to reflect this BLS/O*NET outdoor participation rate and occupation group frequencies. A
17 10,000 person simulation was performed by APEX using this adjusted CHAD activity pattern
18 database designed to simulate outdoor workers and compared with exposure results generated
19 from an identical APEX simulation of all employed persons, though differing by using the
20 standard CHAD database and population-based modeling approach used in the main body REA.
21 Details regarding the development of CHAD activity patterns used as input to simulate outdoor
22 workers, as well as other settings and conditions for APEX is described in Appendix 5G.
23 Estimated exposures are presented in Figure 5-11 for one of two age study groups
24 investigated (results for both age groups were similar) and considering either a longitudinal
25 approach designed specifically to reflect an outdoor worker weekday schedule (left panel) or
26 when using our general population-based modeling approach (right panel). The results indicate
27 that when accounting for a structured schedule that includes repeated occurrences of time spent
28 outdoors for a specified study group, all while simulated individuals are likely to be more
29 consistently performing work tasks that may be at or above moderate or greater exertion levels,
30 there are a greater percent of the study group experiences exposures at or above the selected
31 health effect benchmark levels than that estimated using our general population-based modeling
32 approach. Keep in mind outdoor workers are expected to experience more exposures at or above
33 benchmark levels, though represent a fraction of the total employed population. It is possible
22 U.S. employment data by SOC codes were obtained from: http://www.bls.gov/emp/#tables: Table 1.2
Employment by occupation, 2010 and projected 2020.
23 Additional information is available at http://www.onetonline.org.
5-43
-------
1 that, in using the general population-based approach along with the longitudinal algorithm that
2 accounts for within and between variability in outdoor time, a number of outdoor workers are
3 incidentally simulated and represent a significant portion of those who experienced exposures at
9/1
4 or above benchmark levels. However, the differences between exposures estimated for the two
5 longitudinal approaches become much greater when considering the percent of persons
6 experiencing multiple exposure days at or above benchmark levels, primarily when considering
7 the 60 ppb-8hr benchmark level. For example, < 2% of the general population-based exposure
8 group was estimated to have two or more exposures at or above 60 ppb-8hr, while >17% of
9 specifically simulated outdoor workers were estimated to experience exposures at or above that
10 same level.
24 In this outdoor worker exposure scenario, approximately 30% of our outdoor worker study group ages 19-55 were
estimated to experience at least one exposure at or above 60 ppb-8hr while at moderate or greater exertion.
Assuming outdoor workers constitute approximately 12% of the workforce (Appendix G, Table 5G-8), outdoor
workers experiencing at least one exposure at or above 60 ppb-8hr could contribute 3.6% to a total exposed
population (i.e., 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
between 48-75% of persons experiencing exposures above the 60 ppb benchmark have similar activity pattern
characteristics as outdoor workers.
5-44
-------
Outdoor Worker Scenario-based Approach (ages 19-35)
Atlanta, 2006, Just Meet Existing 75 ppb Standard
123456
Number of Times Bench mark Exceeded (June-August 2006)
General Population-based Approach (ages 19-35)
Atlanta, 2006, Just Meet Existing 75 ppb Standard
< ^
i a '«
OJ OJ 4J
si 2
"
123456
Number of Times Bench mark Exceeded (June-August 2007)
1 Figure 5-11 Percent of workers between ages 19-35 experiencing exposures at or above
2 selected benchmark levels while at moderate or greater exertion using an outdoor worker
3 approach (left panel) and a general population-based approach (right panel) considering
4 air quality adjusted to just meet the existing standard in Atlanta, GA, Jun-Aug, 2006.
5
6 5.4.3.3 Exposures Estimated for All School-age Children When Accounting for Averting
1 Behavior
8 A growing area of air pollution research involves evaluating the actions persons might
9 perform in response to high Os concentration days (ISA, section 4.1.1). Most commonly termed
10 averting behaviors, they can be broadly characterized as personal activities that either reduce
11 pollutant emissions or limit personal exposure levels. The latter topic is of particular interest in
12 this REA due to the potential negative impact it could have on 63 concentration-response (C-R)
13 functions used to estimate health risk and on time expenditure and activity exertion levels
14 recorded in the CHAD diaries used by APEX to estimate Os exposures. To this end, we have
15 performed an additional review of the available literature here beyond that summarized in the
16 ISA to include several recent technical reports that collected and/or evaluated averting behavior
17 data (Graham, 2012). The purpose was to generate a few reasonable quantitative approximations
18 that allow us to better understand how averting behavior might affect time-location-activity
19 patterns, and then simulate how such personal adjustments might affect our population exposure
20 estimates.
21 Based on the elements evaluated in our literature review (i.e., air pollution awareness,
22 prevalence and duration of averting response), we conclude that most people are aware of alert
23 notification systems (in particular those persons having compromised health and reside in an
24 urban area). We approximate that 30% of all asthmatics (or 15% of the general population) may
25 reduce their outdoor activity level on alert days (e.g., KS DOH, 2006; McDermott et al., 2006;
26 Wen et al., 2009; Zivin and Neidell, 2009) and that outdoor time/exertion during afternoon hours
27 may be reduced by about 20-40 minutes in response to an air quality alert notification
5-45
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
1
2
3
4
(Bresnahan et al., 1997; Mansfield et.al, 2006, Neidell, 2010; Sexton, 2011). We used these
literature derived estimates to generate an adjusted activity diary pool used by APEX to simulate
a 2-day exposure period (August 1-August 2, 2007) in Detroit to approximate the effect 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
5-12, left 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
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 5-12, right panel).
60 70
8-hour Exposure Benchmark (ppb)
Figure 5-12 Percent of all school-age children (left panel) and asthmatic school-age
children (right panel) having daily maximum 8-hr average Os concentration at or above
benchmark levels during a 2-day simulation in Detroit, base air quality, August 1-2, 2007.
Red bars indicate exposure results when considering effect of averting.
6 5.4.3.4 Comparison of APEX Estimated Exposures Using Three Different Base Case Air
1 Quality Data Sets: AQS, VNA, andEVNA
8 For this exposure assessment, we elected to use a modeling approach to estimate the
9 ambient input concentration field and better account for spatial gradients that may exist (Chapter
10 4). To support the selection of VNA, we compared exposure results separately generated using
11 ambient monitor (AQS), eVNA, and VNA as input to APEX for three study areas: Atlanta,
12 Detroit, and Philadelphia. All APEX settings were generally consistent with the simulations
13 discussed previously, though the air quality data differed in that the year selected was 2005
14 (based on the available CMAQ data) and that a 4 Km grid was used to define the spatial area for
15 this evaluation rather than census tracts. Daily maximum 8-hr average exposures were estimated
5-46
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
for asthmatic school-age children residing in the same census tracts comprising each air quality
domain and summarized in Figure 5-13.
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 persons 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).
Atlanta
Philadelphia
60 70 80 60 70 80 60 70 80
Daily Maximum 8-hr Average Ozone Exposure Benchmark (ppb)
Figure 5-13 Comparison of APEX exposure results generated for three study areas
(Atlanta, Detroit, and Philadelphia) using three different 2005 air quality input data
sets: AQS, VNA, and eVNA.
5-47
-------
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
5.4.3.5 Comparison of APEX Estimated Exposures Using Two Different Adjusted Air Quality
Data Sets: Quadratic Rollback andHDDM
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 current 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 first draft REA.
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 5-14.
Quadratic Rollback Approach
% of Asthmatic School-Age Children with 1 or more
Exposures > Benchmark while at Moderate or
Greater Exertion and Just Meeting Existing Standarc
^ h
3 w cn u> ro i
X ^s ^s ^s ^s c
-
|
|
|
HDDM-Model Simulation
—
n
i i — i
60 70 80 60 70 80
Daily Maximum 8-hr Average Ozone Exposure Benchmark (ppb)
Figure 5-14 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 (right panel).
The quadratic adjusted air quality resulted in slightly fewer percent of asthmatic school-
age children exposed at or above the highest benchmark (80 ppb-8hr) 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 benchmark (60 ppb-8hr)
using the quadratic approach. This is because the quadratic approach generally targets the highest
5-48
-------
1 concentrations for adjustment, while the HDDM approach accounts for changes across the full
2 concentration distribution to meet the adjusted concentration level of interest.
3 5.4.4 Limited Performance Evaluations
4 5.4.4.1 Personal Exposure Comparisons
5 A new evaluation of APEX was performed using a subset of personal 63 exposure
6 measurements obtained from the Detroit Exposure and Aerosol Research Study (DEARS) (Meng
7 et. al, 2012). For five consecutive days, personal Os outdoor concentrations along with daily
8 time-location activity diaries were collected from 36 adult study participants in Wayne County
9 Michigan during July and August 2006. An APEX simulation was performed considering these
10 same geographic and temporal features, followed with the sub-setting of APEX output data
11 according to important personal attributes of the DEARS study participants (5-day collection
12 study periods, age/sex distributions, outdoor time, ambient concentrations, and air exchange
13 rate). A comparison sample was generated randomly from the complete simulation, selecting for
14 50 APEX simulated individuals.
15 For both data sets and considering the two output variables separately (outdoor time and
16 daily exposure), the median daily values for each study participant were ranked, then plotted
17 along with each individual's corresponding minimum and maximum value using each
18 individual's 5 person-days of data (Figure 5-15). In spite of the distinct matching of influential
19 personal attributes, over 50% of APEX simulated individuals had median daily 63 exposure
20 concentrations above 10 ppb, while only 3% of DEARS participants' median values exceeded 10
21 ppb. The reason(s) for this difference is being investigated.
22
5-49
-------
100 -
90 -
0_
£40 -
I-X)
o
n -
X
x
x x y
x x y^
v A i
/
/
/
* + + +
DEARS
/7~~~
s +
+ +
+
+
p50
100 -
90 -
0_
£40 -
I*-
§20
O
n -
x x
X
X
* Xx
APEX
x
X
x
X
X
x „ x X
X
x f'
x J
x ^y
x 7
7 +
K ,-
V
X7
/ ;
/ ++
' + + +
+
++
+
•^
+ + +
+
£
+
+
+ +
+
+ ++
+
+
+
+
+
+
+
+ max
100 - =
90 --
10 15 20 25
Daily Mean O3 Exposure(ppb)
30
100 - =
90 -
80 -
;S70 -
o
5 50
10 15 20 25
DailyMean O3 Exposure(ppb)
30
10 -
10 -
120 180 240 300 360 420 480 540
Daily Total Outdoor Time (minutes)
120 180 240 300 360 420 480 540
DailyTotal Outdoor Time(minutes)
1 Figure 5-15 Distribution of daily average O3 exposures (top panels) and daily afternoon
2 outdoor time (bottom panels) and for DEARS study participants (left panels) and APEX
3 simulated individuals (right panels) in Wayne County, MI, July-August 2006.
4
5 APEX modeled exposures have previously been compared with personal exposure
6 measurements for 63 (US EPA, 2007b). Briefly, APEX 63 simulation results were compared
7 with 6-day personal 63 concentration measurements for children ages 7-12 (Xue et al., 2004;
8 Geyh et al., 2000). Two separate areas of San Bernardino County were surveyed: urban Upland
9 CA, and the combined small mountain towns of Lake Arrowhead, Crestline, and Running
10 Springs, CA. Available ambient monitoring data for these locations during the same study years
11 (1995-1996) were used as the air quality input to APEX. APEX predicted personal exposures,
12 averaged similarly across a 6-day period, matched reasonably well for much of the concentration
13 distribution considering both locations, but tended to underestimate exposures at the upper
14 percentiles of the distribution. The average difference between the 6-day means was less than 1
15 ppb, with a range of -11 ppb to +8 ppb, though predicted upper bounds for a few averaged
16 exposures having higher exposure concentrations were under-predicted by up to 24 ppb (e.g.,
17 Figure 5-16). In addition, modeled exposure concentration variability was less than that observed
18 in the personal exposure measurements. At the time of analysis, these differences were proposed
5-50
-------
1 to be largely driven by under-estimation of the spatial variability of the outdoor concentrations
2 used by APEX (US EPA, 2007b).
4
5
6
Dncentration (ppb)
K3 J^ O) 00
3 O O O O
! — i i i
0 u
me
T — "
IT ;; T][TI[IlIin Jl : ^
-~v* niiIliJslHlil^*i"^ilH?ttjilftl
[•Bissii*ftI4*tI*li?ii*lI"iA1llIJ.J-JL"'-j.lj- ^ -i
T T
range j
asured -1 -L APEX * Measured - APEX
I
i
Figure 5-16 Means (and range) of 6-day average personal O3 exposures, measured
and modeled (APEX), Upland Ca. Obtained from Figure 8-22 of US EPA (2007b).
7 5.4.4.2 Ventilation Rate Comparisons
8 The algorithm used by APEX to estimate minute-by-minute ventilation rate serves as the
9 basis for recent updates to the ventilation rate distributions provided in EPAs Exposure Factors
10 Handbook (U.S. EPA, 2009b; US EPA, 2011). During the development of the ventilation
11 distributions for EPA at that time, two peer-reviewed studies were identified as providing
12 somewhat relevant measurement data to evaluate the APEX energy expenditure and ventilation
13 algorithm (see Graham, 2009 for additional comparison details). The results of this evaluation
14 are summarized below.
15 Briefly, Brochu et al. (2006a,b) presents data for ventilation rates derived from tracking
16 doubly-labeled water (DLW) consumption/elimination to estimate energy expenditure in healthy
17 normal-weight males and females, ages from 1 month to 96 years (n=l,252). Estimates of energy
18 expended were combined with a fixed oxygen uptake factor (H=0.21) and using a fixed
r\C ^^
19 ventilatory equivalent (VQ) of 27. The DLW measurement period ranged from 7-21 days,
20 resulting in time-averaged metrics that may in some instances provide reasonable estimates for a
21 mean daily ventilation rate, but not useful for estimating variability in an individual's ventilation
22 rate over shorter time periods (as is needed by APEX). Further, while DLW is considered by
23 some as a 'gold standard' for measuring energy expenditure, this characterization would not
24 necessarily be directly transferable to approximations that use this measured value (i.e.,
25 ventilation rate in Brochu et al. (2006a,b) is a calculated value, not measured). Reported
25 The ventilatory equivalent (VQ) is the ventilation rate (VE) divided by the oxygen consumption rate (VO2)
5-51
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
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 persons were simulated by APEX to generate a
reasonable number of persons within each year of age and other potential categorical variables
(e.g., 100-200, although a few older age groups resulted in having fewer persons). 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 5-17 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 persons above age 64 and for both sexes.
ass Normalized Mean Daily
ntilation (m3/day-kg)
3OOOOOO->--
jj^cncn^bocob-
— cu u.o -
^ u.2 -
0
m o -, .
n n -
i
-
A
_I
t
A
^ •
AA i
AAA
-
-
-
A>AAAAAAAJ
ccS»Ho=a
44
XO
yyVAAAAA
UULJUUUU"
*****
! 1 ! 1
**-—
! 1 !
! 1
•APEX Females
o Brochu etal. (2006) Females
* APEX Males
^ Brochu etal. (2006) Males
i i i
j 1
axxxxxxxxJi
A
jb*M!7«ti
1. 1 !
S"!".»»
M^x •
! 1
10 20 30 40 50 60
Age (years)
70 80 90 100
Figure 5-17 Comparison of body mass normalized mean daily ventilation rates
estimated by APEX (closed symbols) and by Brochu et al., 2006 (open symbols).
5-52
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
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 5-18 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.
_>»
'«j
1.1
1.0
_ 0.9
| | 0.8
N <">
lS.
11
•^ +•«
ro
W
w
0.5
0.4
^ 5 0.3
•o 0.2
O
m 01
0.0
-
1
i
(
-
!
* 1
D \
1
$ '
i
<
k
1 \
\ !
> i
j
!
' •.
• APEX Females
o Brochu etal. (2006) Females- Corrected VQ
* APEX Males
•i Brochu etal. (2006) Males - Corrected VQ
i i i i i i i i
,
' i
i i
i — i
i
i 4
i
1 t
-*--i
! 1 1
\ /
5 i
/
3 C
* * 1
1 1 1
i
?. s
> 1
! 1
i
>. !
> (
!
^
3.
k
1
7 8 9 10 11 12 13 14 15 16 17 18 19
Age (years)
Figure 5-18 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.
5-53
-------
1 In a second study identified for comparison with APEX estimates, Arcus-Arth and
2 Blaisdell (2007) provide ventilation estimates for children <19 years of age using energy intake
3 (El, or calories consumed) and body mass data provided by the USDA's Continuing Survey of
4 Food Intake for Individuals (CSFII; USD A, 2000). Two-day daily average Els were combined
5 with a values of H (i.e., 0.22 for infants, 0.21 for non infants) and VQ (i.e., 33.5 for children 0-8,
6 30.6 for boys 9-18, 31.5 for girls 9-18 years old). Again, time-averaging of the data may provide
7 reasonable estimates of a daily mean, but offer no variability in ventilation estimates for shorter
8 durations. Furthermore, data for both sexes are combined and reported by age, with stratified
9 results by sex reported only for aggregated age groups (males and females, 9-18 years old).
10 A 2-day model simulation was performed by APEX to generate ventilation estimates for
r\r
11 children to compare with results of Arcus-Arth and Blaisdell (2007). APEX ventilation
12 estimates were time-averaged to generate mean daily values, and since the data reported in
13 Arcus-Arth and Blaisdell (2007) were not separated by sex (outside of broad age categories), the
14 APEX estimates were also combined by sex to provide a comparable mean estimate for each
15 year of age (5-18). Body mass was also not used as a categorical variable in Arcus-Arth and
16 Blaisdell (2007), therefore all APEX simulated individuals were used, regardless of whether they
17 could be classified as overweight or of normal weight. In addition, daily ventilation rates for a
18 few age groups of children were obtained from Tables 3 and 4 of Brochu et al. (2006a), though
19 considering both estimates for normal and overweight individuals (there were no combined data
20 available). The Brochu et al. (2006a) results have been corrected for VQ as noted above using
21 VQ estimates of Arcus-Arth and Blaisdell (2007) and added for comparison.
22 Figure 5-19 illustrates ventilation rate estimates from the APEX simulation, along with
23 associated data for school-age children (ages 5-18) obtained from the two publications. Daily
24 mean ventilation estimates are quite similar at each year of age, with slightly higher estimates by
25 Arcus-Arth and Blaisdell (2007) at ages 9 and above, particularly when compared with APEX
26 ventilation estimates. Ventilation estimates are remarkably similar for school-age children for all
27 three sources of data, particularly when considering the differences in the type of input data used
28 and the varied approaches of APEX, Brochu et al. (2006a), and Arcus-Arth and Blaisdell (2007).
29 This overall agreement suggests reasonable confidence can be conferred to the algorithm used by
30 APEX to estimate at a minimum, daily mean ventilation rates.
26 Table III, page 103 of Arcus-Arth and Blaisdell (2007) provided body mass normalized ventilation rates.
5-54
-------
1.0 -
_O
m no
ass Normalized Ventik
(m3/day-kg)
3 O O O O O C
J co ji. cn CD ^ b
r
£ 0.1-
O
CO nn -
_
-
-
t
-
-
-
!>----.*
c
>
^
? "---..<
, <
>— -.-J
t-.
-•-APEX Mean
o Brochuetal. (2006) Normals Mean- Corrected VQ
-*- Arcus-Arth and Blaiedell (2007) Mean
^ Brochuetal. (2006)OW/OB Mean- CorrectedVQ
^~~^~)
'::::S
v^
^
L
N i
("— 1
K-,<
""""-<
i
t j
i <
» ,
3666
i
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
l Age (years)
2 Figure 5-19 Comparison of body mass normalized daily mean ventilation rates in
3 school-age children (5-18) estimated using APEX and literature reported values.
4
5 5.4.4.3 Evaluation of Longitudinal Profile Methodology
6 We evaluated the APEX approach used for linking together cross-sectional activity
7 pattern diaries to generate longitudinal profiles for our simulated individuals (Appendix 5G,
8 Section 5G-3). Of particular interest were how well variability in outdoor participation rate and
9 the amount of time expended were represented in our population-based exposure simulations.
10 Our goal in developing the most reasonable longitudinal profiles is to capture expected,
11 important features of population activity patterns, i.e., there is correlation within an individual's
12 day-to-day activity patterns (though neither exactly repeated nor entirely random for individuals)
13 and variability across the modeled study group in day-to-day activity patterns (i.e., not every
14 simulated individual in the study group does the same activity on the same day).
15 The simulated longitudinal profiles indicate the method for linking together cross-
16 sectional diaries generates a diverse mixture of persons having variable, though expected,
17 activity patterns: A small fraction of the simulated population spend a limited amount of
18 afternoon time outdoors and occurring at a low frequency across an Os season, a small fraction
19 consistently spends a greater amount (> 2 hours) of time outdoors and occurring at greater
20 frequency (e.g., 4/5 days per week), while the remaining simulated individuals fall somewhere in
21 between regarding participation and total time. While we are not aware of a population database
5-55
-------
1 available to compare with these simulated results, we are comfortable with the method
2 performance in representing the intended variability in longitudinal activity patterns (see section
3 5G-3 for details).
4 5.5 VARIABILITY AND UNCERTAINTY
5 An important issue associated with any population exposure or risk assessment is the
6 characterization of variability and uncertainty. Variability refers to the inherent heterogeneity in
7 a population or variable of interest (e.g., residential air exchange rates). The degree of variability
8 cannot be reduced through further research, only better characterized with additional
9 measurement. Uncertainty refers to the lack of knowledge regarding the values of model input
10 variables (i.e., parameter uncertainty), the physical systems or relationships used (i.e., use of
11 input variables to estimate exposure or risk or model uncertainty), and in specifying the scenario
12 that is consistent with purpose of the assessment (i.e., scenario uncertainty). Uncertainty is,
13 ideally, reduced to the maximum extent possible through improved measurement of key
14 parameters and iterative model refinement. The approaches used to assess variability and to
15 characterize uncertainty in this REA are discussed in the following two sections. Each section
16 also contains a concise summary of the identified components contributing to uncertainty and
17 how each source may affect the estimated exposures.
18 5.5.1 TREATMENT OF VARIABILITY
19 The purpose for addressing variability in this REA is to ensure that the estimates of
20 exposure and risk reflect the variability of ambient 63 concentrations, population and lifestage
21 characteristics, associated Os exposure and dose, and potential health risk across the study area
22 and for the simulated at-risk study groups. In this REA, there are several algorithms that account
23 for variability of input data when generating the number of estimated benchmark exceedances or
24 health risk outputs. For example, variability may arise from differences in the population
25 residing within census tracts (e.g., age distribution) and the activities that may affect population
26 and lifestage exposure to Os (e.g., time spent inside vehicles, time performing moderate or
27 greater exertion level activities outdoors). A complete range of potential exposure levels and
28 associated risk estimates can be generated when appropriately addressing variability in exposure
29 and risk assessments; note however that the range of values obtained would be within the
30 constraints of the input parameters, algorithms, or modeling system used, not necessarily the
31 complete range of the true exposure or risk values.
32 Where possible, we identified and incorporated the observed variability in input data sets
33 to estimate model parameters within the exposure assessment rather than employing standard
34 default assumptions and/or using point estimates to describe model inputs. The details regarding
5-56
-------
1 variability distributions used in data inputs are described in Appendix 5B. To the extent possible
2 given the data available for the assessment, we accounted for variability within the exposure
3 modeling. APEX has been designed to account for variability in some of the input data,
4 including the physiological variables that are important inputs to determining ventilation rates.
5 As a result, APEX addresses much of the variability in factors that affect human exposure.
6 Important sources of the variability accounted for in this analysis are summarized in Appendix
7 5D.
8 5.5.2 CHARACTERIZATION OF UNCERTAINTY
9 While it may be possible to capture a range of exposure or risk values by accounting for
10 variability inherent to influential factors, the true exposure or risk for any given individual within
11 a study area is unknown, though can be estimated. To characterize health risks, exposure and risk
12 assessors commonly use an iterative process of gathering data, developing models, and
13 estimating exposures and risks, given the goals of the assessment, scale of the assessment
14 performed, and limitations of the input data available. However, significant uncertainty often
15 remains and emphasis is then placed on characterizing the nature of that uncertainty and its
16 impact on exposure and risk estimates.
17 The REA's for the previous 63, NC>2, 862, and CO NAAQS reviews each presented a
18 characterization of uncertainty of exposure modeling (Langstaff, 2007; US EPA 2008, 2009a,
19 2010). The qualitative approach used in this and other REAs is described by WHO (2008).
20 Briefly, we identified the key aspects of the assessment approach that may contribute to
21 uncertainty in the exposure and risk estimates and provided the rationale for their inclusion.
22 Then, we characterized the magnitude and direction of the influence on the assessment results
23 for each of these identified sources of uncertainty. Consistent with the WHO (2008) guidance,
24 staff scaled the overall impact of the uncertainty by considering the degree of uncertainty as
25 implied by the relationship between the source of uncertainty and the exposure concentrations. A
26 qualitative characterization of low, moderate, and high was assigned to the magnitude of
27 influence and knowledge base uncertainty descriptors, using quantitative observations relating to
28 understanding the uncertainty, where possible. A summary of the key findings of those prior
29 characterizations that are most relevant to the current Os exposure assessment are provided in
30 Table 5-6.
5-57
-------
Table 5-6 Characterization of Key Uncertainties in Historical and Current APEX Exposure Assessments
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 O3
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 2nd
draft REA, local-
scale air quality was
estimated using
VNA (see below).
5-58
-------
Sources of Uncertainty
Category
Adjustment of Air
Quality to Simulate
Just Meeting the
Current Standard
APEX: General Input
Databases
Element
Spatial Representation:
Local Scale VNA estimates
Spatial Representation:
Vertical Profile
Quadratic Approach
HDDM Simulation
Approach
Population Demographics
and Commuting (US
Census)
Historical Uncertainty Characterization
Influence of Uncertainty
on Exposure/Intake
Dose Estimates
Direction
Both
Both
Both
Both
Under
Magnitude
Low
Moderate
Low-
Moderate
Low-
Moderate
Low
Knowledge-
base
Uncertainty
Low -
Moderate
Moderate
Moderate
Low-
Moderate
Low
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 5-1 3).
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., persons outdoors).
Variable differences (e.g., none to a factor of
two or three) in the estimated number of
persons 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-
14). Variable differences remain (e.g., none to
a factor of two or three) in the estimated
percent of persons 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.
Comprehensive and subject to quality control.
Differences in 2000 data versus modeled
years (2006-2010) are likely small when
estimating percent of population exposed.
Is rating
appropriate for
current APEX C>3
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.
Yes. No further
characterization
needed.
5-59
-------
Sources of Uncertainty
Category
Element
Activity Patterns (CHAD)
Meteorological (NWS)
Poverty Status (US
Census) Weighted Asthma
Prevalence (CDC)
Historical Uncertainty Characterization
Influence of Uncertainty
on Exposure/Intake
Dose Estimates
Direction
Both
Both
Both
Magnitude
Low-
Moderate
Low
Low
Knowledge-
base
Uncertainty
Low-
Moderate
Low
Low
Comments
Comprehensive and subject to quality control.
Significantly increased number of diaries
used to estimate exposure from prior review
and 1st draft REA 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).
Remaining uncertainty with other influential
factors that cannot be accounted for (e.g.,
SES, region/local outdoor participation rates)
Comprehensive and subject to quality control,
few missing values. Limited application in
selecting CHAD diaries and AERs.
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.
Is rating
dpprupridic lur
current APEX C>3
exposure
assessment?
Yes. Newly
evaluated.
Yes. No further
characterization
needed.
New. Could possibly
use further
characterization
though typically
available local
prevalence rates
are not well
stratified by
influential variables.
5-60
-------
Sources of Uncertainty
Category
APEX:
Microenvironmental
Conc6ntrations
Element
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
Over
Both
Magnitude
Low
Low
Low
Knowledge-
base
Uncertainty
Low-
Mnrlpratp
IVIUUd QIC
Low
Moderate
Comments
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 Oj for persons
residing near roads not modeled here, but
when included, there is a small reduction
(-3%) in the number of persons 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 hr"1. 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 Oj,
exposure.
Is rating
appropriate for
current APEX C>3
exposure
assessment?
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
5-61
-------
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 cities 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 REA (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 C>3
exposure
assessment?
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
5-62
-------
Sources of Uncertainty
Category
APEX: Simulated
Activity Profiles
Element
Longitudinal Profiles
Commuting
Historical Uncertainty Characterization
Influence of Uncertainty
on Exposure/Intake
Dose Estimates
Direction
Under
Both
Magnitude
Low-
Moderate
Low
Knowledge-
base
Uncertainty
Moderate
Moderate
Comments
Depending on the longitudinal profile method
selected, the number of persons 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 REA results
does not assign rigid schedules, for example
explicitly representing a 5-day work week for
employed persons. 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 5-1 1 ),
while both percent of children experiencing
single and multiday exposures were
increased by about 30% when simulating a
rigid schedule (Figure 5-10).
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 persons exceeding benchmark levels
(section 5.3.2).
Is rating
appropriate for
current APEX C>3
exposure
assessment?
Yes. Newly
evaluated.
Yes. Newly
evaluated.
5-63
-------
Sources of Uncertainty
Category
APEX: Physiological
Processes
Exposure Benchmark
Level
Element
At-Risk Population and
Lifestages
Body Mass (NHANES)
NV02max
RMR
METS distributions
Ventilation rates
EVR characterization of
moderate or greater
exertion
Historical Uncertainty Characterization
Influence of Uncertainty
on Exposure/Intake
Dose Estimates
Direction
Both
Unknown
Unknown
Unknown
Over
Over
Over
Magnitude
Low
Low
Low
Low
Low-
Moderate
Low-
Moderate
Moderate
Knowledge-
base
Uncertainty
Low-
Moderate
Low
Low
Low
Low-
Moderate
Low-
Moderate
Low-
Moderate
Comments
An updated evaluation shows activity patterns
of asthmatics are similar to that of non-
asthmatics (section 5.3.1, Tables 5G-2 to 5G-
5).
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).
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 Figures 5-18 and 5-
19). 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 persons performing at moderate or
greater exertion and is a lower bound value
(~5th percentile), the simulated number of
persons achieving this level of exercise is
possibly overestimated.
Is rating
appropriate for
current APEX C>3
exposure
assessment?
Yes. Newly
evaluated.
Yes. No further
characterization
needed.
Yes. No further
characterization
needed.
Newly identified.
May need additional
characterization.
Yes. Given judged
impact to exposure,
additional
characterization is
needed.
Yes. Additional
characterization
would be warranted
if minute or hourly
ventilation rate data
were available.
Newly identified.
May need additional
characterization.
5-64
-------
1 5.6 KEY OBSERVATIONS
2 Two additional tables are provided to additionally summarize the exposure results across
3 all study areas and years of air quality data: Table 5-7 contains the percent of all school-age
4 children experiencing at least one exposure at or above the three exposure benchmark levels,
5 while Table 5-8 contains the percent of all school-age children experiencing at least two
6 exposures at or above the three exposure benchmark levels, with both tables considering results
7 associated with each of the adjusted air quality scenarios.27 Two descriptive statistics are
8 provided from the exposure results for each study area: the mean percent of persons exposed in
9 each study area averaged across the 5 years simulated and the maximum percent of persons
10 exposed in each study area, representing the worst year of air quality simulated. Figure 5-20
11 illustrates the estimated mean and maximum percent of all school-age children exposed for each
12 study area when considering the 60 ppb-8hr benchmark and adjusted air quality scenarios, and
13 using the data provided in Table 5-7 and Table 5-8.
14 Presented below are key observations resulting from the O^ exposure analysis:
15 • General: The estimated percent of any study group exposed at least once at or above the
16 selected benchmark levels were highest considering the base air quality though percent
17 exposed varied by study area, year, and benchmark level (Appendix 5F). Very few
18 persons within any study group (all are estimated to be < 0.3%) experienced any
19 benchmark exceedances when considering an alternative standard level of 55 ppb-8hr
20 (data not shown).
21 • Study Group: The percent of all school-age children exposed at or above the selected
22 benchmark levels across all study areas, years, and air quality scenarios were similar to
23 exposures for asthmatic school-age children (e.g., Figure 5-5 and Figure 5-6,
24 respectively) with both of these study groups having consistently higher percent of
25 persons exposed than that estimated for asthmatic adults and all older adults (Figure 5-7
26 and Figure 5-8, respectively), generally by about a factor of three or more. The percent of
27 all older adults at or above any benchmark level tended to be only a few percentage
28 points or less when compared with corresponding benchmark exceedances for asthmatic
29 adults.
30 • 80 ppb-8hr Exposure Benchmark: In general, less than 1% of any study group,
31 including all school-age children and any study area, was exposed at least once at or
32 above the highest exposure benchmark, 80 ppb-8hr, when considering the existing
The maximum sample size is 6 years based on years simulated, and for a few instances varied based on available
air quality (e.g., Chicago does not have 3 years simulated for just meeting the current standard during 2008-2010
period because air quality was below the current standard, thus the total sample size for this study area is 3.
5-65
-------
1 standard air quality scenario (Table 5-7). When considering a standard level of 70 ppb-
2 8hr, < 0.2% of any study group and any study area was exposed at least once at or above
3 that same benchmark.
4 • 70 ppb-8hr Exposure Benchmark: Less than 10% of any study group, including all
5 school-age children and any study area, was exposed at least once at or above an
6 exposure benchmark of 70 ppb-8hr, when considering the existing standard air quality
7 scenario (Table 5-7). When considering a standard level of 70 ppb-8hr, < 3.5% of any
8 study group and in any study area was exposed at least once at or above that same
9 benchmark. A standard level of 65 ppb-8hr is estimated to reduce the percent of persons
10 at or above an exposure benchmark of 70 ppb-8hr to <0.5% of any study group and in
11 any study area.
12 • 60 ppb-8hr Exposure Benchmark: In general, no more than 26% of any study group in
13 any study area was exposed at least once at or above the lowest exposure benchmark, 60
14 ppb-8hr, when considering the existing standard air quality scenario (Table 5-7, Figure
15 5-20). When considering a standard level of 70 ppb-8hr, < 20% of any study group in
16 any study area was exposed at least once at or above that same benchmark. A standard
17 level of 65 ppb-8hr is estimated to reduce the percent of persons at or above an exposure
18 benchmark of 60 ppb-8hr to < 10% of any study group and study area.
19 • Multi-day Benchmark Exceedances: When considering air quality adjusted to just meet
20 the existing standard, multi-day exposure benchmark exceedances are largely limited to
21 two or more exceedances at the 60 ppb-8hr benchmark, all occurring for < 15% of any
22 study group in any study area (e.g., Table 5-8, Figure 5-9). There were no persons
23 estimated to experience any multi-day exposures at or above 80 ppb-8hr for any study
24 group in any study area, while < 2.2% of persons were estimated to experience two or
25 more exposures at or above 70 ppb-8hr, each considering any adjusted air quality
26 scenario.
27 • Targeted Data Evaluations: Afternoon time spent outdoors, along with ambient Os
28 concentrations are the most influential factors when considering those persons highest
29 exposed. There is no apparent temporal trend in the amount of outdoor time or
30 participation rate when comparing historical CHAD diaries (1980s studies) to recently
31 collected diary data (2000s studies); regardless, majority of CHAD data are from studies
32 conducted since 2000. Use of activity pattern data from non-asthmatics to represent
33 asthmatics appears reasonably justified based on an evaluation indicating their having
34 similar outdoor time expenditure and attaining similar activity levels. APEX estimated
35 daily exposures are somewhat comparable to personal exposure measurements; however,
36 both over- and under-estimations occurred to varying degrees (Figure 5-15; Figure 5-16).
5-66
-------
1 APEX estimated ventilation rates were comparable to literature provided estimates,
2 particularly those of school-age children (Figure 5-19).
3 • Targeted Exposure Scenarios: When considering a modeling approach that more
4 rigidly schedules longitudinal time location activity patterns compared with the standard
5 longitudinal approach used by APEX, a greater percent of persons experience at least one
6 or more exposures at or above benchmark levels. For example, an APEX model
7 simulation using only summer time (no school) CHAD diary days for non-working
8 school-age children generated approximately 30% more persons at or above exposure
9 benchmark levels compared with exposures estimated using our population-based
10 modeling approach (Figure 5-10). When accounting for a fraction of the population to
11 avert in response to a bad air quality day, approximately 1-2 percentage point fewer
12 persons experienced exposures at or above benchmark levels compared with exposures
13 estimated using our population based modeling approach (Figure 5-12).
14
5-67
-------
Table 5-7 Mean and Maximum Percent of all School-age Children Estimated to
Experience at Least One Daily Maximum 8-hr Average Exposure to Os at or Above
Selected Health Benchmark Levels
Study Area
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
Adjusted
Air
Quality
Scenario
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
Percent of All School-Age Children Experiencing At Least
One Exposure At or Above Selected Benchmark Level1
60 ppb-8hr
mean
14.8
7.5
2.9
12.2
7.1
3.0
13.8
9.0
3.4
13.7
9.2
4.2
10.2
4.2
1.1
12.9
7.5
3.0
17.0
10.2
3.8
14.1
7.3
2.9
11.4
6.6
2.7
9.5
4.4
1.1
10.9
3.3
0
13.8
7.1
2.4
10.3
5.8
2.7
16.3
10.2
3.9
13.2
6.6
2.3
max
19.3
10.8
4.8
19.0
11.8
5.4
21.9
15.7
6.7
24.7
16.0
8.1
18
9.3
3.0
22.9
16.0
7.6
25.6
18.9
9.5
19.1
10.3
4.6
17.8
11.9
5.7
10.2
5.0
1.5
19.0
6.6
0.1
20.5
11.8
4.6
16.5
10.0
4.7
25.8
16.9
7.3
23.4
12.5
5.0
70 ppb-8hr
mean
2.8
0.7
0.2
2.0
0.7
0.2
2.8
1.2
0.2
3.2
1.0
0.2
1.4
0.3
0.1
1.9
0.6
0.1
1.7
0.5
0.1
2.4
0.5
0.1
2.3
0.8
0.1
0.6
0.1
0
1.6
0.2
0
2.1
0.6
0.1
1.6
0.4
0.1
3.3
1.0
0.1
2.4
0.6
0.1
max
4.4
1.4
0.5
4.0
1.2
0.3
6.6
3.2
0.5
7.5
2.7
0.4
3.7
0.9
0.2
4.5
1.5
0.3
4.1
1.7
0.4
4.2
0.9
0.2
5.5
2.1
0.4
1.0
0.2
0
3.7
0.5
0
4.2
1.5
0.3
2.7
0.9
0.2
8.1
2.7
0.4
6.0
1.4
0.2
80 ppb-8hr
mean
0.3
0.1
0
0.2
0.1
0
0.3
0.1
0
0.2
0
0
0.1
0
0
0.1
0
0
0.1
0
0
0.1
0
0
0.3
0
0
0
0
0
0.1
0
0
0.2
0
0
0.1
0
0
0.3
0.1
0
0.2
0
0
max
0.7
0.2
0
0.4
0.1
0
1.0
0.2
0
0.7
0.1
0
0.2
0
0
0.3
0.1
0
0.5
0.1
0
0.2
0
0
0.7
0.1
0
0.1
0
0
0.3
0
0
0.4
0.1
0
0.2
0
0
1.1
0.2
0
0.8
0.1
0
1 The mean is the arithmetic average of the estimated
year air quality; max is the highest estimated percent
percent of all school-age children exposed across 2006-2010
of all school-age children exposed in a year.
5-68
-------
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
Percent of All School-Age Children with at Least One 8-hr Daily Max Exposure >= 60 ppb
standard level (ppb) I I 60 L I 65 I I 70 I I 75
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
0°
~
~
~
~
~
~
I
1
^0
/ 707 81/
9°/
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
'o 0°
|
/ p/ 2o/ 3o/ 4o/ «/ go/ ?o/ go/ no/ IQO/ n'o/ 12'o/ 13'o/ ijo/ 150
Percent of All School-Age Children with at Least Two 8-hr Daily Max Exposure >= 60 ppb
standard level (ppb) ^^D 60 ^^D 65 ^^D 70 ^^D 75
Figure 5-20 Incremental increases in percent of all school-age children exposed to Os at or above 60 ppb-8hr for each study
area, year 2006-2010 air quality. Average percent (left panels), maximum percent (right panels), at least one exposure (top
panels), at least two exposures (bottom panels) per year.
5-69
-------
1
2
Table 5-8 Mean and Maximum Percent of All School-age Children Estimated to
Experience at Least Two Daily Maximum 8-hr Average Exposures to Os At or
Above Selected Health Benchmark Levels
Study Area
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
Adjusted
Air
Quality
Scenario
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
75
70
65
Percent of All School-Age Children Experiencing At Least
Two Exposures At or Above Selected Benchmark Level1
60 ppb-8hr
mean
6.0
2.1
0.4
4.6
1.8
0.4
4.5
2.2
0.4
5.3
2.5
0.8
3.1
0.9
0.1
4.8
2.2
0.5
7.6
3.5
0.7
5.0
1.9
0.4
3.8
1.5
0.3
4.1
1.6
0.3
3.4
0.5
0
5.0
1.7
0.3
3.7
1.5
0.4
7.0
3.2
0.7
5.5
2.0
0.4
max
8.9
3.3
0.8
8.4
3.7
0.9
9.7
5.5
1.1
11.6
5.7
1.8
7.5
2.6
0.5
12.2
7.1
2.0
14.4
9.2
2.8
8.6
3.6
1.1
6.3
2.9
0.7
4.5
1.8
0.3
8.0
1.4
0
8.7
3.3
0.6
7.4
3.4
0.9
13.8
7.0
2.0
12.5
5.0
1.2
70 ppb-8hr
mean
0.4
0
0
0.2
0
0
0.3
0.1
0
0.5
0.1
0
0.1
0
0
0.2
0
0
0.2
0
0
0.3
0
0
0.2
0
0
0.1
0
0
0.1
0
0
0.2
0
0
0.2
0
0
0.6
0.1
0
0.4
0
0
max
0.7
0.1
0
0.5
0.1
0
1.1
0.4
0
1.3
0.2
0
0.5
0
0
0.8
0.1
0
0.4
0.1
0
0.8
0.1
0
0.6
0.1
0
0.1
0
0
0.4
0
0
0.5
0.1
0
0.5
0.1
0
2.2
0.3
0
1.4
0.1
0
80 ppb-8hr
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
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
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0.1
0
0
0
0
0
4
5
6
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.
5-70
-------
1 5.7 REFERENCES
2 Ainsworth, B. E.; W.L. Haskell; A. S. Leon; D. R. Jacobs Jr. and H. J. Montoye; J. F. Sallis and
3 R. S. Paffenbarger Jr. 1993. "Compendium of Physical Activities: Classification of Energy
4 Costs of Human Physical Activities." Medicine & Science in Sports Exercise, 25:71-80.
5 Arcus-Arth, A. and J. Blaisdell. 2007. "Statistical Distributions of Daily Breathing Rates for
6 Narrow Age Groups of Infants and Children." Risk Analysis, 27(1):97-110.
7 Bresnahan, B. W.; M. Dickie and S. Gerking. 1997. "Averting Behavior and Air Pollution." Land
8 Economics, 73:340-357.
9 Brochu, P.; J.F. Ducre-Robitaille and J. Brodeur. 2006a. "Physiological Daily Inhalation Rates
10 for Free-living Individuals Aged 1 Month to 96 Years, Using Data from Doubly Labeled
11 Water Measurements: A Proposal for Air Quality Criteria, Standard Calculations and
12 Health Risk Assessment." Human and Ecological Risk Assessment, 12:675-701.
13 Brochu, P.; J. F. Ducre-Robitaille and J. Brodeur. 2006b. "Supplemental Material for
14 Physiological Daily Inhalation Rates for Free-living Individuals Aged 1 Month to 96
15 Years, Using Data from Doubly Labeled Water Measurements: A Proposal for Air Quality
16 Criteria, Standard Calculations and Health Risk Assessment." Human and Ecological Risk
17 Assessment, 12:1-12.
18 Burmaster, D.E. 1998. "LogNormal Distributions for Skin Area as a Function of Body Weight."
19 Risk Analysis, 18(l):27-32.
20 EPRI. 1988. A Study of Activity Patterns Among a Group of Los Angeles Asthmatics. Research
21 Project 940-5. Electric Power Research Institute. Prepared by Roth Associates, November
22 1988.
23 EPRI. 1992. A Survey of Daily Asthmatic Activity Patterns in Cincinnati. TR-101396. Research
24 Project 940-05. Electric Power Research Institute. Prepared by Roth Associates. November
25 1992.
26 Ford, E. S.; G. W. Heath; D. M. Mannino and S. C. Redd. 2003. "Leisure-time Physical Activity
27 Patterns Among U.S. Adults with Asthma." Chest, 124:432-437.
28 George, B. J. and T. McCurdy. 2009. "Investigating the American Time Use Survey from an
29 Exposure Modeling Perspective." Journal of Exposure Science and Environmental
30 Epidemiology, 21:92-105.
31 Geyh, A. S.; J. Xue, H. Ozkaynak and J. D. Spengler. 2000. "The Harvard Southern California
32 Chronic Ozone Exposure Study: Assessing Ozone Exposure of Grade-school-age Children
33 in Two Southern California Communities." Environmental Health Perspectives, 108:265-
34 270.
35 Glen, G.; L. Smith; K. Isaacs; T. McCurdy and J. Langstaff. 2008. "A New Method of
36 Longitudinal Diary Assembly for Human Exposure Modeling." Journal of Exposure
3 7 Science and Environmental Epidemiology, 18:299-311.
5-71
-------
1 Graham, S. E. and T. McCurdy. 2004. "Developing Meaningful Cohorts for Human Exposure
2 Models." Journal of Exposure Analysis and Environmental Epidemiology, 14:23-43.
3 Graham, S. E. and T. McCurdy. 2005. Revised Ventilation Rate (VE) Equations for Use in
4 Inhalation-oriented Exposure Models. Washington, DC: EPA Office of Air and Radiation.
5 (EPA document number EPA/600/X-05/008, Appendix A). Available at:
6 .
7 Graham, S. 2009. Response to Peer-review Comments on Appendix A, prepared by S. Graham
8 (U.S. EPA). Response is Appendix D of US EPA (2009b). Research Triangle Park, NC:
9 EPA Office of Air and Radiation. Available at:
10 .
11 Graham, S. E. 2012. Comprehensive Review of Published Averting Behavior Studies and
12 Available Technical Documents. Memo to Bryan Hubbell, Group Leader, Risk and
13 Benefits Group, Office of Air Quality Planning and Standards. (EPA Docket # EPA-HQ-
14 OAR-2008-0699-0085).
15 Isaacs, K. and L. Smith. 2005. New Values for Physiological Parameters for the Exposure Model
16 Input File Physiology, text memorandum submitted to the U.S. Environmental Protection
17 Agency under EPA Contract EP-D-05-065. NERL WA 10. Alion Science and Technology.
18 Found in US EPA . (2009). Risk and Exposure Assessment to Support the Review of the
19 SO2 Primary National Ambient Air Quality Standard. (EPA document number EPA-
20 452/R-09-007, August 2009). Available at:
21 .
22 Isaacs, K.; G. Glen; T. McCurdy and L. Smith. 2008. "Modeling Energy Expenditure and
23 Oxygen Consumption in Human Exposure Models: Accounting for Fatigue and EPOC."
24 Journal of Exposure Science and Environmental Epidemiology, 18:289-298.
25 Kansas Department of Health and Environment. 2006. Environmental Factors, Outdoor Air
26 Quality, and Activity Level. Results from 2005 Kansas Behavioral Risk Factor Surveillance
27 System. Office of Health Promotion, KDOH, available at:
28 .
29 Langstaff, J. E. 2007. Analysis of Uncertainty in Ozone Population Exposure Modeling,
30 OAQPS Staff Memorandum to Ozone NAAQS Review, January 31. Washington, DC:
31 Office of Air Radiation. (EPA docket number OAR-2005-0172). Available at:
32 .
33 Lioy, PJ. 1990. "The Analysis of Total Human Exposure for Exposure Assessment: A Multi-
34 discipline Science for Examining Human Contact with Contaminants." Environmental
3 5 Science and Technology, 24: 93 8-945.
36 Mansfield, C.; F. R. Johnson and G. Van Houtven. 2006. "The Missing Piece: Averting Behavior
37 for Children's Ozone Exposures." Resource Energy Economics, 28:215-228.
38 Marino, A. J.; E. N. Fletcher; R. C. Whitaker and S. E. Anderson. 2012. "Amount and
39 Environmental Predictors of Outdoor Playtime at Home and School: Across-sectional
5-72
-------
1 Analysis of a National Sample of Preschool-aged Children Attending Head Start." Health
2 & Place, 18: 1224-1230.
3 McCurdy, T. 2000. "Conceptual Basis for Multi-route Intake Dose Modeling Using an Energy
4 Expenditure Approach." Journal of Exposure Analysis and Environmental Epidemiology,
5 10:1-12.
6 McCurdy, T.; G. Glen; L. Smith and Y. Lakkadi. 2000. "The National Exposure Research
7 Laboratory's Consolidated Human Activity Database." Journal of Exposure Analysis and
8 Environmental Epidemiology, 10:566-578.
9 McDermott, M.; R. Srivastava and S. Croskell. 2006. "Awareness of and Compliance with Air
10 Pollution Advisories: A Comparison of Parents of Asthmatics with Other Parents." Journal
11 of'Asthma, 43:235-239.
12 McDonnell, W. F.; H. R. Kehrl; S. Abdul-Salaam; P. J. Ives; L. J. Folinsbee; R. B. Devlin; J. J.
13 O'Neil and D. H. Horstman. 1991. "Respiratory Response of Humans Exposed to Low
14 Levels of Ozone for 6.6 Hours." Archives of Environmental Health, 46(3): 145-150.
15 Meng, Q.; R. Williams and J. P. Pinto. (2012). "Determinants of the Associations Between
16 Ambient Concentrations and Personal Exposures to PM2.5, NO2, and Os During DEARS."
17 Atmospheric Environment, 63:109-116.
18 Montoye, H. J.; H. C. G. Kemper; W. H. N. Saris and R. A. Washburn. 1996. Measuring
19 Physical Activity and Energy Expenditure. Human Kinetics: Champaign, IL.
20 National Research Council. 1991. Human Exposure Assessment for Airborne Pollutants:
21 Advances and Opportunities. Washington, DC: National Academy of Sciences.
22 Neidell, M. 2010. "Air Quality Warnings and Outdoor Activities: Evidence from Southern
23 California Using a Regression Discontinuity Approach Design." Journal of Epidemiology
24 Community Health, 64:921-926.
25 Santuz, P.; E. Baraldi; M. Filippone and F. Zacchello. 1997. "Exercise Performance in Children
26 with Asthma: Is it Different from that of Healthy Controls?" European Respiratory
27 Journal, 10:1254-1260.
28 Schofield, W. N. 1985. "Predicting Basal Metabolic Rate, New Standards, and Review of
29 Previous Work." Human Nutrition - Clinical Nutrition, 39C(S1):5-41.
30 Sexton, A. L. (2011). "Responses to Air Quality Alerts: Do Americans Spend Less Time
31 Outdoors?" Available at:
32 .
34 Shamoo, D. A.; W. S. Linn; R. C. Peng; J. C. Solomon; T. L. Webb; J. D. Hackney and H. Hong.
35 1994. "Time-activity Patterns and Diurnal Variation of Respiratory Status in a Panel of
36 Asthmatics: Implications for Short-term Air Pollution Effects." Journal of Exposure
3 7 Analysis and Environmental Epidemiology, 4(2): 13 3 -148.
5-73
-------
1 U.S. Bureau of Labor Statistics. 2012. American Time Use Survey User's Guide. "Understanding
2 ATUS 2003 to 2011," August 2012. Data and documentation available at:
3 .
4 U.S. Census Bureau. 2007. "2000 Census of Population and Housing. Summary File 3 (SF3)
5 Technical Documentation," available at:
6 . Individual SF3 files '30' (for
7 income/poverty variables pctSO) for each state were downloaded from:
8 .
9 U.S. Department of Agriculture. 2000. "Continuing Survey of Food Intake by Individuals
10 (CSFII) 1994-96," 1998. Agricultural Research Service, CD-ROM, available at:
11 .
12 U.S. Environmental Protection Agency. 1986. Air Quality Criteria for Ozone and Other
13 Photochemical Oxidants. Research Triangle Park, NC: Office of Health and Environmental
14 Assessment, Environmental Criteria and Assessment Office. (EPA document number EPA-
15 600/8-84-020aF-eF). Available from the National Technical Information Service,
16 Springfield, VA. (NTIS publication number PB87-142949).
17 U.S. EPA. 1996a. Review of National Ambient Air Quality Standards for Ozone: Assessment of
18 Scientific and Technical Information - OAQPS Staff Paper. Research Triangle Park, NC:
19 EPA Office of Air Quality Planning and Standards. (EPA document number EPA/452/R-
20 96-007). Available at: .
21 U.S. EPA. 1996b. Air Quality Criteria for Ozone and Related Photochemical Oxidants. Research
22 Triangle Park, NC: EPA Office of Research and Development, National Center for
23 Environmental Assessment. (EPA document number EPA/600/P-93/004aF-cF). Available
24 at: .
25 U.S. EPA. 2002. Consolidated Human Activities Database (CHAD) Users Guide. Database and
26 documentation available at: .
27 U.S. EPA. 2005. Guidance on Selecting Age Groups for Monitoring and Assessing Childhood
28 Exposures to Environmental Contaminants. (EPA document number EPA/630/P-03/003F).
29 Available at: .
30 U.S. EPA. 2006. Air Quality Criteria for Ozone and Related Photochemical Oxidants (Final).
31 Research Triangle Park, NC: EPA National Center for Environmental Assessment. (EPA
32 document number EPA/600/R-05/004aF-cF). Available at:
33 .
34 U.S. EPA. 2007a. Review of National Ambient Air Quality Standards for Ozone: Policy
3 5 Assessment of Scientific and Technical Information - OAQPS Staff Paper. Research
36 Triangle Park, NC: EPA Office of Air Quality Planning and Standards. (EPA document
37 number EPA-452/R-07-007). Available at:
38
-------
1 U. S. EPA. 2007b. Ozone Population Exposure Analysis for Selected Urban Areas. Research
2 Triangle Park, NC: EPA Office of Air Quality Planning and Standards. Available at:
3 .
4 U. S. EPA. 2008. Risk and Exposure Assessment to Support the Review of the NO2 Primary
5 National Ambient Air Quality Standard. Washington, DC: EPA Office of Air and
6 Radiation. (EPA document number EPA-452/R-08-008aO, November). Available at:
7 .
8 U. S. EPA. 2009a. Risk and Exposure Assessment to Support the Review of the SO 2 Primary
9 National Ambient Air Quality Standard. (EPA document number EPA-452/R-09-007,
10 August). Available at:
11 .
12 U. S. EPA. 2009b. Metabolically Derived Human Ventilation Rates: A Revised Approach Based
13 Upon Oxygen Consumption Rates. Washington, DC: EPA National Center for
14 Environmental Assessment. (EPA document number EPA/600/R-06/129F). Available at:
15 .
16 U.S. EPA. 2010. Quantitative Risk and Exposure Assessment for Carbon Monoxide - Amended.
17 Research Triangle Park, NC: EPA Office of Air Quality Planning and Standards. (EPA
18 document number EPA-452/R-10-009, July). Available at:
19 .
20 U.S. EPA. 2011. Exposure Factors Handbook, 2011 edition. Washington, DC: EPA National
21 Center for Environmental Assessment. (EPA document number EPA/600/R-09/052F).
22 Available at: .
23 U.S. EPA. 2012a. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
24 Documentation (TRIM.Expo / APEX, Version 4.4) Volume I: User's Guide. Research
25 Triangle Park, NC: EPA Office of Air Quality Planning and Standards. (EPA document
26 number EPA-452/B-12-001a). Available at:
27 .
28 U.S. EPA. (2012b. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
29 Documentation (TRIM.Expo / APEX, Version 4.4) Volume II: Technical Support
30 Document. Research Triangle Park, NC: EPA Office of Air Quality Planning and
31 Standards. (EPA document number EPA-452/B-12-001b). Available at:
32 .
33 U.S. EPA. 2013. Integrated Science Assessment of Ozone and Related Photochemical Oxidants.
34 Research Triangle Park, NC: EPA National Center for Environmental Assessment. (EPA
35 document number EPA 600/R-10/076F). Available at:
36 .
37 van Gent, R.; K. van der Ent; L. E. M. van Essen-Zandvliet; M. M. Rovers; J. L. L. Kimpen; G.
38 de Meer and P. H. C. Klijn. 2007. "No Difference in Physical Activity in (Un)diagnosed
39 Asthma and Healthy Controls." Pediatric Pulmonology, 42:1018-1023.
5-75
-------
1 Wen, X. J.; L. Balluz and A. Mokdad. 2009. "Association Between Media Alerts of Air Quality
2 Index and Change of Outdoor Activity Among Adult Asthma in Six States," BRFSS, 2005.
3 Journal of Community Health, 34:40-46.
4 Whitfield, R.; W. Biller; M. Jusko and J. Keisler. 1996. A Probabilistic Assessment of Health
5 Risks Associated with Short- and Long-Term Exposure to Tropospheric Ozone. Argonne,
6 IL: Argonne National Laboratory.
7 World Health Organization. 2008. Uncertainty and Data Quality in Exposure Assessment. "Part
8 1: Guidance Document on Characterizing and Communicating Uncertainty in Exposure
9 Assessment." Available at:
10
-------
1 6 CHARACTERIZATION OF HEALTH RISKS BASED ON
2 CONTROLLED HUMAN EXPOSURE STUDIES
3 6.1 INTRODUCTION
4 This chapter presents information regarding the methods and results for a controlled
5 human exposure-based Os (Cb) health risk assessment that builds upon the methodology used in
6 the assessment conducted as part of the 63 NAAQS review completed in 2008 and also
7 introduces a new method for estimating risk. In the previous review, EPA conducted a health risk
8 assessment that produced risk estimates for the number and percent of school-aged children,
9 asthmatic school-aged children, and the general population experiencing lung function
10 decrements associated with Os exposures for 12 urban areas, where lung function is measured as
11 forced expiratory volume in one second (FEVi). That portion of the risk assessment was based
12 on exposure-response relationships developed from analysis of data from several controlled
13 human exposure studies which were combined with population-level exposure distributions
14 developed for children and adults. Risk estimates for lung function decrements were developed
15 for recent air quality levels and for just meeting the existing 8-hour standard and several
16 alternative 8-hour standards. The methodological approach followed in the last risk assessment
17 and risk estimates resulting from that assessment are described in the 2007 Staff Paper (U.S.
18 EPA, 2007a).
19 The goals of the current O?, risk assessment are to provide estimates of the number and
20 percentage of persons that would experience adverse respiratory effects associated with recent 63
21 levels and with meeting the existing and potential alternative Os standards in specific urban
22 areas; and to develop a better understanding of the influence of various inputs and assumptions
23 on the risk estimates. The current assessment includes estimates of risks of lung function
24 decrements in school-aged children (ages 5 to 18), asthmatic school-aged children, and the adult
25 population (19 and above). We recognize that there are many sources of uncertainty in the inputs
26 and approach used in this portion of the health risk assessment which make the specific estimates
27 uncertain, however, we have sufficient confidence in the magnitude and direction of the
28 estimates provided by the assessment for it to serve as a useful input to decisions on the
29 adequacy of the Os standard and risk reductions associated with alternative standards.
30 We are estimating lung function risk using two methodologies in this review. The
31 primary results are based on a new model that estimates FEVi responses for individuals
32 associated with short-term exposures to Os (McDonnell, Stewart, and Smith, 2012). We refer to
33 this model as the McDonnell-Stewart-Smith (MSS) model. We also provide estimates following
34 the methodology used in previous reviews which provides population level estimates of the
6-1
-------
1 percent and number of people at risk. We refer to this model as the Exposure-Response (E-R)
2 model used in previous reviews. Both of these models are implemented in the air pollution
3 exposure model APEX (EPA, 2012b,c). Following this introductory section, this chapter
4 discusses the scope of the controlled human exposure study based risk assessment, describes the
5 risk models, and provides key results from the assessment. The results of sensitivity analyses are
6 reported and key uncertainties are identified and summarized. More detailed descriptions of
7 several parts of the analyses are included in appendices that accompany the REA.
8 6.1.1 Development of Approach for Current Risk Assessment
9 The lung function risk assessment described in this chapter builds upon the methodology
10 and lessons learned from the risk assessment work conducted for previous reviews (EPA, 1996,
11 2007a). The current risk assessment also is based on the information evaluated in the ISA (EPA,
12 2013a). The general approach used in the current risk assessment was described in the Scope and
13 Methods Plan for Health Risk and Exposure (EPA, 2011), that was released to the CASAC and
14 general public in April 2011 for review and comment and which was the subject of a
15 consultation with the CASAC Oi Panel in May 2011. The first draft REA was reviewed by
16 CASAC in September 2012. The approach used in the current risk assessment reflects
17 consideration of the comments offered by CASAC members and the public on the Scope and
18 Methods Plan and the first draft REA.
19 Controlled human exposure studies involve volunteers who are exposed while engaged in
20 different exercise regimens to specified levels of 63 under controlled conditions for specified
21 amounts of time. For the current health risk assessment, we are using probabilistic exposure-
22 response relationships based on analysis of individual data that describe the relationship between
23 measures of personal exposure to 63 and measures of lung function recorded in the studies.
24 Therefore, a risk assessment based on exposure-response relationships derived from controlled
25 human exposure study data requires estimates of personal exposure to ambient 63. Because data
26 on personal hourly exposures to OT, of ambient origin are not available, estimates of personal
27 exposures to varying ambient concentrations are derived through exposure modeling, as
28 described in Chapter 5. While the quantitative risk assessment based on controlled human
29 exposure studies addresses only lung function responses, it is important to note that other
30 respiratory responses have been found to be related to 63 exposures in these types of studies,
31 including increased lung inflammation, increased respiratory symptoms, increased airway
32 responsiveness, and impaired host defenses. Sufficient information is not available to
33 quantitatively model these other endpoints. Section 6.2 of the ISA provides a discussion of these
34 additional health endpoints which are an important part of the overall characterization of risks
35 associated with ambient 03 exposures.
6-2
-------
1 6.1.2 Comparison of Controlled Human Exposure- and Epidemiologic-based Risk
2 Assessments
3 In contrast to the exposure-response relationships derived from controlled human
4 exposure studies, epidemiological studies provide estimated concentration-response
5 relationships based on data collected in real world community settings. The assessment of health
6 risk based on epidemiological studies is the subject of Chapter 7. The characteristics that are
7 relevant to carrying out a risk assessment based on controlled human exposure studies versus one
8 based on epidemiology studies can be summarized as follows:
9 • The relevant controlled human exposure studies in the ISA provide data that can be used
10 to estimate exposure-response functions, and therefore a risk assessment based on these
11 studies requires as input (modeled) personal exposures to ambient O?,. The relevant
12 epidemiological studies in the ISA provide concentration-response functions, and,
13 therefore, a risk assessment based on these studies requires as input (actual monitored or
14 adjusted based on monitored) ambient O^ concentrations, and personal exposures are not
15 required as inputs to the assessment.
16 • Epidemiological studies are carried out in specific real world locations (e.g., specific
17 urban areas). To minimize extrapolation uncertainty, a risk assessment based on
18 epidemiological studies is best performed in locations where the studies took place.
19 Controlled human exposure studies, carried out in laboratory settings, are generally not
20 specific to any particular real world location. A risk assessment based on controlled
21 human exposure studies can therefore appropriately be carried out for any location for
22 which there are adequate air quality and other data on which to base the modeling of
23 personal exposures.
24 • To derive estimates of risk from concentration-response relationships estimated in
25 epidemiological studies, it is usually necessary to have estimates of the baseline
26 incidences of the health effects involved. Such baseline incidence estimates are not
27 needed in a controlled human exposure studies-based risk assessment.
28 6.2 SCOPE OF LUNG FUNCTION HEALTH RISK ASSESSMENT
29 The current controlled human exposure-based 63 health risk assessment is one approach
30 used to estimate risks associated with exposure to ambient Os in a number of urban areas
31 selected to illustrate the public health impacts of this pollutant. The short-term exposure related
32 health endpoints selected for this portion of the 63 health risk assessment include those for which
33 the ISA concludes that the evidence as a whole supports the general conclusion that 03, acting
34 alone and/or in combination with other components in the ambient air pollution mix is causal or
35 likely to be causally related to the endpoint.
36 In the 2007 03 NAAQS review, the controlled human exposure-based health risk
37 assessment involved developing risk estimates for lung function decrements (> 10, > 15, and
38 > 20% changes in FEVi) in school-aged children (ages 5 to 18 years old). The strong emphasis
6-3
-------
1 on children reflects the finding of previous O^ NAAQS reviews that children are an important at-
2 risk group. Due to the increased amount of time spent outdoors engaged in relatively high levels
3 of physical activity (which increases intake), school-aged children as a group are particularly at
4 risk for experiencing Os-related health effects.
5 Outdoor workers and other adults who engage in moderate exertion for prolonged
6 periods or heavy exertion for shorter periods during the day also are clearly at risk for
7 experiencing similar lung function responses when exposed to elevated ambient Os
8 concentrations. In this second draft REA, we focus the quantitative risk assessment for lung
9 function decrements on all and asthmatic school-aged children (ages 5-18), and the adult
10 population (ages 19 and above).
11 For the second draft assessment, lung function risks are estimated for 15 cities, Atlanta,
12 Baltimore, Boston, Chicago, Cleveland, Dallas, Denver, Detroit, Houston, Los Angeles, New
13 York, Philadelphia, Sacramento, St. Louis, and Washington, DC.
14 6.2.1 Selection of Health Endpoints
15 The ISA identifies several responses to short-term 63 exposure that have been evaluated
16 in controlled human exposure studies (US EPA, 2013, sections 6.2.1.1, 6.2.2.1, 6.2.3.1, and
17 6.3.1). These include decreased inspiratory capacity; decreased forced vital capacity (FVC) and
18 forced expiratory volume in one second (FEVi); mild bronchoconstriction; rapid, shallow
19 breathing patterns during exercise; symptoms of cough and pain on deep inspiration (PDI);
20 increased airway responsiveness; and pulmonary inflammation. Such studies provide direct
21 evidence of relationships between short-term Os exposure and an array of respiratory-related
22 effects, however, there are only sufficient exposure-response data at different concentrations to
23 develop quantitative risk estimates for (Vrelated decrements in FEVi. Other responses to 63
24 which may be equally or more important then FEVi decrements (e.g., inflammation) do not
25 necessarily correlate with FEVi responses (ISA, section 6.2.3.1) and this risk assessment is not
26 able to address these other responses.
27 As stated in the 2006 Criteria Document (Table 8-3, p.8-68) for adults with lung disease,
28 even moderate functional responses (e.g., FEVi decrements > 10% but < 20%) would likely
29 interfere with normal activities for many individuals, and would likely result in more frequent
30 medication use. In a recent letter to the Administrator, the CASAC 63 Panel stated that
31 '"Clinically relevant' effects are decrements > 10%, a decrease in lung function considered
32 clinically relevant by the American Thoracic Society" (Samet, 2011, p.2). The CASAC O3 Panel
33 also stated that:
34 a 10% decrement in FEVi can lead to respiratory symptoms, especially in
35 individuals with pre-existing pulmonary or cardiac disease. For example,
36 people with chronic obstructive pulmonary disease have decreased ventilatory
6-4
-------
1 reserve (i.e., decreased baseline FEVi) such that a > 10% decrement could lead
2 to moderate to severe respiratory symptoms (Samet, 2011, p.7).
3
4 This is consistent with the most recent official statement of the American Thoracic Society on
5 what constitutes an adverse lung function health effect of air pollution:
6 The committee recommends that a small, transient loss of lung function, by
7 itself, should not automatically be designated as adverse. In drawing the
8 distinction between adverse and nonadverse reversible effects, this committee
9 recommended that reversible loss of lung function in combination with the
10 presence of symptoms should be considered adverse (ATS, 2000, p.672).
11
12 For this lung function risk assessment, a focus on the mid- to upper-end of the range of
13 moderate levels of functional responses and higher (FEVi decrements > 15%) is appropriate for
14 estimating potentially adverse lung function decrements in active healthy adults, while for people
15 with asthma or lung disease, a focus on moderate functional responses (FEVi decrements down
16 to 10%) may be appropriate.
17 6.2.2 Approach for Estimating Health Risk Based on Controlled Human Exposure
18 Studies
19 The major components of the health risk assessment based on data from controlled
20 human exposure studies are illustrated in Figure 3-3 in Chapter 3. As shown in this figure, under
21 this portion of the risk assessment, exposure estimates for a number of different air quality
22 scenarios (i.e., recent year of air quality, just meeting the existing 8-hour and alternative
23 standards) are combined with probabilistic exposure-response relationships derived from the
24 controlled human exposure studies to develop risk estimates associated with recent air quality
25 and after simulating just meeting the existing and alternative standards. The health effect
26 included in this portion of the risk assessment is lung function decrement, as measured by
27 changes in FEVi. The population risk estimates for a given lung function decrement (e.g., > 15%
28 reduction in FEVi) are estimates of the expected number of people who will experience that lung
29 function decrement, the number of times that people experience repeated occurrences of given
30 lung function decrements, and the number of occurrences (person-days) of the given lung
31 function decrement. The air quality and exposure analysis components that are integral to this
32 portion of the risk assessment are discussed in Chapters 4 and 5.
33 We used two approaches to estimate health risk. As done for the risk assessment
34 conducted during the previous O?, NAAQS review, a Bayesian Markov Chain Monte Carlo
35 approach was used to develop probabilistic exposure-response functions. These functions were
36 then applied to the APEX estimated population distribution of 8-hour maximum exposures for
37 persons at or above moderate exertion (> 13 L/min-m2 body surface area) to estimate the number
6-5
-------
1 of persons expected to experience lung function decrements. The primary approach, based on the
2 McDonnell-Stewart-Smith FEVi model, uses the time-series of 63 exposure and corresponding
3 ventilation rates for each APEX simulated individual to estimate their personal time-series of
4 FEVi reductions, selecting the daily maximum reduction for each person. A key difference
5 between these approaches is that the previous method estimates a population distribution of
6 FEVi reductions, where the MSS model estimates FEVi reductions at the individual level. Each
7 of these approaches is discussed in detail below.
8 6.2.3 Controlled Human Exposure Studies
9 Modeling of risks of lung function decrements as a function of exposures to 63 is based
10 on application of results from controlled human exposure studies. As discussed in Chapter 6 of
11 the ISA (EPA, 2013a), there is a significant body of controlled human exposure studies reporting
12 lung function decrements and respiratory symptoms in adults associated with 1- to 8-hour
13 exposures to O?,. In the ISA sections on controlled human exposure (Sections 6.2.1.1, 6.2.2.1,
14 6.2.3.1, and 6.3.1) over 140 references to human clinical studies are reported.
15 6.2.3.1 Life Stages
16 Consistent with the approach used in the previous 63 NAAQS review and lacking a
17 significant body of controlled human exposure studies on children, we judge that it is reasonable
18 to estimate exposure-response relationships for lung function decrements associated with 03
19 exposures in children 5-18 years old based on data from young adult subjects (18-35 years old).
20 As discussed in the ISA (EPA, 2013a), findings from clinical studies for children and summer
21 camp field studies of children 7-17 years old in at least six different locations in the U.S. and
22 Canada found lung function decrements in healthy children similar to those observed in healthy
23 young adults exposed to 03 under controlled chamber conditions. There are fewer studies of
24 young children than adolescents to draw upon, which may add to uncertainties in the modeling.
25 Additional uncertainties are likely introduced since the lungs and airways of children are
26 developing, while development is complete in adults (Dietert et al., 2000). The primary period of
27 alveolar development is from birth to around eight years of age, but there is evidence for
28 continued development through adolescence. The adult number of alveoli is reached by 2-3
29 years of age and the size and surface area of the alveoli increase until after adolescence (Hislop,
30 2002; Narayanan etal., 2012).
31 Lung function responses to O?, exposure for adults older than 18 decrease with age until
32 around age 55, when responses are minimal. "Children, adolescents, and young adults appear, on
33 average, to have nearly equivalent spirometric responses to Os, but have greater responses than
34 middle-aged and older adults when similarly exposed to Os" (ISA p. 6-21). "In healthy
35 individuals, the fastest rate of decline in 63 responsiveness appears between the ages of 18 and
6-6
-------
1 35 years (Passannante et al., 1998; Seal et al., 1996), more so for females then males (Hazucha et
2 al., 2003). During the middle age period (35-55 years), 63 sensitivity continues to decline, but at
3 a much lower rate. Beyond this age (>55 years), acute Os exposure elicits minimal spirometric
4 changes" (ISA p. 6-23).
5 6.2.3.2 Asthma
6 There have been several controlled human exposure studies of the effects of 63 on
7 asthmatic subjects, going back to 1978 (Linn et al., 1978). In reference to these studies, the ISA
8 states that "[b]ased on studies reviewed in the 1996 and 2006 O?, AQCDs, asthmatic subjects
9 appear to be at least as sensitive to acute effects of 63 as healthy nonasthmatic subjects" (ISA p.
10 6-20). Studies published since the 2006 O3 AQCD do not alter this conclusion (ISA, p. 6-20 to 6-
11 21). In the 2010 O3 NAAQS proposal (75 FR 2969-2972), EPA describes the evidence that
12 people with asthma are as sensitive as, if not more sensitive than, normal subjects in manifesting
13 Os-induced pulmonary function decrements.
14 In reference to epidemiologic studies, the ISA states that "[t]he evidence supporting
15 associations between short-term increases in ambient 63 concentration and increases in
16 respiratory symptoms in children with asthma is derived mostly from examination of 1-h max,
17 8-h max, or 8-h avg 63 concentrations and a large body of single-region or single-city studies.
18 The few available U.S. multicity studies produced less consistent associations." (ISA, p. 6-101 to
19 6-102). "Although recent studies contributed mixed evidence, the collective body of evidence
20 supports associations between increases in ambient 63 concentration and increased asthma
21 medication use in children" (ISA, p. 6-109).
22 6233 Ethnicity
23 There are two controlled human exposure studies that have assessed differences in lung
24 function responses comparing ethnic groups (ISA, p. 6-23 to 6-24). Both of these studies show
25 greater FEVi decrements in blacks than whites, however, epidemiologic studies were less
26 supportive of this difference in response. The data available are insufficient to quantify any
27 differences that might exist due to the limited number of studies and a lack of consistency
28 between disciplines.
29 6234 Body Mass Index
30 Some studies have found greater FEVi decrements to be associated with increasing BMI.
31 BMI was included in some of the models of McDonnell et al. (2012); however, the BMI terms
32 were found to be statistically insignificant, indicating that the effect of BMI on FEVi in the
33 presence of 03 is likely to be small, within the range of BMIs of the subjects studied.
6-7
-------
1 6 2 3 5 Outdoor Workers
2 Although there are no controlled human exposure studies that have had specifically
3 outdoor workers as subjects, the studies are applicable to outdoor workers: the 6.6-hour
4 experimental protocol was intended to simulate the performance of heavy physical labor for a
5 full workday (ISA, p. 6-9).
6 6 2 3 6 Variability of Responses
7 Responses to 63 exposure are variable within the population, even within cohorts of
8 similar people (e.g., healthy young adult white males) (ISA, p. 6-16 to 6-20). Factors which
9 contribute to interindividual variability include health status, body mass index, age, sex,
10 race/ethnicity, and the intrinsic responsiveness of individuals. Other factors which contribute to
11 the variability of responses include the duration and concentration of Os exposure, the level of
12 exercise and breathing rate, attenuation due to repeated exposures, and co-exposures with other
13 pollutants. For specific individuals, lung function responses tend to be reproducible over a period
14 of several months.
15 6.2.4 The McDonnell-Stewart-Smith (MSS) Model
16 In this review, EPA is investigating the use of a new model that estimates FEVi
17 responses for individuals associated with short-term exposures to Oj (McDonnell, Stewart, and
18 Smith, 2007; McDonnell, Stewart, and Smith, 2010). This is a fundamentally different approach
19 than the previous approach, for which the exposure-response function is at a population level, not
20 an individual level. This model was developed using the controlled human exposure data
21 described in Section 6.2.5 as well as incorporating several additional data sets from studies using
22 shorter exposure durations and different exertion levels and breathing rates. These data were
23 from 15 controlled human Os exposure studies that included exposure of 541 volunteers (ages 18
24 to 35l) on a total of 864 occasions. These data are described in McDonnell et al. (1997).
25 Schelegle et al. (2009) found that there appears to be a delay in response when modeling FEVi
26 decrements as a function of cumulative dose and estimated a threshold associated with the delay.
27 McDonnell et al. (2012) refit their 2010 model using data from eight additional studies with 201
28 subjects and incorporating a threshold parameter into the model. Their threshold parameter
29 allows for modeling a delay in response until cumulative dose rate (taking into account decreases
30 over time according to first order reaction kinetics) reaches a threshold value and is found by
31 McDonnell et al. (2012) to slightly improve model fit. That latest model is the model described
1 The ages in these studies range from 18 years 1 month to 35 years 1 month.
6-8
-------
2
3
4
5
6
7
8
9
10
1 1
12
13
14
15
16
17
18
19
20
21
22
23
24
25
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 cumulative 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 cumulative 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 cumulative dose rate. Also,
the Schelegle et al. model's threshold is based on cumulative intake dose (the integral of
concentration x volume inhaled), where the MSS model's threshold is based on net cumulative
dose rate (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 cumulative amount of
exposure to 63 (exposure concentration times ventilation rate, loosely speaking a measure of
dose rate) 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)
dX/dt=C(t)V(t)-aX
x(t)
logistic
response
aX (metabolism)
Figure 6-1. Two-Compartment Model
C is exposure concentration, V is ventilation rate, t is time, X is an intermediate quantity,
a is a decay constant. Adapted from Figure 1 in McDonnell et al. (1999).
6-9
-------
1 X is given by the solution of the differential equation (6-1):
2 dX Of (Equation 6-1)
— = ccono* - /?5*(t)
at
o
4 X(^) increases with "dose" (C-r ) over time for an individual and allows for removal of
5 63 with a half-life of 1/Ps through the 2nd term in equation (6-1). In APEX, because the exposure
6 concentration, exertion level, and ventilation rate are constant over an event, this equation has an
7 analytic solution for each event ("events" in APEX are intervals of constant activity and
8 concentration, where an individual is in one microenvironment, and range in duration from one
9 to 60 minutes):
Jf(t) = X(t0)
10 5 (Equation 6-2)
1 1 This model calculates the FEVi decrement due to Os exposure (compartment 2) as:
12 ™ ™ (Equation 6-3)
13 where T^ = max{0, Xp - Pg}. Pg is a threshold parameter which allows X to increase up to the
14 threshold before the median response is allowed to exceed zero.
15 The variables in the above equations are defined as:
16 The indices i,j,k refer to the /'th subject at they'th time for the Mi experiment for that subject,
17 C(t) is the O^ exposure concentration at time t (ppm) during the event,
18 V(t) = VE(t)IBSA is the ventilation rate normalized by body surface area at time t
19 (L/min-m2),
20 VE(t) is the expired minute volume at time t (L min"1),
21 BSA is the body surface area (m2),
22 t is the time (minutes), to is the time at the start of the event,
23 Age-± is age in years of the /'th subject in the Mi study,
24 A is an age parameter (taken to be the approximate mean age of the clinical study subjects in
25 the McDonnell, Stewart, and Smith 2007 (A=25), 2010 (A=25), and 2012 (A=23.8) papers),
26 U; is a subject-level random effect (between-individual variability not otherwise captured by
27 the model), and
28 Sift is a variability term, which includes measurement error and intra-individual variability
29 not otherwise captured by the model.
6-10
-------
1 The PS and the variances of the {U;} and {Syk} are fitted model parameters (see
2 McDonnell, et al. (2007, 2010, and 2012) for details). In APEX, values of U; and sijk are drawn
3 from Gaussian distributions with mean zero and variances var(U) and var(e), constrained to be
4 within ±2 standard deviations from the means. The values of U; are chosen once for each
5 individual and remain constant for individuals throughout the simulation. The Syk are sampled
6 daily for each individual. The best fit values (based on maximum likelihood) for these
7 parameters are listed in Table 6-1. The values in parentheses are standard errors of the estimates
8 (given here to two significant digits; the values in the papers are given to up to five significant
9 digits). Although some of the parameters are quite different in the three models in Table 6-1, the
10 predictions of these three models are similar. The relative influences of the parameters are
11 discussed in Section 6.5.1.
12
Table 6-1. Estimated Parameters in the MSS Models
13
14
15
16
17
Model
P2
P4
P9
var(U) var(E)
20071,
20 102
20123
9.9047
(0.61)
9.8057
-0.4106
(0.11)
-0.1907
0.0164
(0.0030)
0.01839
46.9397
(7.3)
65.826
0.003748
(0.00027)
0.003191
0.9123
(0.054)
0.8753
0.835
(0.080)
0.9449
13.8279
(0.36)
17.120
2012T4
10.916
(0.84)
-0.2104
(0.31)
0.01506
(0.0033)
13.497
(4.7)
0.003221
(0.00021)
0.8839
(0.065)
59.284
(10)
0.9373
(0.082)
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.
18 We are using this model to estimate lung function decrements for people ages 5 and
19 older. However this model was developed using only data from individuals aged 18 to 35 and the
20 age adjustment term [Pi + PZ (Age^ - A)] in the model is not appropriate for all ages. In addition
21 to this age term, the effects of age are also taken into account through the dependence of
22 ventilation rate and body surface area on age. The APEX estimates of lung function risk for
23 different age groups are also influenced by the time spent outdoors and the activities engaged in
24 by those groups, which vary by age (see Appendix 6-E).
25 Clinical studies data for children which could be used to fit the model for children are not
26 available at this time. In the absence of data, we are extending the model to ages 5 to 18 by
27 holding the age term constant at the age 18 level. Since the response increases as age decreases
28 in the range 18 to 35, this trend may extend into ages of children, in which case the responses of
6-11
-------
1 children could be underestimated. However, the slope of the age term in the MSS model is
2 estimated based on data for ages 18 to 35 and does not capture differences in age trend within
3 this range; in particular, we don't know at what age the response peaks, which could be above or
4 below 18. The evidence from clinical studies indicates that the responsiveness of children to O^
5 is about the same as for young adults (ISA, 2012, p. 6-21). This suggests that the age term for
6 children should not be higher than the age term for young adults.
7 Because the responses to 63 decline from age 18 until around age 55 and for ages older
8 than 55 the response are minimal, we let the age term for ages 35 to 55 linearly decrease to zero
9 and set it to zero for ages > 55.
10 "In healthy individuals, the fastest rate of decline in 63 responsiveness
11 appears between the ages of 18 and 35 years .... During the middle age period
12 (35-55 years), Os sensitivity continues to decline, but at a much lower rate.
13 Beyond this age (>55 years), acute Os exposure elicits minimal spirometric
14 changes." (ISA, 2012, p. 6-23)
15 In order to extend the age term to ages outside the range of ages the MSS model is based
16 on (ages 18-35), we parameterize the age term by [Pi + $2(0-1 Age + 012)], for different ranges of
17 ages (ai and 0.2 depend on age), requiring that these terms match at each boundary to form a
18 piecewise linear continuous function of age. The foregoing assumptions result in the following
19 values of ai and 0.2 for four age ranges (Table 6-2).
20 Table 6-2. Age Term Parameters for Application of the 2012 MSS Threshold
21 Model to All Ages
Age Range
5-17
18-35
36-55
>55
Pi
10.916
10.916
10.916
0
02
-0.2104
-0.2104
-0.2104
0
«i
0
1
2.0341
0
«2
-5.8
-23.8
-59.994
0
22
23 The lung function decrements estimated by the MSS (2010) model for a particular case
24 are illustrated in Figure 6-2 and Figure 6-3. Figure 6-2 shows the predictions of the MSS model
25 for 20-year old individuals with a (typical) body surface area (BSA) of 2 m2 and a target
26 ventilation rate of 40 L/min (moderate exertion) and an O?, exposure level of 100 ppb, under the
27 conditions of a typical 6.6-hour clinical study. Subjects alternated 50 minutes of moderate
28 exercise with 10 minutes of rest for the first three hours, with the exercise occurring first. For the
29 next 35 minutes (lunch), subjects continued exposure at rest. For the remaining three hours of the
30 exposure period, subjects again alternated 50 minutes of exercise with 10 minutes of rest. The
6-12
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
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.
0
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 O3 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.
6-13
-------
7
6
I 4
o
CJ
^^3
PH
0
1
2
3
4
5
6
7
8
9
10
11
12
13
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 from hour 1 to hour 3 and at rest otherwise. There is
a 30-minute delay in response due to the threshold; without the threshold, the response starts
increasing when exercise starts. 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.
6-14
-------
LJJ
2-
1 2
Model
3 4
hour
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 O3 Exposure, 2
Hours Heavy Exercise (30 L/min-m2 BSA).
W
£H
o
PH
Model
3 4
hour
1 no-threshold threshold
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, 2
Hours Heavy Exercise (30 L/min-m2 BSA).
6-15
-------
1 6.2.5 The Exposure-Response Function Approach Used in Prior Reviews
2 As described in section 3.1.2 of the 2007 Risk Assessment Technical Support Document
3 (EPA, 2007b), a Bayesian Markov Chain Monte Carlo approach (Lunn et al., 2012) was used to
4 estimate probabilistic exposure-response relationships for lung function decrements associated
5 with 8-hour Oj, exposures occurring at moderate exertion. In the previous review, summary data
6 from the Folinsbee et al. (1988), Horstman et al. (1990), McDonnell et al. (1991), and Adams
7 (2002, 2003, 2006) studies were combined to estimate exposure-response relationships for 8-
8 hour exposures at moderate exertion for each of the three measures of lung function decrement
9 (> 10, > 15, > 20% decrements in FEVi). In this second draft REA we have updated this
10 exposure-response function with the results from two additional studies (Kim et al., 2011;
11 Schelegle et al., 2009). The controlled human exposure study data were corrected for the effect
12 of exercise in clean filtered air on an individual basis to remove any systematic bias that might
13 be present in the data attributable to an exercise effect (ISA, Section 6.2.1.1). This is done by
14 subtracting the FEVi decrement in filtered air from the FEVi decrement (at the same time point)
15 during exposure to O?,. An example of this calculation is given in Appendix 6-D.
16 Table 6-3 presents a summary of the study-specific results based on correcting all
17 individual responses for the effect on lung function decrements of exercise in clean air.
18
Table 6-3. Study-specific Os Exposure-response 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
rv V, Protocol
Average (J? Exposure
0.04 ppm O3
Adams et al. (2002)
Adams et al. (2006)
Square-wave, face mask
Triangular
Number of Responses"
Exposed A™V^ f™,>
30
30
2(2)
0(0)
0(0)
0(0)
AFEVi >
20%
0(0)
0(0)
0.06 ppm O3
Adams et al. (2006)
Kim etal. (2011)
Schelegle et al. (2009)
Square-wave
Triangular
Square-wave
Variable levels (0.06 ppm avg)
30
30
59
31
2(2)
2(2)
3(6)
4(8)
0(0)
2(2)
1(3)
2(3)
0(0)
0(0)
0(0)
1(1)
0.07 ppm 63
Schelegle et al. (2009)
Variable levels (0.07 ppm avg)
31
6(12)
3(7)
2(3)
0.0 8 ppm 63
Adams et al. (2002)
Square-wave, face mask
30
6(6)
5(5)
2(2)
6-16
-------
Table 6-3. Study-specific Os Exposure-response 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
. n _ Protocol
Average (J? Exposure
Adams et al. (2003)
Adams et al. (2006)
F-H-Mb
Kim etal. (2011)
Schelegle et al. (2009)
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)
Number
Exposed
30
30
30
30
30
30
60
30
31
Number of Responses"
AFEVi >
10%
6(6)
5(5)
6(6)
5(5)
7(7)
9(9)
17(19)
4(6)
10 (15)
AFEVi >
15%
2(2)
2(2)
1(1)
1(1)
2(2)
3(3)
11(14)
1(1)
5(8)
AFEVi >
20%
1(1)
2(2)
1(1)
1(1)
1(1)
1(1)
8(8)
0(0)
4(6)
0.087 ppm 63
Schelegle et al. (2009)
Variable levels (0.087 ppm
avg)
31
14 (17)
10(12)
7(9)
0.1 ppm Os
F-H-Mb
Square-wave
32
13(13)
11(12)
6(9)
0.1 2 ppm 63
Adams et al. (2002)
F-H-Mb
Square-wave, chamber
Square-wave, face mask
Square-wave
30
30
30
17 (17)
21 (21)
18(19)
12 (12)
13(13)
15(15)
10 (10)
7(7)
10 (10)
a. The first number in each cell is the number of responses based on post-exposure decrements in FEVi
(i.e., we used only the last FEVi measurement and the pre-exposure FEVi to obtain a single percentage change in
FEVi for each subject in each experiment). The numbers in parentheses are the numbers of responses based on
maximum FEVi decrements. Specifically, when there were multiple FEVi measurements after the beginning of the
exposure, we calculated multiple FEVi percentage changes for each subject in each experiment and used the
maximum change when calculating the numbers of responses greater than 10%, 15%, and 20%.
b. F-H-M combines data from Folinsbee et al. (1988), Horstman et al. (1990), and McDonnell et al. (1991).
1
2 For the risk assessment conducted during the 2007 O3 NAAQS review (EPA, 2007b),
3 EPA considered both linear and logistic functional forms in estimating the exposure-response
4 relationship and chose a 90 percent logistic/10 percent piecewise-linear split using a Bayesian
5 Markov Chain Monte Carlo approach. This Bayesian estimation approach incorporates both
6 model uncertainty and uncertainty due to sampling variability.
6-17
-------
1 For each of the three measures of lung function decrement, EPA assumed a 90 percent
2 probability that the exposure-response function has the following 3-parameter logistic form:2
a*er(l-efte)
3 y(X, a, p, r)~ r fr+r , (Equation 6-4)
4 where x denotes the Os concentration (in ppm) to which the individual is exposed, y
5 denotes the corresponding response (decrement in FEVi > 10%, > 15% or > 20%), and a, /?, and
6 y are the three parameters whose values are estimated.
7 We assumed a 10 percent probability that the exposure-response function has the
8 following linear with threshold (hockey stick) form:
\a + Bx, for a + Bx> 0
9 y(x;a,/3) = \ , „ _ (Equation 6-5)
[0, for a + fix < 0
10 We assumed that the number of responses, S, out of TV subjects exposed to a given
1 1 concentration, x, has a binomial distribution with response probability given by Eq (6-4) with 90
12 percent probability and response probability given by Eq (6-5) with 10 percent probability. In the
13 2007 review, we also considered 80/20 and 50/50 probabilities for the logistic and hockey stick
14 forms, and ran those as sensitivity analyses. We performed those analyses with the updated data
15 and found that for each of the three exposure-response curves, the 90/10 mix has smaller error in
16 fit (weighted RMSE) than the other two combinations of probabilities, and we are using only that
17 function in this review.
18 In some of the controlled human exposure studies, subjects were exposed to a given Os
19 concentration more than once - for example, using a constant (square-wave) exposure pattern in
20 one protocol and a variable (triangular) exposure pattern in another protocol. However, because
21 there were insufficient data to estimate subject-specific response probabilities, we assumed a
22 single response probability (for a given definition of response) for all individuals and treated the
23 repeated exposures for a single subject as independent exposures in the binomial distribution.
24 For each of the two functional forms (logistic and linear), we derived a Bayesian
25 posterior distribution using this binomial likelihood function in combination with prior
26 distributions for each of the unknown parameters (Box and Tiao, 1973). We assumed lognormal
27 priors with maximum likelihood estimates of the means and variances for the parameters of the
28 logistic function, and normal priors, similarly with maximum likelihood estimates for the means
29 and variances, for the parameters of the linear function. For each of the two functional forms
• The 3-parameter logistic function is a special case of the 4-parameter logistic, in which the function is forced to go
through the origin, so that the probability of response to 0.0 ppm is 0.
6-18
-------
1 considered, we used 1,000 iterations as the "burn-in" period3 followed by 9,000 iterations for the
2 estimation. Each iteration corresponds to a set of values for the parameters of the (logistic or
3 linear) exposure-response function. We combined the 9,000 sets of values from the logistic
4 model runs with the last 1,000 sets of values from the linear model runs to get a single combined
5 distribution of 10,000 sets of values reflecting the 90 percent/10 percent assumption stated
6 above. WinBUGS version 1.4.3 was used for these analyses (WinBUGS; Lunn et al., 2012).
7 For any Oj, concentration, x, we can derive the nth percentile response value, for any n, by
8 evaluating the exposure-response function at x using each of the 10,000 sets of parameter values
9 (9,000 of which were for a logistic model and 1,000 of which were for a linear model). The
10 resulting 2.5th percentile, median (50th percentile), and 97.5th percentile exposure-response
11 functions for changes in FEVi > 10% are shown in Figure 6-6, along with the response data to
12 which they were fit. The corresponding exposure-response functions for changes in FEVi >
13 15% and > 20% are shown in Appendix 6-A. The values of the functions are also provided in
14 Appendix 6-A.
15
3 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.
6-19
-------
1
2
90%
80%
70%
_^ 60%
TO 50%
cc
01
§ 40% -
Q.
01
02 30%
20%
10%
0%
C
S
/
/ ^""
/ 90 10%)
1 f^2/ Median (FEV1>10%)
// V
/ / •
/// • Data for
/ / *
1 1 j »'*331
31S i^t
/ // 90% logistic
/ // 10% linear
^SKso
— ^^^
) 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
Ozone Exposure (ppm)
(FEV1>10%)
FEV1>10%
orm mix:
&
Figure 6-6. Probabilistic Exposure-Response Relationships for FEVi Decrements >
3 10% for 8-Hour Exposures At Moderate Exertion, Ages 18-35. Values associated with data
4 points are the number of subject-exposures at each exposure concentration.
5
6
7 The population risk is estimated by multiplying the expected risk by the number of
8 people in the relevant population, as shown in Equation 6-6 below. The risk (i.e., expected
9 fractional response rate) for the k* fractile, Rk is estimated as:
N
10 Rk = £ P^RR, \ ej ) (Equation 6-6)
1 1 where:
12 6j = (the midpoint of) they'th category of personal exposure to O^;
13 PJ = the fraction of the population having personal exposures to 63 concentration
14 ofe/ppm;
15 RRk 6j = k-fractile response rate at O^ exposure concentration BJ;
16 N= number of intervals (categories) of Os personal exposure concentration.
6-20
-------
1 Exposure estimates used in this portion of the risk assessment were obtained from APEX
2 for each of the fifteen urban areas and the five air quality scenarios. Chapter 5 provides
3 additional details about the inputs and methodology used to estimate population exposure in the
4 urban areas. Exposure estimates for all and asthmatic school-aged children (ages 5 to 18) were
5 combined with probabilistic exposure-response relationships for lung function decrements
6 associated with 8-hour exposure while engaged in moderate exertion. Individuals engaged in
7 activities that resulted in an average equivalent (BSA-normalized) ventilation rate (EVR) for the
8 8-hour period at or above 13 L/min-m2 BSA were included in the exposure estimates for 8-hour
9 moderate or greater exertion. This range was selected based on the EVRs for the group of
10 subjects in the controlled human exposure studies that were the basis for the exposure-response
11 relationships used in this portion of the risk assessment.
12 6.3 O3 RISK ESTIMATES
13 This section provides lung function risk estimates associated with several air quality
14 scenarios: five recent years of air quality as represented by 2006 to 2010 monitoring data, and air
15 quality in those years after simulating just meeting the existing O?, standard and alternative
16 standard levels of 0.070, 0.065, and 0.060 ppm. The risk measures presented here are the
17 percents of the population estimated to experience lung function responses greater then 10, 15,
18 and 20%, one or more times or six or more times during an 63 season, for three age groups:
19 school-aged children (ages 5-18), young adults (ages 19-35) and adults ages 36-55. Results for
20 adults older than 55 are not presented since the responses for this age group are estimated to be
21 minimal. People with multiple events with large lung function decreases are more at risk than
22 those with only one such event during the Os season. Although six events is less than once per
23 month, we see dramatic decreases in population risk in going from one or more to six or more
24 events during a season, which is why we report on six or more events rather than a higher
25 number.
26 In the figures and tables that follow, "base" indicates the base case scenario of recent air
27 quality for the indicated year. "75," "70," "65," and "60" respectively indicate the existing O3
28 standard and alternative standard levels of 0.070, 0.065, and 0.060 ppm. "75 6-8" indicates the
29 0.075 ppm existing 8-hour standard based on rollback for the 2006-2008 period, while "75 8-10"
30 indicates the existing standard scenario based on rollback for the 2008-2010 period. There are
31 two estimates of results for the 2008 existing and alternative standard scenarios (because 2008
32 overlaps the two rollback periods) and one for each of the other four years. These two estimates
33 for 2008 can be quite different because of the relationship between the design value over the
34 three-year period and the amount of adjustment to the air quality distribution in 2008 that can
35 result.
6-21
-------
1 6.3.1 Lung Function Risk Estimates Based on the McDonnell-Stewart-Smith Model
2 Results based on the McDonnell-Stewart-Smith (2012) threshold model are summarized
3 in this section; detailed results can be found in Appendix 6-B.
4 Figure 6-7 shows the results for school-aged children in the same format used in exposure
5 results, explained in Section 5.3.1. Figure 6-7 depicts results for all cities, year, and scenarios for
6 ages 5 to 18 with > 1 occurrences of FEVi decrements > 10, 15, 20% and illustrates the variation
7 of results across cities, year, and scenarios.
8 Figure 6-8 shows the variation across cities (horizontally) and years (vertically) for the
9 percent of school-aged children with > 1 occurrences of FEVi decrements > 10% with air quality
10 just meeting the potential alternative standard of 0.07 ppm. The points above each study area on
11 this graph represent the risk for the six years for the study area (2008 has two points,
12 corresponding to the different 2006-2008 and 2008-2010 design values used to adjust the air
13 quality to meet 0.07 ppm). There is substantial variability both across years and across cities.
14 Denver has the highest overall risks, while Cleveland and New York have the lowest. Los
15 Angeles has the smallest variation across years, with a range of 2.3% (from 14.3% to 16.6%).
16 The other cities have a range of around 4% to 7.5% across years.
17 Table 6-4 and Table 6-5 present summary results (ranges over cities and years) of FEVi
18 decrements > 10 and 15% estimated by the MSS model for the different age groups. The results
19 for asthmatic school-aged children are very similar to the results for all school-aged children and
20 are not presented here.
21 Figure 6-9 illustrates the distribution of responses (FEVi decrements > 10%) across
22 ranges of ambient concentrations of Os for school-aged children for one city and scenario (Los
23 Angeles, 2006 recent air quality). The concentrations are daily 8-hour average ambient
24 concentrations during the 8-hour period with maximum 8-hour average exposure for that day.
25
6-22
-------
xp
(^
O
Al
"S
e
O
(L)
PH
<+H
O
(L)
O
(L)
I
O
O
Al
+-»
'%
c
.£5
O
"o
O
O
GO
18 8 8 8 ~8 89888 8188-6
Fev
- FEV10
-e- FEV15
FEV20
Figure 6-7. Risk results for all school-aged children with > 1 occurrences of FEVi
decrements > 10,15, 20% for all cities, year, and scenarios (y-axis is percent of children affected).
6-23
-------
1
2
3
4
5
21%:
20%:
19%:
18%-
a 17%"
S 16%:
a
a 15%:
"+H !
° 1 4°/ •
1
a 13%:
&
12%:
10%:
9%:
J : *
*
h * *
* * *
ak
ABB
t a o
1 1 s
a t t
n i o
t m n
a o
r
e
9k
9k
9k
*
* $
9k
9k
9k
C C
h 1
i e
C V
a e
g 1
o a
n
d
9k
9k
9k
*
>k
9k
9k
*
ODD
a e e
1 n t
1 v r
a e o
s r i
t
*
H
0
u
s
t
0
n
i
9k
9k
9k
L
0
s
A
n
g
e
1
e
s
9k
9k
9k
$
9k
N
e
w
Y
0
r
k
9k
*
9k
9k
9k
P
h
i
1
a
d
e
1
P
h
i
a
it
9k
9k
9k
S
a
c
r
a
m
e
n
t
0
9k
9k
9k
9k
S
t
L
0
u
i
s
9k
9»
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
cities (horizontally) and years (vertically).
6-24
-------
1
2
3
4
5
Table 6-4. Ranges of percents of population experiencing one 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 cities and years.
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 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 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
7
6-25
-------
1
2
3
4
Table 6-5. Ranges of percents of population experiencing one 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 cities and years.
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 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 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%
5
6
9
10
11
12
13
These concentrations are less than but close to daily maximum 8-hour average ambient
concentrations and are greater than daily maximum 8-hour 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-hour
average ambient concentrations are above 40 ppb for this modeled scenario. This distribution
will be different for different cities, years, and air quality scenarios.
6-26
-------
20;
1 8-
16:
8 14-
a
responses
O K>
«4H
O
1" 8"
! 6-
PH
4-
2;
A-
El
n
i 8-hour average concentration (ppb)
2 Figure 6-9. Distribution of Daily FEVi Decrements > 10% Across Ranges of 8-hour
3 Average Ambient Os Concentrations (Los Angeles, 2006 recent air quality).
4
5 Outdoor workers spend more time outdoors than the general population and therefore are
6 at higher risk for health effects due to 03. We conducted simulations of outdoor workers ages 19-
7 35 for Atlanta (2006) for the current and alternative standards to estimate the risk of this group
8 for experiencing FEVi decrements > 15%. The methodology for simulating outdoor workers
9 involves modifying activity diaries to represent outdoor workers and is described in Section
10 5.3.3 in Chapter 5. Table 6-6 shows the results of these simulations and compares them with the
11 results for the general population, ages 19-35. The percents of people experiencing one or more
12 FEVi decrements > 15% during the 2006 Os season in Atlanta are 3.6 times higher for outdoor
13 workers than for the general population (ages 19-35) under the current standard, and range up to
14 5.3 times higher for the alternative standards. The percents of people experiencing six or more
15 FEVi decrements > 15% during the 2006 Oi season in Atlanta are 20 times higher for outdoor
16 workers than for the general population under the current standard, and range up to 150 times
17 higher for the alternative standards. As expected, we see that the risk of repeated occurrences of
18 FEVi decrements > 15% is much greater for outdoor workers than for the general population.
6-27
-------
1 Part of the reason for this is that APEX tends to underestimate the number of individuals who
2 have very repetitive activity patterns (e.g., 9 to 5 weekdays office workers) when using the
3 CHAD activity database and the method selected for generating longitudinal diary profiles (see
4 Section 5.3.1).
5
6 Table 6-6. Percents of the General Population and Outdoor
7 Workers (ages 19-35) Experiencing 1 or More and 6 or More FEVi
8 Decrements > 15% (based on Atlanta 2006 APEX simulations)
1 or more
Current standard
70 ppb alt. std.
65 ppb alt. std.
60 ppb alt. std.
6 or more
Current standard
70 ppb alt. std.
65 ppb alt. std.
60 ppb alt. std.
General
population
ages 19-35
occurrences
1.2%
0.84%
0.55%
0.32%
occurrences
0.06%
0.018%
0.005%
0.005%
Outdoor workers
ages 19-35
4.3%
3.2%
2.5%
1.7%
1.2%
0.93%
0.74%
0.55%
10 6.3.2 Lung Function Risk Estimates Based on the Exposure-Response Functions
11 Approach Used in Prior Reviews
12 In this section we present lung function risk estimates for all school-aged children
13 following the methodology used in previous reviews, based on updated exposure-response (E-R)
14 functions. In Appendix 6-C we compare these estimates with those from the previous review.
15
16 Table 6-7 provides an overall summary of results for each air quality scenario by
17 tabulating the minimum and maximum estimates over all cities and years of percents of all
18 school-aged children (ages 5 to 18) experiencing one or more days (during the 63 season) with
19 FEVi decrement more than 10 and 15%. This table can be compared with Table 6-4 and Table 6-
20 5, which have analogous results for the MSS model. These results are much lower than the MSS
21 model results. The reasons for this are described in Section 6.3.3 below.
6-28
-------
1
2
3
4
5
6
7
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
Table 6-7. Ranges of percents of school-aged children experiencing one or more
days during the Os season with lung function decrement (AFEVi) more than 10 and 15%.
The numbers in this table are the minimum and maximum percents estimated over all
cities and years.
Scenario
base
75
70
65
60
minimum
percent
experiencing
> 1 day with
A FEVi > 10%
2%
2%
2%
1%
2%
maximum
percent
experiencing
> 1 day with
A FEVi > 10%
11%
6%
6%
5%
3%
minimum
percent
experiencing
> 1 day with
A FEVi > 15%
0%
1%
0%
0%
0%
maximum
percent
experiencing
> 1 day with
A FEVi > 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-
hour average exposures when the 8-hour 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 exposure-response 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 then 10,
15, and 20%. The MSS model estimates are significantly higher than the exposure-response
function approach estimates. In most cases, the MSS model gives results about a factor of three
higher than the exposure-response 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 exposure-response model of previous reviews.
6-29
-------
1
2
Table 6-8. Comparison of responses from the MSS model with responses from the
population exposure-response (E-R) method. 2006 existing standard, ages 5 to 18
Urban 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%
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Since the E-R method of the previous reviews only looks at 8-hour 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-hour average exposure period where the concomitant 8-hour 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-hour period. 4.46% (not among the 15.17%) have instances of AFEVi > 10% but
6-30
-------
1 none of those instances with concomitant EVR > 13. The 11.54% responding is made up of
2 6.67% of the population with instances of AFEVi > 10% concomitant with EVR > 13 and 4.87%
3 with instances of AFEVi > 10% not concomitant with EVR > 13.
4 Table 6-9 and Table 6-10 summarize the pertinent results for the Atlanta and three Los
5 Angeles simulations. Looking at the first rows of these tables shows that these models have
6 similar corresponding results. The broader scope of activity/exposure patterns encompassed by
7 the MSS model, beyond the 8-hour average EVR > 13 restriction of the E-R model, contributes
8 from a third to a half to the total MSS model risk and to a large part explains the differences
9 between the models for ages 18-35. The difference between the MSS and E-R models is larger
10 for school-aged children than for adults ages 18-35 due to the increased EVR and time spent
11 outdoors in children compared to adults.
12
6-31
-------
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 AFEV1 > cutoff
concomitant with 8 -hour EVR > 13
profiles with instances of AFEVi > cutoff never
concomitant with 8 -hour EVR > 13
Final result of each model
MSS model
AFEVj > 10%
6.7%
4.8%
11.5%
E-R model
AFEVj > 10%
5.0%
5.0%
MSS model
AFEVj > 15%
2.1%
1.2%
3.3%
E-R model
AFEVj > 15%
1.8%
1.8%
MSS model
AFEVj > 20%
0.8%
0.5%
1.3%
E-R model
AFEVj > 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 -hour EVR > 13
profiles with instances of AFEVi > cutoff never
concomitant with 8 -hour EVR > 13
Final result of each model
MSS model
AFEVj > 10%
7.9%
6.5%
14.4%
E-R model
AFEVj > 10%
6.2%
6.2%
MSS model
AFEVj > 15%
2.6%
1.8%
4.4%
E-R model
AFEVj > 15%
2.6%
2.6%
MSS model
AFEVj > 20%
1.2%
0.8%
2.0%
E-R model
AFEVj > 20%
1.4%
1.4%
6-32
-------
1
2
3
4
Figure 6-10 compares the E-R function to the response curve of the MSS model restricted
to 8-hour average EVR > 13 and shows that these curves are very close. The MSS model has a
higher response for the low and high ranges of exposure concentrations, while the E-R model is
higher in the mid-range of exposures.
M
I
w
=10% (MSS)
AFEV1>=10% (E-R)
r
30 40 50
Ozone exposure (ppb)
AFEV1>=15% (MSS)
AFEV1>=15% (E-R)
\
60
70
\
80
AFEV1>=20% (MSS)
AFEV1>=20% (E-R)
5 Figure 6-10. Comparison of E-R and MSS Model (restricted to 8-hour average EVR
6 > 13) Response Functions (Atlanta 2006 base case, ages 18-35).
7 Another element of the difference between the models derives from the distribution of
8 EVR in the clinical studies the E-R approach is based on and how this compares to the
9 distribution of EVR in the APEX simulations. Most of the clinical studies are conducted with a
10 target EVR of 20 L/min-m2 BSA and the actual EVRs vary somewhat around this value. The
9 -^^
11 rationale for the cutpoint of 13 L/min-m BSA is described in EPA's responses to comments on
12 the 1996 proposed rule on the NAAQS for O3 (Federal Register, 1996) as "for the 8-hr health
13 risk assessment the range (based on being within 2 standard deviations of the mean) of EVRs
14 observed in the subjects who participated in the study [McDonnell et al., 1991] was 13-27 liters
15 per minute per meter squared [BSA] (L/min-m2)." Figure 6-11 shows the distribution of EVRs >
6-33
-------
1
2
3
4
5
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.
6
7
8
9
EVR
13.5
14.5
15.5
16.5
17.5
18.5
19.5
20.5
21.5
22.5
23.5
24.5
25.5
26.5
27.5
28.5
29.5
30.5
31.5
32.5
33.5
34.5
35.5
1
1
1
1
1
1
zz
HI
H
H
]
1
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
0
200000
400000
600000
800000
1000000
1200000
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
FREQUENCY
Figure 6-11. Distribution of Daily Maximum 8-hour Average EVR For Values of
EVR> 13 (L/min-m2) (midpoints on vertical axis) (Atlanta 2006 base case, ages 18-35).
6-34
-------
1 6.4 EVALUATION OF THE MSS MODEL
2 6.4.1 Summary of Published Evaluations
3 McDonnell et al. (2010) performed a detailed evaluation of their model using two
4 methods: (1) cross-validation and (2) comparison of an independent data set against the
5 predictions of the model.
6 The cross-validation was based on the data set of 15 EPA studies from which their
7 original model was developed (McDonnell et al., 2007). This data set has 541 subjects, each with
8 multiple measurements during single experiments. Subjects were omitted from the data set, one
9 at a time, the model refit to the reduced data set, and the resulting parameters used to predict the
10 FEVi decrements for the omitted subject. The authors then compare the mean predictions and
11 mean observed values for each subject and presented these results in a scatter plot (Figure Ib,
12 McDonnell et al., 2010). The observations exhibit much more variability than the predictions; for
13 observed values of 20%, predicted values range from around 2 to 19%; and all observed values
14 above 20% are underpredicted (the observed values range from -20 to 60%, while the predicted
15 values range from 0 to 20%). These features result from the omission of the inter- and intra-
16 individual variability terms (U; and sp) in the MSS model (equation 6-3), which are accounted
17 for in the risk estimates in this chapter.
18 Model predictions were compared against an independent data set of seven clinical
19 studies with a total of 204 subjects (McDonnell et al., 2010). Graphs of predicted and observed
20 study means vs. time show fair to good model fit. The authors do not present overall fit statistics
21 that are directly commensurate with the statistics of interest in this risk assessment: the
22 proportions of people with FEVi decrements greater than 10, 15, and 20%.
23 McDonnell et al. (2012) do compare observed and predicted proportions of people with
24 FEVi decrements greater than 10, 15, and 20% and provide the corresponding scatter plots
25 (Figure 4). They find the model to be unbiased, with the slopes of the observed vs. predicted
26 lines for 10, 15, and 20% to be around 1.0 and the R2 respectively 0.78, 0.73, and 0.67. The
27 higher observed proportions of people with FEVi decrements greater than 10, 15, and 20%
28 tended to be substantially underpredicted.
29 6.4.2 Children
30 A clinical study with children (ages 8-11; mean, 10 years; n=22), exposed to 120 ppb Os
31 over 2.5 hours at heavy exertion levels was done by McDonnell et al. (1985). This study could be
32 used to fit the model for children if all of the measurements of FEVi and ventilation rates were
33 available. The paper lists the end-of exposure FEVi responses for each individual (but not
34 ventilation rates), which we use to compare with the MSS model with the age term extension
35 described in Section 6.2.4. The numbers of subjects with clean-air adjusted responses greater
6-35
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
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 (EPA, 2011). Details of this comparison can be
found in Appendix 6-D.
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% FEV1 decrement
MSS model
18.4%
McDonnell
et al. (1985)
18.2%
(4 subjects)
> 15% FEV1 decrement
MSS model
6.8%
McDonnell
et al. (1985)
9.1%
(2 subjects)
> 20% FEV1 decrement
MSS model
2.3%
McDonnell
et al. (1985)
4.5%
(1 subject)
15
16
17
18
19
20
21
22
23
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. This is consistent with the logistic form of the model,
where the impact of exposures to low concentrations on risk is small.
6-36
-------
1 Table 6-12. Percents of the population by age group with one or more days during
2 the Os season with lung function (FEVi) decrements more than 10,15, and 20% (Atlanta
3 2006 base case). MSS Threshold model, monitors air quality.4
4
5
6
7
8
9
Age AFEVi > AFEVi > AFEVi >
Group 10% 15% 20%
5 to 18 31% 13% 6.4%
19 to 35 11% 3.1% 1.3%
36 to 55 3.7% 0.60% 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, monitors air quality.
Age AFEVi > AFEVi > AFEVi >
Group 10% 15% 20%
5 to 18 31% 13% 6.6%
19 to 35 11% 3.1% 1.3%
36 to 55 3.8% 0.60% 0.15%
10
11
12 6.5 CHARACTERIZATION OF UNCERTAINTY
13 In the controlled human exposure study based risk assessment, there are two broad
14 sources of uncertainty to the risk estimates. One of the most important sources of uncertainty is
15 the estimation by APEX of the population distribution of individual time series of Oj exposures
16 and ventilation rates. The uncertainty regarding these estimated exposures is discussed in
17 Chapter 5; they are not discussed further here.
18
4 In the first draft REA, 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 second draft REA. 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 6-F).
6-37
-------
1 In this section, uncertainties associated with the second broad source of uncertainty in the
2 risk calculation are discussed, namely, uncertainties in the lung function risk model. The specific
3 sources of uncertainty covered are:
4 • Statistical model form
5 • Convergence of APEX results
6 • Application of model for all lifestages
7 • Application of model for asthmatic children
8 • Interaction between 63 and other pollutants
9 6.5.1 Statistical Model Form
10 The MSS model is a 2-compartment model, the form of which is based on physical
11 considerations. It accomodates these key features of human exposure studies: (1) FEVi
12 responses increase with increasing O^ concentration, ventilation rate, and duration of exposure,
13 (2) the effect of each of these three variables depends on the levels of the other two variables, (3)
14 FEVi responses depend on age, (4) certain individuals are consistently more responsive to Os
15 exposure, and (5) Os -induced FEVi decrements improve within a few hours of cessation of
16 exposure (McDonnell et al, 2007). These considerations support the form of the model, as do
17 model evaluation that have been performed (Section 6.4.1). Although the model does not have
18 good predictive ability for individuals (psuedo-R2 0.28), it does better at predicting the
19 proportion of individuals with FEVi decrements > 10, 15, and 20% (psuedo-R2s of 0.78, 0.74,
20 0.68) (McDonnell et al, 2012).
21 The clinical studies that these models' estimates are based on were conducted with young
22 adult volunteers rather than randomly selected individuals, so it may be that selection bias has
23 influenced the model parameter estimates.
24 The parameter estimates are not very precise, as the result of the likelihood surface being
25 somewhat flat in the neighborhood of the maximum likelihood estimates. Table 6-14 gives 95
26 percent confidence intervals for each of the parameter estimates as percents of the estimates,
27 based on the standard errors reported by McDonnell et al. (2012). Figure 6-10 shows how much
28 the modeled number of children with one or more FEVi decrements > 10% changes when each
29 parameter is increased by five percent (keeping the other parameters fixed at their estimates).
30 The scenario modeled is the Los Angeles 2006 recent conditions base case. The physiological
31 parameter MET, a measure of the level of exertion for a given activity (see Appendix 6-E), is
32 also included here for comparison. MET is a key variable in calculating ventilation rates and is
33 specified by a distribution for each activity. Here we have shifted all MET distributions by +5%
34 of their means.
6-38
-------
1 Table 6-14. MSS threshold model estimated parameters with confidence intervals
parameter
estimate
standard
error
95% conf.
interval
PI
10.916
0.8446
±15%
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%
2 from McDonnell et al. (2012).
3 The most influential parameter in Figure 6-12 is Pe, the power to which ventilation rate is
4 raised in the MSS model. An increase of five percent in p6 leads to 27, 40, and 47 percent
5 increases respectively in the modeled number of children with FEVi decrements > 10, 15 and
6 20%. The next most influential parameter is the variance of E, the intra-individual variability
7 term. The least influential parameter is P2, the slope of the age term. These changes of five
8 percent are much less than the 95 percent confidence intervals of the parameter estimates, so the
9 uncertainty in the risk estimates resulting from parameter uncertainty is likely to be more than is
10 indicated in Figure 6-12.
11 Age Term Significance
12 As discussed in Section 6.5.3 below, there are uncertainties in extrapolating the MSS
13 model down to age 5 from the age range of 18 to 35 to which the model was fit. Further
14 considerations indicating that the uncertainty of the extension to children of the MSS model
15 could be substantial are that the age coefficient P2 = -0.21 (s.e. 0.31) in the MSS model is not
16 statistically significantly different from zero; and when the MSS model is fit to the U.C. Davis
17 clinical data the age term is positive, P2 = +0.19 (0.60), although also not statistically
18 significantly different from zero (McDonnell et al., 2012). Note that, in the previous section, P2
19 was found to be the least influential model parameter.
6-39
-------
1-
2-
3-
4-
5-
7-
8-
9-
10-
beta2
11
( 5%) 0%
5% 10% 15% 20% 25% 30%
Elasticity
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.
6-40
-------
1 The Variability Term s
2 The variability term s in equation 6-3 is assumed by the MSS model to have a Gaussian
3 distribution with mean zero and estimated standard deviation 4.135 (in the threshold model).
4 Since the actual values are bounded, we truncate the variability term distribution at ±2 standard
5 deviations (±8.27), a convention we use for the distributions of several physiological variables
6 input to APEX in the physiology input file. To look at the effect of truncating the variability
7 distribution, we conducted simulations with the variability term truncated at ±20, the range of the
8 actual values of the variability term. We find that this constraint has a very large effect on
9 estimates of percents of the population with FEVi decrements > 10 and 15% and less of an effect
10 for 20%. The percent of children with FEVi decrements > 10% increases from 31% to 92%
11 when increasing the truncation point from 8.27 to 20. Details of this comparison and additional
12 results are presented in Appendix 6-F. The assumption that the distribution of the variability term
13 s is Gaussian is convenient for fitting the model, but is not accurate. The extent to which this
14 mis-specification affects the estimates of the parameters of the MSS model is not clear.
15 6.5.2 Convergence of APEX Results
16 APEX accounts for several sources of variability by drawing random variables from
17 specified distributions. Some variables are drawn once for each simulated individual (e.g., age,
18 location of residence), some are drawn every day or every hour for each simulated individual,
19 and others are drawn more frequently, at the event level (e.g., activity). Increasing the number of
20 individuals simulated in an APEX run increases the accuracy of the modeled variability and the
21 results of the APEX runs are more reproducible. In order to assess the number of individuals to
22 simulate to achieve convergence of APEX results, we perform multiple APEX runs with
23 identical inputs except for the random number seed, and look at the variability of the results of
24 these model runs. Table 6-15 summarizes the results of 40 APEX simulations of the Atlanta
25 2006 base case with 200,000 simulated individuals. For each of these measures, the range of
26 results over the 40 APEX runs is less than one percent. This analysis of the convergence of
27 APEX results shows that modeling 200,000 simulated individuals is adequate for reasonable
28 convergence of the FEVi risk measures.
29
6-41
-------
1 Table 6-15. Convergence results for the Atlanta 2006 base case with 200,000
2 simulated individuals. Percents of the population by age group with one or more days (and
3 six or more days) during the Os season with lung function (FEVi) decrements more than
4 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%
11.5%
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 6.5.3 Application of Model for All Lifestages
7 The exposure-response functions derived from controlled human exposure studies
8 involving 18-35 year old subjects were used to estimate responses for school-aged children (ages
9 5-18). This was in part justified by the findings of McDonnell et al. (1985) who reported that
10 children 8-11 years old experienced FEVi responses similar to those observed in adults 18-35
11 years old when both groups were exposed to 120 ppb 63 at an EVR of 32-35 L/min/m2. In
12 addition, a number of summer camp studies of school-aged children exposed in outdoor
13 environments in the Northeast also showed Cb-induced lung function changes similar in
14 magnitude to those observed in controlled human exposure studies using adults, although the
15 studies may not directly comparable. The MSS model predicts increasing responsiveness with
16 younger participants in the age range of 18-35 years, as shown in Figure 6E-4 (Appendix 6-E),
17 which might indicate that responsiveness would continue to increase as age decreases from 18.
18 In extending the MSS model to children, we fixed the age term in the model at its highest value,
19 the value for age 18. If continuing the MSS model trend were to accurately describe continued
20 increased response in children, then the fixed age term for children may have underestimated the
21 effects on children, and particularly younger children. On the other hand, if FEVi responses for
22 children are similar to those observed in adults 18-35 years old, as the evidence suggests, then
23 our approach to extending the age term would overestimate the response to children (see Table
24 6E-3 in Appendix 6-E).
6-42
-------
1 In considering extending the MSS model to ages older than 36, we note that, in general,
2 Os responsiveness steadily declines for persons aged 35-55, with persons >55 eliciting minimal
3 responsiveness (ISA, section 6.2.1.1). As described in Section 6.2.4, we extended the age term
4 from the value at 36 linearly to zero at age 55, and set it to zero for ages above 55 (see Error!
5 Reference source not found.). The uncertainty of this extrapolation may be substantial, but
6 these age groups are not the primary focus in the clinical risk assessment.
7 6.5.4 Application of Model for Asthmatic Children
8 The risk assessment used the same exposure-response relationship, developed from data
9 collected from healthy study subjects, and applied it to all persons, children, and asthmatic
10 children. Based on limited evidence from a few human exposure studies, it is likely that subjects
11 having asthma are at least as sensitive to acute effects of O^ as other subjects not having this
12 health condition (ISA, page 6-20 to 6-21). An analysis by Romieu et al. (2002) indicated a larger
13 Os-associated decrement in FEVi among children with moderate to severe asthma than
14 among all children with asthma (ISA, page 6-54). This suggests that the lung function
15 decrements presented in this assessment for asthmatic children may be underestimated. The
16 magnitude of influence this element might have on our risk estimates remains unknown at this
17 time. In addition, asthmatic children may have less reserve lung capacity to draw upon when
18 faced with decrements, and therefore a >10% decrement in lung function may be a more adverse
19 event in an asthmatic child than a healthy child.
20 6.5.5 Interaction Between Os and Other Pollutants
21 Because the controlled human exposure studies used in the risk assessment involved only
22 63 exposures, it was assumed that estimates of (Vinduced health responses would not be
23 affected by the presence of other pollutants (e.g., SC^PM^.s, etc). The magnitude of influence
24 that potential interactions might have on our risk estimates remains unknown at this time.
25 6.5.6 Qualitative Assessment of Uncertainty
26 EPA staff have identified key sources of uncertainty with respect to the lung function risk
27 estimates. These are: the physiological model in APEX for ventilation rates, the Os exposures
28 estimated by APEX, the MSS model applied to ages 18 to 35, and extrapolation of the MSS
29 model to children ages 5 to 18. The first two of these are discussed in Chapter 5. At this time we
30 do not have quantitative estimates of uncertainty for any of these. Table 6-16 provides a
31 qualitative assessment of the uncertainty resulting from each of these key sources. The primary
32 source of uncertainty is the MSS model, applied to ages 18 to 35.
33
6-43
-------
Table 6-16. 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 6-E 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).
C>3 exposures
The Os exposures estimated by APEX
and their uncertainties are discussed in
Chapters.
Both
Low-
Medium
Low
T, exposures are a key input to the MSS model.
The McDonnell-
Stewart-Smith (MSS)
FEVi model for ages
18 to 35
The MSS model is integrated into
APEX and predicts FEVi 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 FEVi and
estimated parameters of the model introduce uncertainty
into the model predictions of large FEVi decrements. The
estimated parameters have fairly wide confidence intervals
(Table 6-1) 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 FEVi decrements > 10%.
(The 95 percent confidence interval of this parameter
estimate is ±14%.)
The variability term e [in equation 6-3] 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), a convention
we use for the distributions of several physiological
6-44
-------
Source
Description
Potential influence of
uncertainty on risk
estimates
Direction Magnitude
Knowledge-
Base
uncertainty*
Comments
variables input to APEX in the physiology input file. To
look at the effect of truncating the variability distribution,
we conducted simulations with the variability term
truncated at ±20, the range of the actual values of the
variability term. 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.
Extrapolation of the
MSS model to
children
The MSS model is based on studies
with subjects ranging in age from 18 to
35 years; therefore prediction for
individuals outside this age range
involves assumptions for extrapolation
of the MSS model for individuals <18
and >35 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
6-45
-------
1 6.6 DISCUSSION
2 The second draft lung function risk assessment evaluated risks of lung function
3 decrements due to O?, exposure for all three groups: school-age children ages 5 to 18, young
4 adults ages 19 to 35, and adults ages 36 to 55. Adults older than 55 have minimal (Vinduced
5 lung function risk. Two models were used, one based on application of an individual level
6 exposure-response function, the MSS model introduced in this review, and one based on
7 application of a population level E-R function consistent with the model used in the previous 63
8 review which applies probabilistic population-level exposure-response relationships for lung
9 function decrements (measured as percent reductions in FEVi) associated with 8-hour moderate
10 exertion exposures. The MSS model is preferred, due to its ability to model individual exposures
11 for a wide range of exposure times and levels of exercise (Section 6.2.4; ISA pages 6-15 to 6-
12 16). Both models provide estimates of the percent of the groups experiencing a reduction in lung
13 function for three different levels of impact, 10, 15, and 20% decrements in FEVi. These levels
14 of impact were selected based on the literature discussing the adversity associated with these
15 types of lung function decrements (US EPA, 2012, Section 6.2.1.1; Henderson, 2006). For the
16 second draft assessment, lung function risks were estimated for 15 cities: Atlanta, Baltimore,
17 Boston, Chicago, Cleveland, Dallas, Denver, Detroit, Houston, Los Angeles, New York,
18 Philadelphia, Sacramento, St. Louis, and Washington, DC.
19 Based on the MSS model, the percents of population estimated to experience lung
20 function responses greater then 10, 15, and 20%, associated with 63 exposure while engaged in
21 various levels of exertion, vary considerably for different years and cities under the recent air
22 quality scenarios and also for the existing and alternative standard scenarios (Figure 6-7 and
23 Figure 6-8, Table 6-4 and Table 6-5). The estimates for > 10% FEVi decrement for school-age
24 children for recent O?, concentrations range across cities and years from 11 to 31 percent, and
25 range from 11 to 22 percent after simulating just meeting the existing standard. The estimates for
26 > 15% FEVi decrement for school-age children for recent Os concentrations range across cities
27 and years from 2 to 12 percent, and range from 2 to 6 percent after simulating just meeting the
28 existing standards. The estimates for > 20% FEVi decrement for recent 63 concentrations range
29 across cities and years from 1 to 6 percent, and range from 1 to 3 percent after simulating just
30 meeting the existing standards.
31 Figure 6-13 displays the risks and the incremental increases in risk for increasing
32 standard levels, where risk is taken to be the highest value for each study area (over years) of the
33 percent of school-aged children with FEVi decrement > 10%. The risks in this figure for
34 Washington, DC, for example, are about 9.6% for the alternative standard level of 60 ppb and
35 13.4% for the alternative standard level of 65 ppb. The length of the orange bar is the
6-46
-------
1 incremental risk (3.8%) in going from the 60 ppb to the 65 ppb alternative standards. This figure
2 shows that there are significant increases in incremental risk for all 15 cities in the progression of
3 alternative standard levels from 60 ppb to the level of the existing standard, 75 ppb. The pattern
4 of reductions for lung function decrements larger than 15 and 20% are similar. As discussed in
5 Section 4.3.1, the New York 60 ppb alternative standard was not modeled and the risk for NY for
6 that scenario would not necessarily be zero. Figure 6-14 displays the risks and the incremental
7 increases in risk for increasing standard levels, where risk is taken to be the mean value for each
8 study area (over years) of the percent of school-aged children with FEVi decrement > 10%.
9 Similar to the MSS model results, the percents of school-age children estimated to
10 experience lung function responses greater then 10, 15, and 20% based on the population level
11 E-R function exhibit variation across years and cities. However, the MSS model estimates are
12 significantly higher than the E-R approach estimates. For lung function responses greater than
13 10, 15, and 20% the MSS model gives results typically a factor of three higher than the E-R
14 model for school-aged children. Both models give higher responses for higher concentrations,
15 compared to lower concentrations, as can be seen in Figures 6-6, 6-9, and 6-10.
16 The MSS model was applied to estimate lung function risk for outdoor workers (ages 19-
17 35) in Atlanta for one year (2006). The proportion of outdoor workers with FEVi decrements >
18 15% ranges from 3.6 to 5.3 times the proportion of the general population (ages 19-35) with
19 FEVi decrements > 15% across the different standards simulated. The proportion of outdoor
20 workers with multiple occurrences of FEVi decrements > 15% is much greater than for the
21 general population.
22
6-47
-------
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 Years5
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St Louis
Washington
0% 2%
12% 14% 16% 18% 20% 22% 24%
4% 6% 8% 10%
percent of school-aged children with FEVI decrement > 10%
standard level (ppb) i i 60 i i 65 i i 70 i i 75
5 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.
6-48
-------
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 Years6
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%
percent of school-aged children with FEVI decrement > 10%
18% 20%
' 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.
6-49
-------
1 6.7 REFERENCES
2
3 Adams, W.C. 2002. "Comparison of Chamber and Face-mask 6.6-hour Exposures to Ozone on
4 Pulmonary Function and Symptoms Responses." Inhalation Toxicology, 14:745-764.
5 Adams, W.C. 2003. "Comparison of Chamber and Face Mask 6.6-hour Exposure to 0.08 ppm
6 Ozone via Square-wave and Triangular Profiles on Pulmonary Responses." Inhalation
1 Toxicology, 15: 265-281.
8 Adams, W.C. 2006. "Comparison of Chamber 6.6-h Exposures to 0.04-0.08 ppm Ozone via
9 Square-wave and Triangular Profiles on Pulmonary Responses." Inhalation Toxicology, 18:
10 127-136.
11 American Thoracic Society. 2000. "What Constitutes an Adverse Effect of Air Pollution?"
12 American Journal of Respiratory and Critical Care Medicine, 161: pp. 665-673.
13 Box, G. E. P. and G. C. Tiao. 1973. Bayesian Inference in Statistical Analysis. Wesley
14 Publishing Co.: Wiley Classic Library, ISBN 0-471-57428-7.
15 Burmaster, D.E. 1998. "Lognormal Distributions for Skin Area as a Function of Body Weight."
16 Risk Analysis, 18(l):27-32.
17 Dietert, R.R.; R. A. Etzel; D. Chen, et al. 2000. "Workshop to Identify Critical Windows of
18 Exposure for Children's Health: Immune and Respiratory Systems Work Group Summary."
19 Environ Health Perspect, 108(suppl 3): 483-490.
20 Esmail, S.; Y. Bhambhani and S. Brintnell. 1995. "Gender Differences in Work Performance on
21 the Baltimore Therapeutic Equipment Work Simulator." American Journal of
22 Occupational Therapy, 49: 405 - 411.
23 Federal Register. 1996. Responses to Significant Comments on the 1996 Proposed Rule on the
24 National Ambient Air Quality Standards for Ozone (December 13, 1996; Fed. Reg. 61 FR
25 65716).
26 Folinsbee, L.J., et al. 1988. "Pulmonary Function and Symptom Responses after 6.6-hour
27 Exposure to 0.12 ppm Ozone with Moderate Exercise." Journal of the Air Pollution Control
28 Association, 38: 28-35.
29 Graham, S. and T. McCurdy. 2005. Revised Ventilation Rate (Ve) Equations for Use in
30 Inhalation-Oriented Exposure Models: A NERL Internal Research Report. Research
31 Triangle Park, NC: EPA National Exposure Research Laboratory.
32 Hazucha, M.J.; L. J. Folinsbee and P. A. Bromberg. 2003. "Distribution and Reproducibility of
33 Spirometric Response to Ozone by Gender and Age." Journal of Applied Physiology, vol.
34 95, no. 5, pp. 1917-1925.
35 Hislop, A.A. 2002. "Airway and Blood Vessel Interaction During Lung Development." Journal
36 of Anatomy, 201:325-334. Available at:
37 .
6-50
-------
1 Horstman, D.H. et al. 1990. "Ozone Concentration and Pulmonary Response Relationships for
2 6.6-hour Exposures with Five Hours of Moderate Exercise to 0.08, 0.10, and 0.12 ppm."
3 American Review of Respiratory Disease, 142:1158-1163.
4 Isaacs, K.; G. Glen; T. Mccurdy and L. Smith. 2008. "Modeling Energy Expenditure and Oxygen
5 Consumption in Human Exposure Models: Accounting for Fatigue and EPOC." Journal of
6 Exposure Science and Environmental Epidemiology, 18,289-298.
7 Isaacs, K. and L. Smith. 2005. New Values for Physiological Parameters for the Exposure Model
8 Input File 'Physiology.txt.' Technical memorandum to Tom McCurdy, EPA, December 20.
9 Kim, C.S.; N. E. Alexis; A. G. Rappold; H. Kehrl; M. J. Hazucha; J. C. Lay; M. T. Schmitt; M.
10 Case; R. B. Devlin; D. B. Peden; D. Diaz-Sanchez. 2011. "Lung Function and Inflammatory
11 Responses in Healthy Young Adults Exposed to 0.06 ppm Ozone for 6.6 hours." American
12 Journal of Respiratory and Critical Care Medicine, 183: 1215-1221.
13 Linn, W.S.; R. D. Buckley; C. E. Spier; R. L. Blessey; M. P. Jones; D. A. Fischer and J. D.
14 Hackney. 1978. "Health Effects of Ozone Exposure in Asthmatics." The American Review of
15 Respiratory Disease, vol. 117, no. 5, pp. 835-843.
16 Lunn, D.; C. Jackson; N. Best; A. Thomas and D. Spiegelhalter. 2012. The BUGS Book-A
17 Practical Introduction to Bayesian Analysis. CRC Press: Chapman and Hall.
18 McCurdy, T.; G. Glen; L. Smith; Y. Lakkadi. 2000. "The National Exposure Research
19 Laboratory's Consolidated Human Activity Database." Journal of Exposure Analysis and
20 Environmental Epidemiology, 10: 566-578.
21 McDonnell, W.F.; R. S. Chapman; M. W. Leigh; G. L. Strope; A. M. Collier. 1985. "Respiratory
22 Responses of Vigorously Exercising Children to 0.12 ppm Ozone Exposure." American
23 Review of Respiratory Disease, 132:875-879.
24 McDonnell, W. F.; H. R. Kehrl; S. Abdul-Salaam; P. J. Ives; L. J. Folinsbee; R. B. Devlin; J. J.
25 O'Neil; D. H. Horstman. 1991. "Respiratory Response of Humans Exposed to Low Levels
26 of Ozone for 6.6 hours." Archives of Environmental Health, 46(3): 145-150.
27 McDonnell, W.F. et al. 1993. "Predictors of Individual Differences in Acute Response to Ozone
28 Exposure." American Review of Respiratory Disease, 147:818-825.
29 McDonnell, W.F. et al. 1997. "Prediction of ozone-induced FEVi changes." American Journal of
30 Respiratory and Critical Care Medicine, 156:715:722.
31 McDonnell, W.F.; P. W. Stewart and M. V. Smith. 2007. "The Temporal Dynamics of Ozone-
32 induced FEVi Changes in Humans: An Exposure-response Model." Inhalation Toxicology,
33 19:483-494.
34 McDonnell, W.F.; P. W. Stewart and M. V. Smith. 2010. "Prediction of Ozone-induced Lung
35 Function Responses in Humans." Inhalation Toxicology, 22(2): 160-8.
36 McDonnell, W.F.; P. W. Stewart; M. V. Smith; C. S. Kim and E. S. Schelegle. 2012. "Prediction
37 of Lung Function Response for Populations Exposed to a Wide Range of Ozone
38 Conditions." Inhalation Toxicology, 24:619-633.
39
6-51
-------
1 Narayanan, M.; J. Owers-Bradley; C. S. Beardsmore; M. Mada; I. Ball; R. Garipov; K. S.
2 Panesar; C. E. Kuehni; B. D. Spycher; S. E. Williams; M. Silverman. 2012.
3 "Alveolarization Continues During Childhood and Adolescence: New Evidence from
4 Helium-3 Magnetic Resonance." American Journal of Respiratory and Critical Care
5 Medicine, January 15; 185(2): 86-191. Available at:
6 .
7 Passannante, A. N.; M. J. Hazucha; P. A. Bromberg; E. Seal; L. Folinsbee; G. Koch. 1998.
8 "Nociceptive Mechanisms Modulate Ozone-induced Human Lung Function Decrements."
9 Journal of Applied Physiology, 85:1863-1870.
10 Romieu, I; J. J. Sienra-Monge; M. Ramirez-Aguilar; M. M. Tellez-Rojo; H. Moreno-Macias; N.
11 I. Reyes-Ruiz; B. E. Del Rio-Navarro; M. X. Ruiz-Navarro; G. Hatch; R. Slade; M.
12 Hernandez-Avila. 2002. "Antioxidant Supplementation and Lung Functions Among
13 Children with Asthma Exposed to High Levels of Air Pollutants." American Journal of
14 Respiratory and Critical Care Medicine, 166: 703-709.
15 Samet, J.M. 2011. "Clean Air Scientific Advisory Committee (CASAC) Response to Charge
16 Questions on the Reconsideration of the 2008 Ozone National Ambient Air Quality
17 Standards." (Report number EPA-CASAC-11-004), dated March 30, 2011. Available at:
18 .
20 Schelegle, E.S.; C. A. Morales; W. F. Walby; S. Marion; R. P. Allen. 2009. "6.6-Hour Inhalation
21 of Ozone Concentrations from 60 to 87 ppb in Healthy Humans." American Journal of
22 Respiratory and Critical Care Medicine, 180:265-272.
23 Schelegle, E.S.; W. C. Adams; W. F. Walby and S. Marion. (2012). "Modeling of Individual
24 Subject Ozone Exposure Response Kinetics." Inhalation Toxicology, 24(7):401-415.
25 Schofield, W.N. 1985. "Predicting Basal Metabolic Rate, New Standards and Review of
26 Previous Work." Clinical Nutrition, 1985; 39C (suppl): 5-41.
27 Seal, Jr., E.; W. F. McDonnell; D. E. House. 1996. "Effects of Age, Socioeconomic Status, and
28 Menstrual Cycle on Pulmonary Response to Ozone." Archives of Environmental and
29 Occupational Health, 1996;51:132-137.
30 U.S. Environmental Protection Agency. 1996. Review of National Ambient Air Quality
31 Standards for Ozone: Assessment of Scientific and Technical Information - OAQPS Staff
32 Paper. Research Triangle Park, NC: EPA Office of Air Quality Planning and Standards.
33 (EPA document number EPA/452/R-96-007). Available at:
34 .
35 U.S. EPA. 2007a. Review of National Ambient Air Quality Standards for Ozone: Policy
36 Assessment of Scientific and Technical Information - OAQPS Staff Paper. Research Triangle
37 Park, NC: EPA Office of Air Quality Planning and Standards. (EPA document number
38 EPA-452/R-07-007). Available at:
39 .
40
6-52
-------
1 U. S. EPA. 2007b. Ozone Population Health Risk Assessment for Selected Urban Areas.
2 Research Triangle Park, NC: EPA Office of Air Quality Planning and Standards. (EPA
3 document number EPA-452/R-07-009). Available at:
4 .
5 U. S. EPA. 2009. Metabolically Derived Human Ventilation Rates: A Revised Approach Based
6 Upon Oxygen Consumption Rates. Washington, DC: EPA National Center for
7 Environmental Assessment. (EPA document number EPA/600/R-06/129F). Available at:
8 .
9 U.S. EPA. 2011. Oj National Ambient Air Quality Standards: Scope and Methods Plan for
10 Health Risk and Exposure Assessment. Research Triangle Park, NC: EPA Office of Air
11 Quality Planning and Standards. (EPA document number EPA-452/P-11-001). Available at:
12 .
13 U.S. EPA. 2012b. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
14 Documentation (TRIM.Expo / APEX, Version 4.4) Volume I: User's Guide. Research
15 Triangle Park, NC: EPA Office of Air Quality Planning and Standards(EPA document
16 number EPA-452/B-12-001a). Available at:
17 .
18 U.S. EPA. 2012c. Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
19 Documentation (TRIM.Expo / APEX, Version 4.4) Volume II: Technical Support
20 Document. Research Triangle Park, NC. (EPA document number EPA-452/B-12-001b).
21 Available at: .
22 U.S. EPA. 2013 a. Integrated Science Assessment of Ozone and Related Photochemical Oxidants.
23 EPA National Center for Environmental Assessment. (EPA document number EPA/600/R-
24 10/076F, 2013). Available at:
25 .
26 WinBUGS, version 1.4.3. Available at: .
6-53
-------
1
2 7 CHARACTERIZATION OF HEALTH RISK BASED ON
3 EPIDEMIOLOGICAL STUDIES
4 This chapter provides an overview of the methods used to estimate health risks in
5 selected urban areas based on application of results of epidemiology studies. Section 7.1.1
6 discusses the basic structure of the risk assessment, identifying the modeling elements and
7 related sources of input data needed for the analysis and presenting an overview of the approach
8 used in calculating health effect incidence using concentration-response (C-R) functions based
9 on epidemiological studies. Section 7.2 discusses air quality considerations. Section 7.3
10 discusses the selection of model inputs including: (a) selection of urban study areas, (b) selection
11 of epidemiological studies and specification of C-R functions, (c) specification of baseline health
12 effect incidence and prevalence rates, and (d) estimation of population (demographic) counts.
13 Section 7.4 describes how uncertainty and variability are addressed in the risk assessment,
14 including specification of the sensitivity analyses completed for the risk assessment and how
15 these differ from the core risk estimates. Section 7.5 summarizes the risk estimates that are
16 generated, including both the core estimates and sensitivity analyses. Finally, Section 7.6
17 provides an assessment of overall confidence in the risk assessment together with a set of key
18 observations regarding the risk estimates generated.
19 7.1 GENERAL APPROACH
20 7.1.1 Basic Structure of the Risk Assessment
21 This risk assessment involves the estimation of the incidence of specific health effect
22 endpoints associated with exposure to ambient 63 for defined populations located within a set of
23 urban study areas. Because the risk assessment focuses on health effect incidence experienced by
24 defined populations, it represents a form of population-level risk assessment and does not
25 estimate risks to individuals within the population. Furthermore, because it models risk for
26 residents in a set of urban study areas, it is not intended to provide an estimate of national-level
27 risk1 .
28 The general approach used in both the prior and current 63 risk assessments relies on C-
29 R functions based on effect estimates and model specifications obtained from epidemiological
30 studies. Since these studies derive effect estimates and model specifications using averages of
31 ambient air quality data from fixed-site, population-oriented monitors, uncertainty arising from
1 Chapter 8 provides a limited assessment of national risk focused on the mortality burden associated with recent O3
levels. This risk and exposure assessment does not provide an analysis of the risk reductions that would be
expected for the entire U.S. after meeting either the existing or alternative standards.
7-1
-------
1 the application of these functions in an O^ risk assessment is decreased if, in modeling risk, we
2 also use ambient air quality data at fixed-site, population-oriented monitors to characterize
3 exposure. Therefore, we developed a composite monitor for each urban study area to represent a
4 surrogate population exposure by averaging O?, concentrations across the monitors in that study
5 area to produce a single composite hourly time series of values. The 63 metrics used in
6 evaluating risk are derived from the composite monitor hourly time series distribution (see
7 sections 7.2 and Chapter 4 for additional detail on the characterization of ambient 63 levels).2
8 The general 63 health risk model, illustrated in Figure 7-1, combines 63 air quality data,
9 C-R functions, baseline health incidence and prevalence data, and population data (all specific to
10 a given urban study area) to derive estimates of the annual incidence of specified health effects
11 for that urban study area attributable to OT, exposure. This risk assessment models risk for 12
12 urban study areas we selected to provide coverage for the types of urban 03 scenarios likely to
13 exist across the U.S. (see section 7.3.1). Chapter 8 provides an assessment of the degree to which
14 the 12 selected urban areas are representative of other urban areas in the U.S. that are likely to
15 experience elevated risks from exposure to ambient 63 under recent conditions.
16 This risk assessment provides an updated set of estimates for risk under recent 63
17 conditions and just meeting the existing standard, and additional estimates of risk if alternative
18 standards are just met, with an emphasis on reductions in risk between just meeting the existing
19 standard and just meeting alternative standards (the full set of risk estimates, including
20 simulation of risk under current conditions is presented in Appendix 7-B). The alternative
21 standard levels evaluated are 70, 65 and 60 ppb (expressed using the current form of the 63
22 standard).
23 We simulated just meeting the existing and alternative 63 standards by adjusting hourly
24 63 concentrations measured over the 63 season using a model-based adjustment methodology
25 that estimates Os sensitivities to precursor emissions changes.3 These sensitivities, which
26 estimate the response of 63 concentrations to reductions in anthropogenic NOx and VOC
27 emissions, are developed using the Higher-order Decoupled Direct Method (HDDM) capabilities
28 in the Community Multi-scale Air Quality (CMAQ) model. More details on the HDDM-
29 adjustment approach is presented in Chapter 4 of this RE A and in Simon et al. (2013).
30 As discussed in Chapters 2 and 3, in modeling risk we employ continuous non-threshold
31 C-R functions relating 63 exposure to health effect incidence. The use of non-threshold
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 average of maximum hourly values
derived for each O3 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 REA, 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 this draft, and received
support for the model based approach from CASAC (Frey, H.D., 2012).
7-2
-------
1 functions reflects the discussion of the relevant studies in the Os ISA (see Os ISA, section
2 2.5.4.4, U.S. EPA 2013a). However, also consistent with the conclusions of the Os ISA, we
3 recognize that the evidence from the studies indicates less confidence in specifying the shape of
4 the C-R function at O^ concentrations towards the lower end of the distribution of data used in
5 fitting the curve due to the reduction in the number of data points available. The 63 ISA noted
6 that the studies indicate reduced certainty in specifying the shape of the C-R function
7 specifically for short-term (Vattributable respiratory morbidity and mortality, in the range
8 generally below 20 ppb (for both Shr-maximum and 24hr metrics) (Os ISA, section 2.5.4.4).
9 However, care needs to be taken in interpreting this range of reduced confidence indicated in the
10 studies and applying it to the interpretation of risk estimates generated for a specific urban study
11 area. This is because there is considerable heterogeneity in the effect of Os on mortality across
12 urban study areas (Os ISA section 6.6.2.3). Additionally, it is likely that levels of confidence
13 associated with C-R functions (including ranges of reduced confidence in specifying the
14 function) also vary across urban study areas reflecting underlying differences in factors
15 impacting the exposure-response relationship for 63, such as demographic differences and
16 exposure measurement error. For these reasons, the <20 ppb range discussed in the 63 ISA
17 should be viewed as a more generalized range to be considered qualitatively or semi-
18 quantitatively, along with many other factors, when interpreting the risk estimates rather than as
19 a fixed, bright-line.4
20 Based on comments we received from CAS AC on the 1st draft REA, we are no longer
21 including estimates of risk down to the lowest measured level (LML).5 Instead, through the use
22 of heat map tables, we focus on providing estimates of total risk, and the distribution of risk over
23 concentrations of 63.6 Coupled with information about what the studies indicate about the C-R
24 function at lower 63 concentrations, this provides for a more complete understanding of
25 confidence in estimated risk than simply truncating risk at the LML.
26 In modeling risk for all health endpoints included in the analysis, for recent 63 conditions
27 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 REA, CASACrecommended 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 REA (Frey and Samet, 2012). However, they recommend 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
-------
1 existing and alternative standards, we estimated both total risk as well as the difference in risk,
2 representing the degree of risk reduction associated with just meeting the existing and alternative
3 standard levels. When calculating risk differences, we focus on comparing total risk after just
4 meeting each alternative standard with total risk after just meeting the existing standards. We
5 also evaluate the incremental change in risk from meeting increasingly lower alternative standard
6 levels. Risk results are presented in terms of absolute numbers and changes in the Os attributable
7 incidence of mortality and morbidity, and in terms of the percent of baseline mortality and
8 morbidity attributable to 63. We also provide risks per 100,000 population (to normalize risks
9 across urban areas with different size populations to facilitate comparisons).
10 As with previous NAAQS-related risk assessments, for this analysis we have generated
11 two categories of risk estimates, including a set of core (or primary) estimates and an additional
12 set of sensitivity analyses. The core risk estimates utilize C-R functions based on
13 epidemiological studies for which we have relatively greater overall confidence and which
14 provide the best coverage for the broader Os monitoring period (rather than focusing only on the
15 summer season). Although it is not strictly possible to assign quantitative levels of confidence to
16 these core risk estimates due to data limitations, they are generally based on inputs having higher
17 overall levels of confidence relative to risk estimates that are generated using other C-R
18 functions. Therefore, emphasis is placed on the core risk estimates in making observations
19 regarding total risk and risk reductions associated with recent conditions and after just meeting
20 the existing and alternative standard levels. By contrast, the sensitivity analysis results typically
21 reflect application of C-R functions covering a wider array of design elements which can impact
22 risk (e.g., length of season, copollutants models, lag structures, statistical modeling methods etc).
23 The sensitivity analysis results provide insights into the potential impact of these design elements
24 on the core risk estimates, thereby informing our characterization of overall confidence in the
25 core risk estimates.7 We have significantly expanded our sensitivity analysis relative to that
26 completed for the 1st draft REA to address a wider range of modeling elements which can impact
27 the core risk estimates. Details of the design of the core and sensitivity analyses (including
28 modeling element composition) for each of the health effect endpoints categories covered in this
29 risk assessment are presented in section 7.4.3 and briefly summarized below.
30 For short-term exposure related mortality, our core analysis is based on application of C-
31 R functions obtained from the Smith et al., 2009 epidemiological study (see section 7.3.2). In
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
-------
1 addition, we have completed an expanded array of sensitivity analyses which provide coverage
2 for a number of modeling elements including: (a) time period reflected in risk modeling (summer
3 season versus full monitoring period), (b) peak Os metric (8hr maximum versus 8hr mean) (c)
o
4 use of regional versus national-based Bayesian adjustment in deriving effect estimates, (d) use
5 of single (Os-only) versus copollutant (63 and PMio) models, 9 (e) application of alternative C-R
6 functions based on Zanobetti and Schwartz, 2008 (see section 7.3.2) and (f) size of the urban
7 study area (CBSA versus smaller multi-county study area)10 (see sections 7.4.3 and 7.5.3 for
8 additional detail on the sensitivity analyses completed). In addition to these sensitivity analyses,
9 we have considered alternative methods for adjusting air quality to attain existing and alternative
10 standards (NOx-only versus combination of VOC and NOx reductions). Additional sensitivity
11 analyses exploring lag structure may also provide useful information, but are not possible due to
12 the lack of availability of Bayes adjusted estimates for alternative lag structures.
13 For short-term exposure morbidity, we have effect estimates covering a wide range of
14 design elements including co-/single-pollutant models and lag structure. However, we were not
15 in a position to differentiate between these alternative model forms in terms of overall
16 confidence and have therefore included all of these estimates in the core analysis. This range of
17 risk estimates can also be viewed as a sensitivity analysis where there is no clear "core" estimate
18 and instead, the full range of risk estimates is considered to provide the best overall picture of
19 risk for a specific endpoint (see section 7.3.2 and 7.4.3).
20 Our analysis also includes estimates of long-term exposure related respiratory mortality,
21 including a core estimate based on a co-pollutant model (with PM2.5) together with sensitivity
22 analyses exploring regional heterogeneity in the effect estimate and application of a national-
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 O3-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 copollutants model results are limited by the reduced number of days with copollutants 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 copollutants 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 surburban
areas that are socioeconomically tied to the urban center by commuting. CBSAs tend to be significantly larger
than the study areas used in the epidemiological studies providing effect estimates. We have used risk estimates
based on CBSAs 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
-------
1 level estimates focusing only on Oj, (see section 7.5.3).u The decision to model this endpoint is
2 based on our evaluation of the evidence as summarized in the 63 ISA and comments received
3 from CASAC based on the 1st draft risk assessment (Frey H.D., 2012 p. ).
4 As noted earlier, for this draft, we have modeled all core risk estimates using study areas
5 based on the core-based statistical area (CBSA) regardless of whether the epidemiological
6 studies providing the effect estimates used the CBSA spatial definition or a different spatial
7 study area definition. The decision to use CBSA-based study areas in all core simulations for this
8 draft reflects our desire to better represent the changes in risk that could be experienced in the urban
9 areas and avoid introducing substantial known bias into the risk estimates. As discussed in
10 Chapter 4 (section 4.3.1.2), most nonattaining Oj monitors are not located in the center of the urban
11 area, but instead in the surrounding areas, reflecting the transport and atmospheric chemistry
12 governing OT, formation. The monitors in the urban core areas are usually most affected by local
13 sources of NOx and experience lower concentrations of O^ since the NO is titrating the O?, in these
14 areas. For these monitors, simulating attainment of the existing and alternative standard levels can
15 result in an increase in Os concentrations, while areas further out from the core experience the
16 expected reduction in O^ level. Had we focused risk estimates on the smaller urban core areas
17 used in some of the epidemiological studies, we would not have fully captured the changes in
18 risk estimated to be experienced by the broader urban area since we would have been focusing
19 only on those areas experiencing net increases in Os (when simulating attainment of the existing
20 and alternative standard levels). By modeling risk for the core analysis using the more inclusive
21 CBSA study areas, we insure that risk estimates will include consideration both for the relatively
22 smaller core urban areas experiencing increases in 63 as well as the broader urban and suburban
23 area experiencing risk reductions. We will also insure that, to a greater extent, the analysis
24 includes the county with the design value monitor in the assessment of risk (see section 7.2).
25 There is a degree of uncertainty introduced through application of effect estimates to
26 study areas (i.e., CBSAs) that do not match those used in the underlying epidemiological studies.
27 This uncertainty should be viewed within the context of the overall larger uncertainty associated
28 with transferring effect estimates from the context of the epidemiological studies to the context
29 of the risk assessment. The epidemiological studies used in modeling short-term exposure-related
30 endpoints generate effect estimates based on day to day variation in 63 and health effects, using
31 the area wide average 63 concentrations. Area wide 63 averaging masks the specific population
32 distribution of O^ exposures which reflects the times and durations of exposures to 03 measured
33 at individual monitors in an urban area. We apply those effect estimates to the air quality
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
-------
1 scenarios of just meeting existing and alternative standards, where we are shifting the entire
2 distribution of daily 63 concentrations, and altering the relationships between 63 concentrations
3 at different monitors, and thus likely altering the relationship between area wide average Os and
4 the population distribution of Os exposures. By doing so, we introduce an additional source of
5 exposure measurement error, which goes beyond the impact that measurement error has on the
6 effect estimate, and introduces additional uncertainty into the estimates of risk associated with
7 simulating meeting existing and alternative standards.
8 Our decision to use the CBSA to define the spatial extent of each urban study area
9 reflects the greater weight we place on minimizing biases relative to minimizing uncertainty,
10 although we strive to minimize both where possible. The sensitivity analysis related to using
11 study-based spatial definitions for urban areas shows clearly that using the smaller urban areas
12 biases downward the risk reductions across an urban area. Thus, to avoid this bias in risk
13 estimates we accept a measure of increased uncertainty associated with the application of effect
14 estimates to study areas that are larger than those used in some of the original epidemiological
15 studies providing those effect estimates.
16 Using the CBSA definitions of urban areas can partially address the bias caused by
17 focusing only on urban core areas. However, it does not address this bias fully in some areas
18 because of the unevenness in monitoring throughout urban areas. In some urban areas the
19 monitors are more evenly distributed across the CBSA, while in other areas they are not. For
20 example, in some urban areas, there is a high density of monitors in the urban core counties, with
21 less density of monitors in surrounding counties also in the CBSA. Because we use a simple
22 average (to match the averaging used in the epidemiology studies) of monitors across the CBSA,
23 this means that 63 concentrations in areas where there are more monitors (e.g. in urban core
24 counties) will get a higher weight in the average 63 concentrations relative to 63 concentrations
25 in other parts of the CBSA. To the extent that the area with the higher density of monitors
26 experiences increases in 63 while the remaining area experiences decreases in 63, the overall
27 average Os concentrations applied to populations in the entire CBSA will be weighted more
28 towards O^ increases, which will attenuate the overall risk reduction that may be associated with
29 meeting alternative 63 standards. We are not able to determine the magnitude of this remaining
30 bias; however, it is expected to be higher in locations with a high percentage of total CBSA
31 monitors concentrated in urban core counties.
32 The risk assessment reflects consideration for five years of recent air quality data from
33 2006 through 2010, with these five years reflecting two three-year attainment simulation periods
34 that share a common overlapping year (i.e., 2006-2008 and 2008-2010 - see section 7.2). We
35 selected these two attainment simulation periods to provide coverage for a more recent time
36 period with relatively elevated 03 levels (2006-2008) and recent time period with relatively
7-7
-------
1 lower Os levels (2008-2010). For the REA, we model risk for the middle year of each three-year
2 attainment simulation period in order to provide estimates of risk for a year with generally higher
3 Os levels (2007) and a year with generally lower Os levels (2009). In modeling risk, we matched
4 the population data used in the risk assessment to the year of the air quality data. For example,
5 when we used 2007 air quality data, we used 2007 population estimates. For baseline incidence
6 and prevalence, rather than interpolating rates for the two specific years modeled in the risk
7 assessment, we selected the closest year for which we had existing incidence/prevalence data
8 (i.e., for simulation year 2007, we used available data for 2005 and for simulation year 2009, we
9 used data from 2010). The calculation of baseline incidence and prevalence rates is described in
10 section 7.3.4.
11 The risk assessment procedures described in more detail below are diagramed in Figure
12 7-1. To estimate the change in incidence of a given health effect resulting from a given change in
13 ambient Os concentrations in an assessment location, the following analysis inputs are necessary:
14 • Air quality information including: (1) Os air quality data from each of the
15 simulation years included in the analysis (2007 and 2009) from population-oriented
16 monitors in the assessment location (these are aggregated to form composite monitor
17 values used to represent population exposure), and (2) a method for adjusting the air
18 quality data to simulate just meeting the current or alternative suite of Os standards.
19 (These air quality inputs are discussed in more detail in Chapter 4).
20 • C-R function(s): which provide an estimate of the relationship between the health
21 endpoint of interest and Os concentrations (for this analysis, C-R functions used were
22 applied to urban study areas matching the assessment locations from the
23 epidemiological studies used in deriving the functions, in order to increase overall
24 confidence in the risk estimates generated - see section 7.3.2). For Os,
25 epidemiological studies providing information necessary to specify C-R functions are
26 readily available for Os-related health effects associated with short-term exposures
27 (Section 7.1.2 describes the role of C-R functions in estimating health risks associated
28 with O3). In addition, the Jerrett et al. (2009) study provided a C-R function for
29 modeling mortality risks associated with longer-term exposures to Os.
30 • Population information (baseline health affects incidence and prevalence rates
31 and population): The baseline incidence provides an estimate of the incidence rate
32 (number of cases of the health effect per year or day, depending on endpoint, usually
33 per 10,000 or 100,000 general population) in the assessment location corresponding
34 to recent ambient Os levels in that location. The baseline prevalence rate describes the
35 prevalence of a given disease state or conditions (e.g., asthma) within the population
-------
1 (number of individuals with the disease state/condition, usually per 10,000 or 100,000
2 general population). To derive the total baseline incidence or prevalence per year, this
3 rate must be multiplied by the corresponding population number (e.g., if the baseline
4 incidence rate is number of cases per year per 100,000 population, it must be
5 multiplied by the number of 100,000s in the population) (Section 7.3.4 summarizes
6 considerations related to the baseline incidence and prevalence rates and population
7 data inputs to the risk assessment).
8
9 In addition to the inputs described above, it is also necessary to specify the spatial extent
10 of the study areas that will be modeled. These study areas definitions determine the composition
11 of (a) the composite monitor values (which specific set of monitors are used in constructing the
12 composite monitor, reflecting the area-wide average across monitors for each study area), (b) the
13 specific set of effect estimates that will be used (matching the study areas to the specific set of
14 effect estimates in the epidemiological studies being used to support modeling of endpoints), (c)
15 the baseline incidence data and (d) the population demographic (count) data for each study area.
16 As mentioned earlier, for this REA we have modeled 12 urban study areas and have used the
17 CBSA spatial definition to specify the extent of each of these urban areas (see section 7.3.1 for
18 additional details on study area selection).
7-9
-------
1
! Concentration-Re
Identify
location- specific
1 epidemiological
I studies from the ISA
I
V
Identify appropriate
modeling period
I
AirQuality Inputs (Chapter4)
r ~\
Composite Monitor
03 Metrics for recent
conditions in Urban Case
Study Areas
V J
r 1 '
Composite Monitor
03 Metrics in Urban Case
Study Areas afterjust meeting i
Existing and alternative standards
J
ssponse Functions i
Convert
Identify RKtoS
, Relative Risk
(RR) or slope 1
coefficents (R)
J v
* s- ~*^ 1
Identify / Set of location V
functional form \specific C-R functions p~
'••^ ^ '
I
!
1 Population Information
i f
Daily location-specific
baseline health incidenc
I I
-
•r ;
I j
/-
Population living in
Study area
1 I
L
!
y.
Calculate daily changes in ozone
between just meeting current
' BenMAP
Compute day by day
' ozone attributable
incidence of mortality I
j and morbidity j
I i
Compute day by day
changes in incidence of
I mortality and morbidity <— ' '
attributable to meeting j
alternative standards
_._
V V
Estimates of % attributable ^ f Estimates of ozone- ^
Incidence and change attributable incidence of
in % attributable incidence mortality and morbidity
of mortality and morbidity and change in ozone
forthe modeling period , attributable incidence
\^ forthe modeling period )
Figure 7-1 Flow Diagram of Risk Assessment for Short-term Exposure Studies
7-10
-------
1 This risk assessment was implemented using the EPA's Environmental Benefits Mapping
2 and Analysis Program—Community Edition, Version 0.63 (BenMAP-CE) (U.S. EPA, 2013b).
3 This GIS-based computer program draws upon a database of population, baseline
4 incidence/prevalence rates and effect coefficients to automate the calculation of health impacts.
5 For this analysis, the standard set of effect coefficients and health effect incidence data available
6 in BenMAP has been augmented to reflect the latest studies and data available for modeling Os
7 risk. EPA has traditionally relied upon the BenMAP program to estimate the health impacts
8 avoided and economic benefits associated with adopting new air quality rules. For this analysis,
9 EPA used the model to estimate Os-related risk for the suite of health effects endpoints described
10 in section 3.2. There are three primary advantages to using BenMAP for this analysis, as
11 compared to the procedure for estimating population risk followed in the last review. First, once
12 we have configured the BenMAP software for this particular O^ analysis, the program can
13 produce risk estimates for an array of modeling scenarios across a large number of urban areas.
14 Second, the program can more easily accommodate a variety of sensitivity analyses. Third,
15 BenMAP allowed us to complete the national assessment of 63 mortality described in Chapter 8,
16 which plays in important role in assessing the representativeness of the urban study area analysis.
17 7.1.2 Calculating Os-Related Health Effects Incidence
18 The C-R functions used in the risk assessment are empirically estimated associations
19 between average ambient concentrations of Os and the health endpoints of interest (e.g.,
20 mortality, hospital admissions, emergency department visits). This section describes the basic
21 method used to estimate changes in the incidence of a health endpoint associated with changes in
22 Os, using a "generic" C-R function of the most common functional form.
23 Although some epidemiological studies have estimated linear C-R functions and some
24 have estimated logistic functions, most of the studies used a method referred to as "Poisson
25 regression" to estimate exponential (or log-linear) C-R functions in which the natural logarithm
26 of the health endpoint is a linear function of Cb:
27
28 y = Befk (1)
29 where x is the ambient 63 level, y is the incidence of the health endpoint of interest at 63
30 level x, p is the coefficient relating ambient Os concentration to the health endpoint, and B is the
31 incidence at x=0, i.e., when there is no ambient 03. The relationship between a specified ambient
32 63 level, x0, for example, and the incidence of a given health endpoint associated with that level
33 (denoted as yo) is then
34
35 y0=Be^ (2)
7-11
-------
1
2 Because the log-linear form of a C-R function (equation (1) is by far the most common
3 form, we use this form to illustrate the "health impact function" used in the OT, risk assessment.
4 If we let XQ denote the baseline (upper) 03 level, and xi denote the lower 03 level, and yo
5 and yi denote the corresponding incidences of the health effect, we can derive the following
6 relationship between the change in x, Ax= (XQ- xi), and the corresponding change in y, Ay, from
7 equation (I).12
8 Av = (v — v } = v \] — e~^~\ (3}
9
10 Alternatively, the difference in health effects incidence can be calculated indirectly using
11 relative risk. Relative risk (RR) is a measure commonly used by epidemiologists to characterize
12 the comparative health effects associated with a particular air quality comparison. The risk of
13 mortality at ambient O?, level XQ relative to the risk of mortality at ambient O?, level xi, for
14 example, may be characterized by the ratio of the two mortality rates: the mortality rate among
15 individuals when the ambient OT, level is XQ and the mortality rate among (otherwise identical)
16 individuals when the ambient Os level is xi. This is the RR for mortality associated with the
17 difference between the two ambient 03 levels, XQ and xi. Given a C-R function of the form
18 shown in equation (1) and a particular difference in ambient O^ levels, Ax, the RR associated
19 with that difference in ambient 63, denoted as RRAx, is equal to epAx. The difference in health
20 effects incidence, Ay, corresponding to a given difference in ambient OT, levels, Ax, can then be
21 calculated based on this RRAx as:
22
23 Ay = (y0-yl) = y0[l-(l/RRfr.y\. (4)
24
25 Equations (3) and (4) are simply alternative ways of expressing the relationship between
26 a given difference in ambient Oi levels, Ax > 0, and the corresponding difference in health
27 effects incidence, Ay.13 These health impact equations are the key equations that combine air
28 quality information, C-R function information, and baseline health effects incidence information
29 to estimate ambient 63 health risk.
12 If Ar < 0 - i.e., if Ar = (xr x0) - then the relationship between Ar and Ay can be shown to be
Ay = (yl-y0) = y0[ef**-l\. If Ax < 0, Ay will similarly be negative. However, the magnitude of Ay will be
the same whether Ar>OorAr<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
-------
1 7.2 AIR QUALITY CONSIDERATIONS
2 Air quality data are discussed in detail in Chapter 4 of this report. Here we describe those
3 air quality considerations that are directly relevant to the estimation of health risks in the
4 epidemiology based portion of the risk assessment. As described in section 7.1.1, the risk
5 assessment uses composite (area-wide average) monitor values derived for each urban study area
6 as the basis for characterizing population exposure in modeling risk. The use of composite
7 monitors reflects consideration for the way ambient O?, data are used in the epidemiological
8 studies providing the C-R functions (see section 7.1.1). For the short-term exposure related
9 health endpoints, the composite monitor values derived for this analysis include hourly time
10 series for each study area (where the 63 value for each hour is the average of measurements
11 across the monitors in that study area reporting values for that hour). Once these composite
12 monitor hourly time series are constructed, we can then extract short-term peak O^ metrics
13 needed to model specific health effects endpoints. For short-term Os-attributable endpoints,
14 reflecting consideration for available evidence in the published literature (see section 7.3.2), we
15 have focused the analysis on short-term peak O?, metrics including Ihr maximum, 8hr mean and
16 8hr maximum. The 24 hour average has been deemphasized for this analysis, although it is still
17 used in risk modeling when use of C-R functions based on this metric allow us to cover a
18 specific health effect endpoint/location of particular interest14 (see section 7.3.2).
19 For modeling mortality risk associated with long-term Os-attributable we construct
20 seasonally-averaged maximum hourly O^ values (see section 7.3.2). The derivation of composite
21 monitor distributions used in modeling this health effect endpoint is different than that used for
22 short-term Os-attributable endpoints. Specifically, for the long-term Os-attributable endpoint we
23 first construct the seasonally-averaged peak Os metric for each monitor within a given study area
24 and then average those monitor-specific metric values together to generate a single composite
25 value to use in generating risk estimates for that study area.
26 In applying effect estimates obtained from epidemiological studies we attempted to
27 match the modeling period (e.g. Oj monitoring season) associated with each epidemiology study.
28 This increases overall confidence in the risk compared with using a single more generalized
29 specification of the modeling period. As discussed earlier, we modeled all health effect endpoints
30 for the core analysis using a CBSA-based study area. The use of the CBSA-based study areas
31 addresses potential bias that would have occurred had we focused the risk assessment on the
32 smaller core urban study areas, (see section 7.1.1). Table 7-1 identifies (a) the counties
33 associated with the CBS A definition for each of the 12 urban study areas, (b) the number of Os
34 monitors associated with each CBSA (and a flag for whether the design value monitor is
14 In order to provide estimates of respiratory-related HA for LA, we did include a C-R function based on Linn et al.,
2000, which utilizes a 24 hour average exposure metric.
7-13
-------
1
2
3
4
5
6
contained within the CBSA), (c) the number of monitors associated with the smaller Smith et al.,
2009-based study areas, and (d) the specific 63 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 4.3.2.1).
Table 7-1 Information on the 12 Urban Case 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, Stafford
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
#ofO3
Monitors
within the
CBSAa
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
7
8
9
10
11
12
a - This column presents the number of monitors within each CBSA, whether the design value falls outside
of the CBSA (denoted with an"*") and the number of monitors within the smaller Smith etal., 2009-based study
area (in parenthesis).
We estimate risk associated with recent 63 conditions as well as risk associated with
simulating just meeting the existing and alternative standards. While the derivation of composite
7-14
-------
1 monitor hourly Os distributions (and associated peak exposure metrics) for recent conditions is
2 relatively straightforward, the generation of these estimates for the scenarios of just meeting the
3 existing and alternative standards is more complex. The procedures for simulating attainment of
4 both existing and alternative O^ standards are presented in Chapter 4 and Chapter 4 appendices.
5 Summary statistics for the air metrics used in modeling risk for each of the 12 urban
6 study areas under recent conditions and simulated attainment of the existing and alternative
7 standard levels are presented in Chapter 4 (see section 4.3.3.2, Figures 4-10 (2007) and 4-11
8 (2009)).
9 7.3 SELECTION OF MODEL INPUTS AND ASSUMPTIONS
10 7.3.1 Selection of Urban Study Areas
11 This analysis focuses on modeling risk for a set of urban study areas, reflecting the goal
12 of providing risk estimates that have greater overall confidence due to the use of location-
13 specific data when available for these urban locations. In addition, given the greater availability
14 of location-specific data, a more rigorous evaluation of the impact of uncertainty and variability
15 can be conducted for a set of selected urban study areas than would be possible for a broader
16 regional or national-scale analysis. We considered the following factors in selecting the 12 urban
17 study areas included in this analysis:
18 • Air Quality Data: An urban area has reasonably comprehensive monitoring data for the
19 period of interest (2006-2010) to support the risk assessment. This criterion was
20 evaluated qualitatively by considering the number of monitors within the CBS A of the
21 prospective urban areas. Locations with one or two monitors would be excluded since
22 they had relatively limited spatial coverage in characterizing Oj, levels.
23 • Elevated Ambient 63 Levels: Because we are interested in evaluating the potential
24 magnitude of risk reductions associated with just meeting the existing and alternative Os
25 standard levels, we focus on study areas with elevated ambient 63 levels at or above the
26 existing standard, such that just meeting alternative 63 standard levels would result in
27 some degree of risk reduction.
28 • Location-specific C-R Functions: Given the health endpoints selected for inclusion in
29 the analysis (see section 7.3.2), there are epidemiological studies of sufficient quality
30 available for these urban study areas to provide the C-R functions necessary for modeling
31 risk. This criterion primarily applies to short-term epidemiological studies since the
32 associated health effect endpoints are the primary focus of the REA. Short-term Os-
7-15
-------
1 attributable epidemiological studies often include city-specific effect estimates, and in
2 some cases are multi-city studies that provide estimates for multiple cities.
3 • Baseline Incidence Rates and Demographic Data: The required urban area-specific
4 baseline incidence rates and population data are available for a recent year for at least one
5 ofthehealthendpoints.
6 • Geographic Heterogeneity: Because 63 distributions and population characteristics vary
7 geographically across the U.S., we selected urban study areas to provide coverage for
8 regional variability in factors related to 63 risk including variability in the spatial pattern
9 of 63 in the urban area, population exposure (differences in residential housing density,
10 air conditioning use and commuting patterns), demographic characteristics (baseline
11 incidence rates, SES) and variability in effect estimates. The degree to which the set of
12 urban study areas provided coverage for regional differences across the U.S. in many of
13 these 03 risk-related factors was evaluated as part of the representativeness analysis
14 presented in Chapter 8.
15 Application of the above criteria resulted in the selection of 12 urban study areas for
16 inclusion in the risk assessment including:
17 • Atlanta, GA
18 • Baltimore, MD
19 • Boston, MA
20 • Cleveland, OH
21 • Denver, CO
22 • Detroit, MI
23 • Houston, TX
24 • Los Angeles, CA
25 • New York, NY
26 • Philadelphia, PA
27 • Sacramento, CA
28 • St. Louis, MO
29
30 The specific set of counties used in defining each of the 12 urban study areas based on
31 the CBSA is presented in Table 7-1.
7-16
-------
1 7.3.2 Selection of Epidemiological Studies and Specification of Concentration-Response
2 Functions
3 Once the set of health effect endpoints to be included in the risk assessment has been
4 specified, the next step was to select the set of epidemiological studies that will provide the
5 effect estimates and model specifications used in the C-R functions. This section describes the
6 approach used in completing these tasks and presents a summary of the epidemiological studies
7 and associated C-R functions specified for use in the risk assessment.
8 In Chapter 2, section 2.5 we identified the set of health effect categories and associated
9 endpoints to be included in this assessment, based on review of the evidence provided in the 63
10 ISA (U.S. EPA, 2013a). The selection of specific health effect endpoints to model within a given
11 health effect endpoint category is an iterative process involving review of both the strength of
12 evidence (for a given endpoint) as summarized in the 63 ISA together with consideration for the
13 available epidemiological studies supporting a given endpoint and the ability to specific key
14 inputs needed for risk modeling, including effect estimates and model forms. Ultimately,
15 endpoints are only selected if (a) they are associated with an overarching effect endpoint
16 category selected for inclusion in the risk assessment and (b) they have sufficient
17 epidemiological study support to allow their modeling in the risk assessment. Health effect
18 endpoints selected for inclusion in the second draft REA include:
19
20 Short-term O3-attributable:
21 • Mortality (likely to be a casual relationship)
22 o All-cause (non-accidental)
23 o Cardiovascular
24 o Respiratory
25 • Respiratory effects (causal relationship)
26 o ED (asthma, wheeze, all respiratory symptoms)
27 o HA (COPD, asthma, all respiratory) 15
28 o Respiratory symptoms
29
30 Long-term Os-attributable:
15 Regarding COPD-related HA, the O3 ISA states that "Although limited in number, both single- and multi-city
studies consistently found positive associations between short-term O3 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 REA), 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
-------
1 • Respiratory effects, focusing on respiratory-related mortality (likely causal
2 relationship).16
3
4 We selected epidemiological studies to support modeling of the health effect endpoints
5 listed above by applying a number of criteria including17:
6 • The study was peer-reviewed, evaluated in the 63 ISA, and judged adequate by EPA
7 staff for purposes of inclusion in the risk assessment. We considered the following
8 criteria: whether the study provides C-R relationships for locations in the U.S.,
9 whether the study has sufficient sample size to provide effect estimates with a
10 sufficient degree of precision and power, and whether adequate information is
11 provided to characterize statistical uncertainty.
12 • Preference for multicity studies given that they typically have greater power and
13 reflect patterns of Os related health effects over a range of urban areas (and regions)
14 which can display variability in key risk-related factors such as exposure
15 measurement error. In the case of short-term (Vattributable mortality, we also
16 favored those multi-city studies for which we could obtain Bayesian-adjusted city-
17 specific estimates from the study authors, since these incorporate both city-specific
18 effect information with information from the broader array of cities included in the
19 study. In those instances where we did not have multi-city studies (e.g., with many of
20 the short-term respiratory-related morbidity endpoints) we use single-city studies.
21 • The study design is considered robust and scientifically defensible, particularly in
22 relation to methods for covariate adjustment, including treatment of confounders, as
23 well as treatment of effect modifiers. For example, if a given study used ecological-
24 defined variables (e.g., smoking rates) as the basis for controlling for confounding,
25 concerns may be raised as to the effectiveness of that control.
26 • The study is not superseded by another study (e.g., if a later study is an extension or
27 replication of a former study, the later study would effectively replace the former
28 study), unless the earlier study has characteristics that are clearly preferable (e.g.,
29 inclusion of copollutants models, or use of a peak exposure metric of interest).
16 The O3 ISA classifies long-term O3-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, 2012, Chapter 1). We have focused on
modeling long-term O3-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
-------
1 We applied the above criteria and selected the set of epidemiological studies presented in
2 Table 7-2 for use in specifying C-R functions (Table 7-2 also describes elements of the C-R
3 functions specified using each epidemiological study, as discussed below).
4 As part of methods refinement for this risk assessment, we considered studies that
5 utilized more sophisticated and potentially representative exposure surrogates in characterizing
6 population exposure (e.g., using population-weighted Os monitor values instead of equally-
7 weighted monitors, linking exposures in individual counties or U.S. Census tracts to the nearest
8 monitor, rather than using a composite monitor value to represent the entire study area).
9 However, analysis conducted by EPA demonstrated that use of the simpler composite monitor
10 approach (as used for other short-term (Vattributable morbidity endpoints) generated risk
11 estimates that were very close to those generated using the population-weighted Os metric (see
12 REFERENCE- Karen Wesson???). Therefore, in order to conserve time and resources, we
13 modeled this endpoint using the more generalized composite monitor-based metric. And finally,
14 a number of the long-term Os-attributable morbidity studies originally considered for modeling
15 this endpoint category did involve more complex Oj, metrics (e.g., Atkinbami et al., 2010, Meng
16 et al., 2010, and Moore et al., 2008). However, limitations in the study-level data required to
17 support risk assessment prevents us at this point from completing a quantitative risk assessment
18 for this category of health endpoints with a reasonable degree of confidence.18
19 Based on additional evaluation of the literature, we have substituted Smith et al., 2009 for
20 Bell et al., 2004 as a source of Bayes-adjusted city-specific effect estimates to support modeling
21 short-term Os-attributable mortality. This decision reflects a number of factors. The Smith et al.,
22 2009 study includes a wider range of simulations exploring sensitivity of the mortality effect to
23 different model specifications including (a) regional versus national Bayes-based adjustment, (b)
24 copollutants models considering PMio, and (c) all - year versus Os-season based estimates. This
25 is contrasted with the Bell et al., 2004 study which does not provide this degree of model
26 exploration. In obtaining the city-specific Bayes-adjusted effect estimates for the Smith et al.,
27 2009 study from the study authors, we were provided with estimates reflecting this range of
28 alternative model specifications which allowed us to incorporate them into both the core and
29 sensitivity analysis portions of the REA (see section 7.4.3). In addition, the Smith et al., 2009
30 study does not use the trimmed mean approach employed in the Bell et al., 2004 study in
31 preparing Oj, monitor data. We have a number of concerns regarding the trimmed mean approach
32 including (1) the potential loss of temporal variation in the data when the approach is used (this
33 could impact the size of the effect estimate) and (2) a lack of complete documentation for the
18 However, as noted in section 7.7.3 of the first draft REA, 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
-------
1 approach which prevents us from fully reviewing the technique and using it in preparing O^
2 metrics for the REA. Given these concerns, we view it as advantageous that the Smith et al.,
3 2009 study does not use the trimmed mean approach.
4 With the exception of the trimmed mean approach, the Smith et al., 2009 study was
5 intended to reproduce the results of the Bell et al., 2004 analysis. Thus, the core risk results
6 based on Smith et al 2009 are comparable to the 1st draft REA estimates based on Bell et al 2004,
7 while the alternative models provided in Smith et al 2009 allow for an expanded set of sensitivity
8 analyses. The comparability of the Smith et al 2009 and Bell et al 2004 estimates is confirmed by
9 the graphical comparison in Smith et al 2009 of mortality effect estimates (for the 24hr O^
10 metric) with matching effect estimates from Bell et al., 2004. This comparison demonstrates the
11 close match of the two studies (for this particular scenario).
12 Reflecting the points made above, in modeling short-term Os-attributable mortality, we
13 have included a core analysis based on the national-Bayesian adjusted city-specific effect
14 estimates (reflecting the full Os monitoring period in each city) obtained from Smith et al., 2009.
15 As sensitivity analyses, we have included effect estimates obtained from Smith et al., 2009
16 which reflect application of copollutants models (including PMio), Bayes adjustment using a
17 regional prior,19 and a shorter fixed Oj, measurement period (April-October). In the 1st draft
18 REA, we had also included national Bayes-adjusted effect estimates (reflecting a fixed June-
19 August period) obtained from Zanobetti and Schwartz, 2008 as part of the core analysis.
20 However, we have decided to instead include these as part of the sensitivity analysis in this 2n
21 draft of the REA since these effect estimates cover a more limited warm-weather period and
22 consequently will generate only partial characterizations of mortality risk (since they exclude
23 risk occurring during the non-summer months).
24 We have also included estimates of respiratory-related mortality associated with long-
25 term O?, exposures based on effect estimates obtained from Jerrett et al., 2009. The decision to
26 model long-term (Vattributable mortality reflects consideration for evidence supporting a likely
27 to be a causal relationship for long-term Os-attributable respiratory effects, including mortality
28 (Os ISA, section 2.5.2, U.S. EPA, 2013a). After considering its strengths and weaknesses, we
29 consider the Jerrett et al. (2009) study to be an appropriate basis for estimating long-term 63-
30 related respiratory mortality risk. Key strengths of this study are that it (a) included 1.2 million
31 participants in the American Cancer Society cohort from all 50 states, DC, and Puerto Rico;
32 included 63 data from 1977 (5 years before enrollment in the cohort began) to 2000; (b)
33 considered co-pollutant models that controlled for PM^.s; and (c) explored the potential for a
34 threshold concentration associated with the long-term mortality endpoint. Importantly, this study
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
-------
1 was also the first to explore the relationship between long-term Os exposure and respiratory
2 mortality (rather than focusing on cardiopulmonary mortality). Key limitations are possible
3 exposure misclassification and uncontrolled confounding by temperature, which are endemic to
4 most long-term epidemiological studies. While Jerrett et al. (2009) found negative associations
5 between 63 exposure and cardiovascular mortality when controlling for PM2.s, null or negative
6 associations for Os are consistent with the evidence that PM2.5 is the pollutant most strongly
7 associated with cardiovascular disease (EPA 2009 PM ISA).
8 Our analysis includes a core estimate based on a co-pollutant model (with PM2.5). The
9 seasonal average metrics used in the long-term exposure mortality estimate are not very sensitive
10 to the reduced number of days with co-pollutant monitoring, and as such it is appropriate to
11 include the copollutant model as the core estimate. We also include two sensitivity analyses for
12 long-term Os-attributable respiratory mortality including: (a) application of regionally-
13 differentiated effect estimates (although these do not include a copollutants model specification)
14 and (b) application of a single pollutant (Os-only) national-based effect estimate.
15 The effect estimates used in modeling long-term Os-attributable mortality (see Table 7-2)
16 utilize a seasonal average of peak (Ihr maximum) measurements. These long-term exposure
17 metrics can be viewed as long-term exposures to daily peak Os over the warmer months, as
18 compared with annual average levels such as are used in long-term PM exposure calculations.
19 This increases the need for care in interpreting these long-term Os-attributable mortality
20 estimates together with the short-term Os-attributable mortality estimates, in order to avoid
21 double counting. It is also important to keep in mind that our estimates of short-term 63-
22 attributable mortality are for all-causes, while estimates of long-term Os-attributable mortality
23 are focused on respiratory-related mortality. This further limits the ability to compare estimates
24 of long-term and short-term exposure related mortality.
25 Once the set of epidemiology studies described above was selected, the next step was to
26 specify C-R functions for use in the risk assessment. Several factors were considered in
27 identifying the effect estimates and model forms used in specifying C-R functions for each
28 endpoint. These factors are described below:
29
30 • Os Exposure Metric: In the risk assessment supporting the previous Os NAAQS review,
31 for short-term exposure, we included C-R functions based on 24hr averages as well as a
32 number of peak 63 measurements. However, given that the the current 63 NAAQS
33 standard uses an 8hr form and given that many of the clinical studies involving Os also
34 utilize shorter exposures (on the order of 2 to 8 hrs - see Os ISA, section 6.2.1.1), we
35 wanted to see if the latest epidemiological studies for Os also supported use of an 8hr
36 averaging time in modeling risk. Several epidemiological studies completed since the last
7-21
-------
1 review provide limited support for stronger associations between health endpoints and
2 peak Os metrics (i.e., Ihr maximum, 8hr maximum and 8hr means) relative to 24hr
3 averages. Specifically, a study of respiratory ED visits in Atlanta (Darrow et al., 2011)
4 found stronger associations with peak metrics (including Ihr and 8hr maximum
5 measurements) compared with 24hr averages (see Os ISA section 6.2.7.3 and Figure 6-
6 17, U.S. EPA, 2013a). Similarly, for short-term exposure-related mortality, there are also
7 a limited number of epidemiologic studies that have compared mortality associations
8 with peak Oj, metrics and the 24hr average metric. Although the 63 ISA recognizes that
9 24hr exposure metrics when used in time series studies may result in smaller risk
10 estimates, ultimately it concludes that "Overall, the evidence from time-series and panel
11 epidemiologic studies does not indicate that one exposure metric is more consistently or
12 strongly associated with mortality or respiratory-related health effects" (U.S. EPA,
13 2013a, section 2.5.4.2). Based on consideration for the evidence summarized in the Os
14 ISA, we have decided to focus on peak exposure metrics because of the limited evidence
15 that these metrics may be associated with higher risk estimates relative to the 24 hr
16 exposure metric. However, we recognize that, as summarized in the 63 ISA, there is only
17 weak support for differentiating between these two categories of short-term exposure
18 metric.
7-22
-------
Table 7-2 Overview of Epidemiological Studies Used in Specifying C-R Functions
Epidemiological
study
(stratified by Qy-
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 (^^-attributable mortality
Smith etal., 2009
Zanobetti and
Schwartz (2008)
Non-
accidental,
respiratory,
cardiovascul
ar
Non-
accidental,
respiratory,
cardiovascul
ar
95 large urban
communities
(provides
coverage for all
12 urban study
areas)
48 U.S. cities
(provides
coverage for the
12 urban study
areas)
24hr avg, 8hr max,
Ihrmax. April
through October
and all year
8hr 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 0-3d, 0-20 and
4-20 day). Age range: all ages
Focused on the 8hr max-based metric C-R
functions for the REA (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 copollutants
(PMio) models, and (c) full monitoring period
versus summer only (April-October). For the core
analysis, we focused on the single pollutant (Qj-
only) model covering the full monitoring period.
The copollutants model (with PMio) 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 8hr mean Os
levels measured between June and August.
Estimates were generated for each study area using
this constrained warm-season period.
Short-term ^^-attributable morbidity - HA for respiratory effect)
Medina-Ramon et
al., 2006.
Linn et al., 2000
HA: COPD,
pneumonia
HA:
unscheduled
36 cities
(provides
coverage for all
12 urban study
areas)
LA only
8hr mean, warm
(May-September),
cool (October-
April), all year
24hr mean, LA QT,
season (all year),
Distributed lag (0-1 day). Age
range: > 65yrs. Controlled for day
of the week and weather (including
temperature).
Lag 0. Age range: all ages. Used
subgroup analysis to explore the
Generated risk estimates based on warm season for
COPD only (May-September).
Included effect estimate based on 24hr avg metric
(for summer) since this provided additional
7-23
-------
Epidemiological
study
(stratified by Oj-
attributable health
endpoints)
Lin et al, 2008
Katsouyanni et al
2009
Silverman et al.,
2010
Health
endpoints
for
pulmonary
illness
HA:
respiratory
disease
HA:
cardiovascul
ar disease,
chronic
obstructive
pulmonary
disease,
pneumonia,
all
respiratory
HA: asthma
(ICU and
non-ICU)
Location
(urban study
area(s)
covered)
NY State (used
to cover NYC)
14 cities
(provides
coverage for
Detroit only)
NYC
Exposure metric
(and modeling
period)
winter, spring,
summer and
autumn
lhrmax(for
10am-6pm
interval), warm
season (April-
October)
Ihrmax. Summer
only and all year
8hr max. Warm
season (April-
August)
Additional study design details
effect of temporal variation,
weather and autocorrelation on ©3
effect.
Lag 0, 1,2, 3. Age range: <18yrs.
Models adjusted for the
confounding effects of
demographic characteristics,
particulate matter(PM10),
meteorological conditions,
day of the week, seasonality, long-
term trends, and different
lag periods of exposure.
Lag 0-lday. Age range: > 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 forPM25. Lag 0-1
day. Age range: children 6-1 8yrs.
The model adjusted for temporal
trends, weather, and day of the
week.
Notes regarding application in the analysis
coverage for HA in L.A. Modeled using air quality
for June- August.
Used Ihr max metric applied to the city-specific
Os season for NYC (April-October).
C-R function applied only for all respiratory
endpoint. Used June-August-based composite
monitor.
Applied C-R function (for QT, and QT, with control
for PM2 5) to the city-specific Os season for NYC
(slightly longer than the modeling period used in
the study).
Short-term O^-attributable morbidity- ED andER visits (respiratory)
Ito et al., 2007
Tolbertetal.,2007
ED: asthma
ED: all
respiratory
NYC
Atlanta
8hr max. Warm
season (April-
September)
8hr max. 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,PM10and NO2/NO2.
Age range: all ages. Model controls
for temporal trends, temperature,
Applied C-R functions (for Oj alone and Oj with
control for listed pollutants) to the city-specific Os
season for NYC (slightly longer than the modeling
period used in the study).
Applied C-R functions (for QT, alone and QT, with
control for listed pollutants) to the city-specific Os
season for Atlanta.
7-24
-------
Epidemiological
study
(stratified by Oj-
attributable health
endpoints)
Strickland etal.,
2010
Darrowetlal.,2011
Health
endpoints
ER:
respiratory
ED: all
respiratory
Location
(urban study
area(s)
covered)
Atlanta
Atlanta
Exposure metric
(and modeling
period)
8hr max (based on
population
weighted average
across monitors).
Warm season
(May to October)
and cool
(November to
April)
Shrmax, Ihrmax,
24hr avg for
summer (March-
October).
Additional study design details
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: Iday. 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.
Notes regarding application in the analysis
Included effect estimates based on both lag
structures and used composite monitor values for
city-specific ©3 season.
Used city-specific ©3 season-based composite
monitor values.
Short-term 0 ^-attributable morbidity - respiratory symptoms
Gent et al., 2003
Respiratory
symptoms:
wheeze,
persistent
cough, chest
tightness,
shortness of
breath
Springfield MA
(study used to
cover Boston)
Ihrmax, Shrmax
Lag: 0 and 1 day. Age range:
asthmatic children <12 yrs. Model
adjusted for temperature.
Included effect estimates for different symptoms
based on both 8hr max and Ihr max metrics (for
city-specific QT, 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.
Long-term O^-attributable respiratory mortality
Jerrett etal., 2009
Respiratory,
cardiovascul
ar,
cardiopulmo
nary
96 metropolitan
statistical areas
(provides
coverage for all
12 study areas)
Seasonal average
(i.e. Apr-Sep) of
the peak daily Ihr
max values.
>30 yrs of age, includes national-
level and regional effect estimates
(only national-level estimate has
copollutants modeling considering
PM2.5 along with Os). Modeling
included consideration for a range
of potential confounders evaluated
Included national copollutants model-based effect
estimates in core analysis and single-pollutant
model regional effect estimates and national effect
estimates as sensitivity analyses.
7-25
-------
Epidemiological
study
(stratified by Oj-
attributable health
endpoints)
Health
endpoints
Location
(urban study
area(s)
covered)
Exposure metric
(and modeling
period)
Additional study design details
at both the ecological level and
personal level.
Notes regarding application in the analysis
7-26
-------
1 • Single-and Multi-pollutant Models (pertains to both short-term and long-term
2 exposure studies): Epidemiological studies often consider health effects associated with
3 ambient 63 using both single-pollutant and co-pollutant models. To the extent that any of
4 the co-pollutants present in the ambient air may have contributed to health effects
5 attributed to 63 in single pollutant models, risks attributed to 63 may be overestimated or
6 underestimated if C-R functions are based on single pollutant models. This would argue
7 for inclusion of models reflecting consideration of co-pollutants. Conversely, in those
8 instances where co-pollutants are highly correlated with 63, inclusion of those pollutants
9 in the health impact model can produce unstable and statistically insignificant effect
10 estimates for both 63 and the co-pollutants. Furthermore, there are often significant
11 differences in sampling frequencies for each pollutant included in copollutants models,
12 which can lead to a loss of statistical power in copollutants models (relative to single
13 pollutant models). These last points could argue for inclusion of a model based
14 exclusively on 03. Given that single and multi-pollutant models each have potential
15 advantages and disadvantages, to the extent possible, given available information we
16 have included both types of C-R functions in the risk assessment.
17 • Multiple Effect Estimates within a Given CBSA-based Study Area: As noted earlier
18 in section 7.1.1, for this analysis, all health endpoints, including short-term 03-
19 attributable mortality are modeled using CBSA-based study areas. In the case of both
20 Smith et al., 2009 and Zanobetti and Schwartz 2008, these CBSA-based study areas are
21 larger than the study areas used in these epidemiological studies to derive effect
22 estimates. Furthermore, for some of the CBSA-based urban study areas, several of the
23 smaller study areas evaluated in the epidemiological study fall within a single larger
24 CBSA-based study area. For example, with the Smith et al., 2009 study, multiple effect
25 estimates are available for the CBSA-defmed study areas of Los Angeles and New York
26 City. Specifically, the Smith et al., 2009 study provides separate effect estimates for (a)
27 Santa Anna/Anaheim and Los Angeles study areas, both of which fall within the larger
28 CBSA-based Los Angeles study area and (b) New York, Jersey City and Newark study
29 areas, all of which fall within the larger CBSA-defmed New York study area (see Table
30 7-3). This raises the question of how to specify the effect estimate for these larger CBSA-
31 based study areas when there are multiple effect estimates available from the
32 epidemiological study. For this analysis, in those instances where there are multiple effect
33 estimates, we have decided to use the effect estimate that represents the largest number of
34 residents within each CBSA-based study area. There is uncertainty associated with this
35 decision which is discussed both in section 7.4.2 and section 7.5.3 (as part of the air
36 quality-related sensitivity analysis discussion).
7-27
-------
1
2
Table 7-3 CBSA-based Study Areas with Multiple Effect Estimates from the Smith et
al., 2009 Study*
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
CBSA
Study Area
New York
City
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.
* 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 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.
7-28
-------
1 • Multiple Lag Models: Based on our review of evidence provided in the Os ISA, we
2 believe there is increased confidence in modeling both short-term (Vattributable
3 mortality and respiratory morbidity risk based on exposures occurring up to a few days
4 prior to the health effect, with less support for associations over longer exposure periods
5 or effects lagged more than a few days from the exposure (see 63 ISA section 2.5.4.3,
6 U.S. EPA, 2013a). Consequently, we have favored C-R functions reflecting shorter lag
7 periods (e.g., 0, 1 or 1-2 days). With regard to the specific lag structure (e.g, single day
8 versus distributed lags), the 63 ISA notes that epidemiological studies involving
9 respiratory morbidity have suggested that both single day and multi-day average
10 exposures are associated with adverse health effects (see 63 ISA section 2.5.4.3).
11 Therefore, when available both types of lag structures where considered in specifying C-
12 R functions for short-term Os-attributable mortality and morbidity.
13 • Seasonally-differentiated Effects Estimates: The previous 03 Air Quality Criteria
14 Document (AQCD) (published in 2006) concluded that aggregate population time-series
15 studies demonstrate a positive and robust association between ambient Os concentrations
16 and respiratory-related hospitalizations and asthma ED visits during the warm season (see
17 O3 ISA section 2.5.3.1 U.S. EPA, 2013a). The current O3 ISA notes that recent studies of
18 short-term Os-attributable respiratory mortality in the U.S. suggest that the effect is
19 strengthened in the summer season (Os ISA section 2.5.3.1, U.S. EPA, 2013a). In
20 addition, many of the key epidemiological studies discussed in the current Os ISA
21 exploring both short-term exposure related mortality and morbidity have larger (and more
22 statistically significant) effect estimates when evaluated for the summer (63) season,
23 relative to the full year (see O3 ISA Figures 6-20 and 6-27, U.S. EPA, 2013a). However,
24 if we focus the assessment of risk on the warm season, we bias our estimate by excluding
25 potential effects associated with cooler (non-summer) months. Given our desire to
26 provide a more complete picture of overall risk in each of the study areas, we have
27 favored (for the core analysis) effect estimates that cover the full 63 monitoring period
28 specific to each study area, rather than the more limited warm (summer) period.
29 • Shape of the Functional Form of the Risk Model (including threshold): The current
30 63 ISA concludes that there is little support in the literature for a population threshold for
31 short-term (Vattributable effects. However, specifically in relation to mortality, the 63
32 ISA concludes that a national or combined analysis may not be appropriate to identify
33 whether a threshold exists (see O3 ISA, section 2.5.4.4, U.S. EPA, 2013a).20 Given the
20 Specifically, given the multi-city nature of these mortality studies combined with the variability in O3 and other
factors related to exposure and risk, the O3 ISA concludes that these studies are not well positioned to evaluate
the potential for a threshold in the mortality effect.
7-29
-------
1 above general observation from the Os ISA regarding the low potential for thresholds, we
2 did not include C-R functions for any of the short-term (Vattributable health endpoints
91
3 modeled that incorporated a threshold.
4 Application of the above criteria resulted in an array of C-R functions specified for the
5 risk assessment (see Table 7-2), including functions covering short-term (Vattributable
6 mortality and morbidity and long-term Os-attributable mortality. In presenting the C-R functions
7 in Table 7-2, we have focused on describing key attributes of each C-R function (and associated
8 source epidemiological study) relevant to a review of their use in the risk assessment. More
9 detailed technical information including effect estimates and model specification is provided in
10 Appendix 7A. Specific summary information provided in Table 7-2 includes:
11 • Health endpoints: identifies the specific endpoints evaluated in the study. Generally
12 we included all of these in our risk modeling, however, when a subset was modeled,
13 we reference that in the "Notes" column (last column in the table).
14 • Location: identifies the specific urban areas included in the study and maps those to
15 the set of 12 urban study areas included in the risk assessment.
16 • Exposure metric: describes the exposure metric used in the study, including the
17 specific modeling period (e.g., 63 season, warm season, full year). We developed two
18 categories of composite monitor values to match the modeling periods used in the two
19 short-term Os-attributable mortality studies providing C-R functions for the analysis.
20 For the remaining morbidity endpoints, we mapped specific C-R functions to
21 whichever of these two composite monitor categories most closely matched the
22 modeling period used in the underlying epidemiological study. This mapping (for
23 morbidity endpoint C-R functions) is described in the "Notes" column (the seasons
24 reflecting in modeling for each C-R function are also presented in Appendix 7A).
25 • Additional study design details: this column provides additional information primarily
26 covering the lag structure and age ranges used in the study.
27 • Notes regarding application in second draft analysis: as the name implies, this
28 column provides notes particular to the application of a particular epidemiological
29 study and associated C-R functions in the risk assessment.
21 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. 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
-------
1 7.3.3 Baseline Health Effect Incidence and Prevalence Data
2 As discussed earlier (section 7.1.2), the most common epidemiological-based health risk
3 model expresses the change in health risk (Ay) associated with a given change in 63
4 concentrations (Ax) as a percentage of the baseline incidence (y). To accurately assess the impact
5 of Os air quality on health risk in the selected urban areas, information on the baseline incidence
6 of health effects (i.e., the incidence under recent air quality conditions) in each location is
7 needed. In some instances, health endpoints are modeled for a population with an existing health
8 condition, necessitating the use of a prevalence rate. Where at all possible, we use county-
9 specific incidences or incidence rates (in combination with county-specific populations). In some
10 instances, when county-level incidence rates were not available, BenMAP can employ more
11 generalized regional rates (see BenMAP Guidance Manual for additional detail, Abt Associates,
12 Inc. 2010). For prevalence rates (which were only necessary for modeling respiratory symptoms
13 among asthmatic children using Gent et al., (2008) - see Table 7-2), we utilized a national-level
14 prevalence rate appropriate for the age group being modeled. A summary of available baseline
15 incidence data for specific categories of effects (and prevalence rates for asthma) is presented
16 below:
17 • Baseline incidence data on mortality: County-specific (and, if desired, age- and race-
18 specific) baseline incidence data are available for all-cause and cause-specific
99
19 mortality from CDC Wonder. The most recent year for which data are available
20 online is 2005 and this was the source of incidence data for the risk assessment.23
21 • Baseline incidence data for hospital admissions and emergency room (ER) visits:
22 Cause-specific hospital admissions baseline incidence data are available for each of
23 40 states from the State Inpatient Databases (SID). Cause-specific ER visit baseline
24 incidence data are available for 26 states from the State Emergency Department
25 Databases (SEDD). SID and SEDD are both developed through the Healthcare Cost
26 and Utilization Project (HCUP), sponsored by the Agency for Healthcare Research
27 and Quality (AHRQ). In addition to being able to estimate State-level rates, SID and
28 SEDD can also be used to obtain county-level hospital admission and ER visit counts
29 by aggregating the discharge records by county.
30 • Asthma prevalence rates: state-level prevalence rates that are age group stratified are
31 available from the Centers for Disease Control and Prevention (CDC) Behavioral
32 Risk Factor Surveillance System (BRFSS) (U.S. CDC, 2010).
http://wonder.cdc.gov/mortsql.html
23 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
-------
1 Incidence and prevalence rates are presented as part of the full set of model inputs
2 documented in Appendix 7A. The incidence rates and prevalence rates provided in Table 7A-1
3 are weighted average values for the age group associated with each of the C-R functions. These
4 weighted averages are calculated within BenMAP using more refined age-differentiated
5 incidence and prevalence rates originally obtained from the data sources listed in the bullets
6 above.
7 7.3.4 Population (demographic) Data
8 To calculate baseline incidence rates, in addition to the health baseline incidence data we
9 also need the corresponding population. We obtained population data from the 2010 U.S. Census
10 (http://www.census.gov/popest/counties/asrh/). These data are then used as the basis for back-
11 casting estimates for simulation years (in this case, 2007 and 2009) (see Appendix J of the
12 BenMAP User's Manual for additional detail, U.S. EPA, 2012b). Total population counts used in
13 modeling each of the health endpoints evaluated in the analysis (differentiated by urban study
14 area and simulation year) are provided as part model inputs presented in Appendix 7A.
15 7.4 ADDRESSING VARIABILITY AND UNCERTAINTY
16 An important component of a population risk assessment is the characterization of both
17 uncertainty and variability. Variability refers to the heterogeneity of a variable of interest within
18 a population or across different populations. For example, populations in different regions of the
19 country may have different behavior and activity patterns (e.g., air conditioning use, time spent
20 indoors) that affect their exposure to ambient 63 and thus the population health response. The
21 composition of populations in different regions of the country may vary in ways that can affect
22 the population response to exposure to 63 - e.g., two populations exposed to the same levels of
23 Os might respond differently if one population is older than the other. Variability is inherent and
24 cannot be reduced through further research. Refinements in the design of a population risk
25 assessment are often focused on more completely characterizing variability in key factors
26 affecting population risk - e.g., factors affecting population exposure or response - in order to
27 produce risk estimates whose distribution adequately characterizes the distribution in the
28 underlying population(s).
29 Uncertainty refers to the lack of knowledge regarding the actual values of inputs to an
30 analysis. Models are typically used in analyses, and there is uncertainty about the true values of
31 the parameters of the model (parameter uncertainty) - e.g., the value of the coefficient for 63 in a
32 C-R function. There is also uncertainty about the extent to which the model is an accurate
33 representation of the underlying physical systems or relationships being modeled (model
34 uncertainty) - e.g., the shapes of C-R functions. In addition, there may be some uncertainty
7-32
-------
1 surrounding other inputs to an analysis due to possible measurement error—e.g., the values of
2 daily 63 concentrations in a risk assessment location, or the value of the baseline incidence rate
3 for a health effect in a population.24 In any risk assessment, uncertainty is, ideally, reduced to the
4 maximum extent possible through improved measurement of key variables and ongoing model
5 refinement. However, significant uncertainty often remains, and emphasis is then placed on
6 characterizing the nature of that uncertainty and its impact on risk estimates. The characterization
7 of uncertainty can be both qualitative and, if a sufficient knowledgebase is available,
8 quantitative.
9 The selection of urban study areas for the O^ risk assessment was designed to cover the
10 range of (Vrelated risk experienced by the U.S. population and, in general, to adequately reflect
11 the inherent variability in those factors affecting the public health impact of Os exposure.
12 Sources of variability reflected in the risk assessment design are discussed in section 7.4.1, along
13 with a discussion of those sources of variability which are not fully reflected in the risk
14 assessment and consequently introduce uncertainty into the analysis.
15 The characterization of uncertainty associated with risk assessment is often addressed in
16 the regulatory context using a tiered approach in which progressively more sophisticated
17 methods are used to evaluate and characterize sources of uncertainty depending on the overall
18 complexity of the risk assessment (WHO, 2008). Guidance documents developed by EPA for
19 assessing air toxics-related risk and Superfund Site risks (U.S.EPA, 2004 and 2001, respectively)
20 as well as recent guidance from the World Health Organization (WHO, 2008) specify multi-
21 tiered approaches for addressing uncertainty.
22 The WHO guidance, in particular, presents a four-tiered approach for characterizing
23 uncertainty (see Chapter 3, section 3.2.6 for additional detail on the four tiers included in the
24 WHO's guidance document). With this four-tiered approach, the WHO framework provides a
25 means for systematically linking the characterization of uncertainty to the sophistication of the
26 underlying risk assessment. Ultimately, the decision as to which tier of uncertainty
27 characterization to include in a risk assessment will depend both on the overall sophistication of
28 the risk assessment and the availability of information for characterizing the various sources of
29 uncertainty. We used the WHO guidance as a framework for developing the approach used for
30 characterizing uncertainty in this risk assessment.
31 The overall analysis in the Os NAAQS risk assessment is relatively complex, thereby
32 warranting consideration of a full probabilistic (WHO Tier 3) uncertainty analysis. However,
33 limitations in available information prevent this level of analysis from being completed at this
24 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
-------
1 time. In particular, the incorporation of uncertainty related to key elements of C-R functions
2 (e.g., competing lag structures, alternative functional forms, etc.) into a full probabilistic WHO
3 Tier 3 analysis would require that probabilities be assigned to each competing specification of a
4 given model element (with each probability reflecting a subjective assessment of the probability
5 that the given specification is the "correct" description of reality). However, for many model
6 elements there is insufficient information on which to base these probabilities. One approach that
7 has been taken in such cases is expert elicitation; however, this approach is resource- and time-
8 intensive and consequently, it was not feasible to use this technique in the current Oj NAAQS
9S
9 review to support a WHO Tier 3 analysis.
10 For most elements of this risk assessment, rather than conducting a full probabilistic
11 uncertainty analysis, we have included qualitative discussions of the potential impact of
12 uncertainty on risk results (WHO Tierl). As discussed in section 7.1.1, for this draft of the risk
13 assessment, we have also expanded the sensitivity analysis considerably to cover a range of
14 model elements (this represents a WHO Tier 2 analysis). The specific modeling elements
15 covered in the sensitivity analysis for each health effects endpoint together with the specification
16 of the core analysis is presented in section 7.4.3. As part of the sensitivity analysis, we have also
r\r
17 completed an influence analysis using estimated elasticities of response designed to determine
18 which of the input factors used in calculating risk are primarily responsible for inter-city
19 variability in risk. This influence analysis focuses on the response of core short-term exposure-
20 related mortality risk to inputs since this is one of the key risk metrics completed for the REA
21 (see section 7.4.3).
22 In addition to the qualitative and quantitative treatment of uncertainty and variability
23 which are described here, we have also completed an analysis to evaluate the representativeness
24 of the selected urban study areas against national distributions for key Oj risk-related attributes
25 to determine whether they are nationally representative or more focused on a particular portion
26 of the distribution for a given attribute (see Chapter 8, section 8.2.1). In addition, we have
27 completed a second analysis addressing the representativeness issue, which identified where the
28 12 urban study areas included in this risk assessment fall along a distribution of national-level
29 short-term and long-term exposure-related mortality risk (see Chapter 8, section 8.2.2). This
30 analysis allowed us to assess the degree of which the 12 urban study areas capture locations
31 within the U.S. likely to experience elevated levels of risk related to Os exposure (for both short-
32 term and long-term Os-attributable mortality).
25 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.
26 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
-------
1 The remainder of this section is organized as follows. Key sources of variability which
2 are reflected in the design of the risk assessment, along with sources excluded from the design,
3 are discussed in section 7.4.1. A qualitative discussion of key sources of uncertainty associated
4 with the risk assessment (including the potential direction, magnitude and degree of confidence
5 associated with our understanding of the source of uncertainty - the knowledge base) is
6 presented in section 7.4.2. The design of the core analysis and sensitivity analysis completed for
7 each of the health effect endpoint categories modeled in the risk assessment is discussed in
8 section 7.4.3.
9 7.4.1 Treatment of Key Sources of Variability
10 The risk assessment was designed to cover the key sources of variability related to
11 population exposure and exposure response, to the extent supported by available data. Here, the
12 term key sources of variability refers to those sources that we believe have the potential to play
13 an important role in impacting population incidence estimates generated for this risk assessment.
14 Specifically, hawse have concluded that these sources of variability, if fully addressed and
15 integrated into the analysis, could result in adjustments to the core risk estimates which might be
16 relevant from the standpoint of interpreting the risk estimates in the context of the Os NAAQS
17 review. The identification of sources of variability as "key" reflects consideration for sensitivity
18 analyses conducted for previous 63 NAAQS risk assessments, which have provided insights into
19 which sources of variability can influence risk estimates, as well as information presented in the
20 O3 ISA.
21 As with all risk assessments, there are sources of variability which have not been fully
22 reflected in the design of the risk assessment and consequently introduce a degree of uncertainty
23 into the risk estimates. While different sources of variability were captured in the risk
24 assessment, it was generally not possible to separate out the impact of each factor on population
25 risk estimates, since many of the sources of variability are reflected collectively in a specific
26 aspect of the risk model. For example, inclusion of urban study areas from different regions of
27 the country likely provides some degree of coverage for a variety of factors associated with Os
28 risk (e.g., air conditioner use, differences in population commuting and exercise patterns,
29 weather). However, the model is not sufficiently precise or disaggregated to allow the individual
30 impacts of any one of these sources of variability on the risk estimates to be characterized.
31 Key sources of potential variability that are likely to affect population risks are discussed
32 below, including the degree to which they are captured in the design of the risk assessment:
33 • Heterogeneity in the Effect of Os on Health Across Different Urban Areas: A
34 number of studies cited in the Os ISA have found evidence for regional heterogeneity
35 in the short-term (Vattributable mortality effect (Smith et al., 2009 and Bell and
7-35
-------
1 Dominici, 2008, Bell et al., 2004, Zanobetti an Schwartz 2008 - see O3 ISA section
2 6.6.2.2, U.S. EPA, 2013a). These studies have demonstrated that differences in effect
3 estimates between cities can be quite substantial (see Os ISA Figures 6-32 and 6-33).
4 Therefore, for the short-term Os-attributable mortality endpoint modeled using Smith
5 et al., 2009-based effect estimates, we have included Bayes-adjusted city-specific
6 effect estimates reflecting application of both a regional- and national-prior, both of
7 which are intended to capture cross-city differences in effect estimates for the
8 mortality endpoint, while still reflecting input from the more stable regional, or
9 national-level signal. The national-prior based estimates are included in the core
10 analysis since they have greater overall power, while the regional-prior based
11 estimates are included as sensitivity analyses to explore the impact of using regional
97
12 prior in developing the Bayes-adjusted estimates (see section 7.4.3). For short-term
13 morbidity endpoints, typically we have used city-specific effect estimates; however,
14 for most endpoints, we only have estimates for a subset of the urban study areas
15 (typically NYC, Atlanta and/or LA). Therefore, although our risk estimates do reflect
16 the application of city-specific effect estimates, because we do not have estimates for
17 all 12 urban study areas, we do not provide comprehensive coverage for
18 heterogeneity in modeling the respiratory morbidity endpoint category. Long-term
19 Os-attributable mortality has been shown to demonstrate regional heterogeneity.
20 Specifically, Jerrett et al., 2009 presented regional effect estimates that demonstrated
21 considerable heterogeneity ranging from essentially a no-effect (for the Northeast and
22 Industrial Midwest) to effects substantially larger than the national effect (Southeast,
23 Southwest and Upper Midwest) (see Table 4 in Jerrett et al., 2009). There are many
24 potential explanations for regional heterogeneity including differences in 63-
25 attributable factors and potential confounding, potential for the presence of (and
26 regional differences in) averting behavior, and variation in sample sizes which can
27 impact stability of effect estimates. For the core analysis, we use a national effect
28 estimate in modeling long-term exposure related mortality. Consideration of regional
29 effect estimates are included as a sensitivity analysis (see section 7.4.3 and 7.5.3).
30 • Exposure Measurement Error Associated with Oi Effect Estimates: Exposure
31 measurement error refers to uncertainty associated with using ambient monitor based
32 exposure surrogate metrics to represent the actual exposure of an individual or
33 population. As such, this factor can be an important contributor to variability in
34 epidemiological study results across locations, and uncertainty in results for any
27 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
-------
1 specific city (O^ ISA, p. Ixii). Exposure measurement error can result from a number
2 of factors (e.g., central site monitors not representing actual patterns of personal
3 exposure including activity patterns, presence of non-ambient sources of exposure for
4 the pollutant of interest) (O^ ISA, Ixii). These factors can vary across urban study
5 areas (and even within urban study areas), thereby contributing to differences in the
6 nature and magnitude of exposure measurement error across locations and ultimately
7 to differences in effect estimates and associated confidence levels. Exposure
8 measurement error is related to heterogeneity in effect estimates, since regional
9 differences in effect estimates can result in part, from differences in exposure
10 measurement error as noted here.
11 • Intra-urban Variability in Ambient Oi Levels: The picture with regard to within
12 city variability in ambient O?, levels and the potential impact on epidemiologic-based
13 effect estimates is somewhat more complicated. The 63 ISA notes that spatial
14 variability in Os levels is dependent on spatial scale with Os levels being more
15 homogeneous over a few kilometers due to the secondary formation nature of Os,
16 while levels can vary substantially over tens of kilometers. Community exposure may
17 not be well represented when monitors cover large areas with several subcommunities
18 having different sources and topographies as exemplified by Los Angeles which
19 displays significantly greater variation in inter-monitor correlations than does, for
20 example, Atlanta or Boston (see O3 ISA section 4.6.2.1 U.S. EPA 2013a). Despite the
21 potential for substantial variability across monitors the 63 ISA notes that studies have
22 tended to demonstrate that monitor selection has only a limited effect on the
23 association of short-term Os exposure with health effects. The likely explanation for
24 this is that, while absolute values for a fixed point in time can vary across monitors in
25 an urban area, the temporal patterns of Os variability across those same monitors
26 tends to be well correlated. Given that most of the short-term Os-attributable Os
27 epidemiological studies are time series in nature, the 63 ISA notes that the stability of
28 temporal profiles across monitors within most urban areas means that monitor
29 selection will have little effect on the outcomes of an epidemiological study
30 examining short-term Os-attributable mortality or morbidity (see 63 ISA section
31 4.6.2.1 U.S. EPA 2013a). For this reason, we conclude that generally intra-city
32 heterogeneity in 63 levels is not a significant factor likely to impact estimates of
33 short-term Os-attributable risk. One exception is LA which, due to its size and
34 variation in O^ sources and other factors impacting O^ patterns such as topography,
35 displays significant variation in ambient 63 levels with a subsequent impact on risk.
36 However, in the case of LA (as with the other urban study areas), we model risk using
7-37
-------
1 composite monitors which do not provide spatially-differentiated representations of
2 exposure and consequently, we do not address this source of variability in the risk
3 assessment. As discussed in the uncertainty section, short-term exposure mortality
4 effect estimates for the New York CBSA (Smith et al., 2009) display significant
5 variability. However, it is not clear which factors are primarily responsible for this
6 heterogeneity (e.g., differences in the urban structure, residential behavior, or ambient
7 Os levels within the CBSA). The potential for intra-city heterogeneity in Os levels to
8 affect risk is more pronounced with long-term Os-attributable mortality where the
9 relationship between annual trends in ambient Os (as represented using composite
10 monitor values) and annual mortality is compared between urban study areas in order
11 to derive effect estimates. Here, pronounced heterogeneity in Os levels within a given
12 city can result in exposure misclassification, if that heterogeneity is not well
13 represented by the composite monitor for that city. Different degrees of exposure
14 misclassification across urban study areas can introduce uncertainty into the overall
15 national-level effect estimate for long-term exposure-related mortality. Furthermore,
16 if that exposure measurement error has a regional trend, then measurement error can
17 potentially result in apparent regional heterogeneity in the effect estimates. The
18 degree to which there is true regional heterogeneity is made uncertain by the presence
19 of differential measurement error across regions.
20 • Variability in the Patterns of Ambient Os Reduction Across Urban Areas: The
21 simulated patterns of ambient Os concentrations across an urban area can vary based
22 on the methodology used to adjust ambient Os concentrations to represent just
23 meeting the current or alternative suites of standards. For the 1st draft REA, we used
24 a statistical approach called the "quadratic rollback" method for simulating just
25 meeting the current Os standard. Although the quadratic rollback method replicates
26 historical patterns of air quality changes better than some alternative methods, its
27 implementation relies on a statistical relationship instead of on a mechanistic
28 characterization of physical and chemical processes in the atmosphere. Because of its
29 construct as a statistical fit to measured Os values, the quadratic rollback technique
30 cannot capture spatial and temporal heterogeneity in Os response and also cannot
31 account for nonlinear atmospheric chemistry that causes increases in Os as a result of
32 emissions reductions of certain Os precursors under some circumstances. As noted in
33 section 7.1.1, for this draft of the REA, we have employed a model-based Os
34 adjustment methodology in the risk assessment for simulating Os concentrations
35 under current and alternate standard levels. Use of this model-based approach allows
36 the risk assessment results to more fully account for non-linearities in Os formation
7-38
-------
1 and to reflect spatial and temporal heterogeneity in 03 response, including NOx
2 titration conditions under which a reduction in NOx causes an increase in 63
3 concentrations, in some core urban locations.
4 • Demographics and Socioeconomic-status (SES)-related Factors: Variability in
5 population density, particularly in relation to elevated levels of 63 has the potential to
6 influence population risk, although the significance of this factor also depends on the
7 degree of intra-urban variation in 03 levels (as discussed above). In addition,
8 community characteristics such as pre-existing health status, ethnic composition, SES
9 and the age of housing stock (which can influence rates of air conditioner use thereby
10 impacting rates of infiltration of 63 indoors) can contribute to observed differences in
11 O3-related risk (discussed in O3 ISA - section 2.5.4.5, U.S. EPA, 2013a). Some of the
12 heterogeneity observed in effect estimates between cities in the multicity studies may
13 be due to these community characteristics, and while we cannot determine how much
14 of that heterogeneity is attributable to these factors, the degree of variability in effect
15 estimates between cities in our analysis should help to capture some of the latent
16 variability in these community characteristics.
17 • Baseline Incidence of Disease: We collected baseline health effects incidence data
18 (for mortality and morbidity endpoints) from a number of different sources (see
19 section 7.3.4). Often the data were available at the county-level, providing a relatively
20 high degree of spatial refinement in characterizing baseline incidence given the
21 overall level of spatial refinement reflected in the risk assessment as a whole.
22 Otherwise, for urban study areas without county-level data, either (a) a surrogate
23 urban study area (with its baseline incidence rates) was used, or (b) less refined state-
24 level or national incidence rate data were used.
25 7.4.2 Qualitative Assessment of Uncertainty
26 As noted in section 7.4, we have based the design of the uncertainty analysis carried out
27 for this risk assessment on the framework outlined in the WHO guidance document (WHO,
28 2008). That guidance calls for the completion of a Tier 1 qualitative uncertainty analysis,
29 provided the initial Tier 0 screening analysis suggests there is concern that uncertainty associated
30 with the analysis is sufficient to significantly impact risk results (i.e., to potentially affect
31 decision making based on those risk results). Given previous sensitivity analyses completed for
32 prior O^ NAAQS reviews, which have shown various sources of uncertainty to have a potentially
33 significant impact on risk results, we believe that there is justification for conducting a Tier 1
34 analysis.
7-39
-------
1 For the qualitative uncertainty analysis, we have described each key source of uncertainty
2 and qualitatively assessed its potential impact (including both the magnitude and direction of the
3 impact) on risk results, as specified in the WHO guidance. Similar to our discussion of
4 variability in the last section, the term key sources of uncertainty refers to those sources that the
5 we believe have the potential to play an important role in impacting population incidence
6 estimates generated for this risk assessment (i.e., these sources of uncertainty, if fully addressed
7 could result in adjustments to the core risk estimates which might impact the interpretation of
8 those risk estimates in the context of the Os NAAQS review). These key sources of uncertainty
9 have been identified through consideration for sensitivity analyses conducted for previous Os
10 NAAQS risk assessments, together with information provided in the final Os ISA and comments
11 provided by CASAC on the analytical plan for the risk assessment.
12 Table 7-4 includes the key sources of uncertainty identified for the Os REA. For each
13 source of uncertainty, we have (a) provided a description, (b) estimated the direction of influence
14 (over, under, both, or unknown) and magnitude (low, medium, high) of the potential impact of
15 each source of uncertainty on the risk estimates, (c) assessed the degree of uncertainty (low,
16 medium, or high) associated with the knowledge-base (i.e., assessed how well we understand
17 each source of uncertainty), and (d) provided comments further clarifying the qualitative
18 assessment presented.
19 The categories used in describing the potential magnitude of impact for specific sources
20 of uncertainty on risk estimates (i.e., low, medium, or high) reflect our consensus on the degree
21 to which a particular source could produce a sufficient impact on risk estimates to influence the
r\Q
22 interpretation of those estimates in the context of the Os NAAQS review. Sources classified as
23 having a "low" impact would not be expected to impact the interpretation of risk estimates in the
24 context of the 63 NAAQS review; sources classified as having a "medium" impact have the
25 potential to change the interpretation: and sources classified as "high" are likely to influence the
26 interpretation of risk in the context of the Os NAAQS review. Because this classification of the
27 potential magnitude of impact of sources of uncertainty is not based on our direct quantitative
28 assessments, we use qualitative judgments, in some cases informed by other relevant quantitative
29 analyses. Therefore, the results of the qualitative analysis of uncertainty are not useful for
30 making quantitative estimates of confidence, e.g. probabilistic statements about risk. However,
31 they can be used to support the interpretation of the risk estimates, including the assessment of
32 overall confidence in the risk estimates. In addition, they can also be used in guiding future
33 research to reduce uncertainty related to O^ risk assessment. As with the qualitative discussion of
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
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?
7-40
-------
1 sources of variability included in the last section, the characterization and relative ranking of
2 sources of uncertainty addressed here is based on our consideration of information provided in
3 previous Os NAAQS risk assessments (particularly past sensitivity analyses), the results of risk
4 modeling completed for the current O^ NAAQS risk assessment and information provided in the
5 third draft 63 ISA as well as earlier 63 Criteria Documents. Where appropriate, in Table 7-4, we
6 have included references to specific sources of information considered in arriving at a ranking
7 and classification for a particular source of uncertainty.
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 Os to simulate
attainment of both
existing and alternative
standard levels
See Chapter 4 for details
Both
Low-
Medium
Low-medium
See Chapter 4 for more details (specific call-outs to be added)
B. UseofCBSA-based
study areas in modeling
risk (i.e., potential
mismatch between study
areas used in the REA
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 current draft of the REA, 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 REA by focusing
risk estimates on that subpopulation living in areas likely to experience
potential increases in Os (and excluding the larger population of urban
and suburban areas likely to experience reductions in ambient QT,
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 REA 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 REA 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 Oj
monitors in an urban
area (as reflected in the
The effect estimates used in this risk
assessment reflect a specific spatial
and temporal pattern of ambient Oj
(as represented by the particular
monitoring network providing data
for the underlying epidemiological
study). However, if the spatial and
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 QT, 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
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)
epidemiological study
providing the effect
estimates) to a simulated
change in the patterns of
those correlations when
we estimate risk in the
REA.
temporal pattern of Oj 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.
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 QT, concentrations.
D. Characterizing intra-
urban population
exposure in the context
of epidemiology studies
linking Qy to specific
health effects
Exposure misclassification within
communities that is associated with
the use of generalized population
monitors (which may miss important
patterns of exposure within urban
study areas) introduces uncertainty
into the effect estimates obtained
from epidemiology studies.
Under
(generally)
Low-
medium
Medium
KB and INF: Despite the potential for substantial variability in Os
levels across monitors (particularly in larger urban areas with greater
variation in sources and topography such as L.A.), the Oj 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 (see 63 ISA section 4.6.2.1, US EPA, 2013a). However,
this issue could be more of a concern in larger urban areas which may
exhibit greater variation in Oj, levels due to diverse sources,
topography and patterns of commuting.
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 effect
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 Os 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 Os-attributable studies (Os ISA, p. Ixii). 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 (Oj, ISA, p.
Ixix). In addition, exposure measurement error can vary across
different populations even within the same urban study area. For
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)
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 Qy. In this example,
an effect estimate derived for that specific population based on Oj
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.
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 QT, NAAQS
regulatory 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 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.
G. Addressing co-
pollutants
The inclusion or exclusion of co-
pollutants which may confound, or
in other ways, affect the QT, effect,
introduces uncertainty into the
analysis.
Both
Low-
medium
Medium
KB and INF: The Os ISA notes that across studies, the potential
impact of PM indices on (^-mortality risk estimates tended to be
much smaller than the variation in Os-mortality risk estimates across
cities. This suggests that Oj effects are independent of the relationship
between Os and mortality. However, interpretation of the potential
confounding effects of PM on (^-mortality risk estimates requires
caution. This is because the PM-Os correlation varies across regions,
due to the difference in PM components, complicating the
interpretation of the combined effect of PM on the relationship between
Os 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 (O3ISA section
2.5.4.5, US EPA, 2013a).
H. Specifying lag
structure (short-term
exposure studies)
There is uncertainty associated with
specifying the exact lag structure to
use in modeling short-term Oj-
Both
Low-
Medium
Low
KB and INF: The majority of studies examining different lag models
suggest that QT, effects on mortality occur within a few days of
7-44
-------
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)
attributable mortality and
respiratory-related morbidity.
exposure. Similar, studies examining the impact of Oj exposure on
respiratory-related morbidity endpoints suggests a rather immediate
response, within the first few days of ©3 exposure (see O3ISA section
2.5.4.3, US EPA, 2013a). 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 ©3 exposure 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)
7-45
-------
1 7.4.3 Description of Core and Sensitivity Analyses
2 As discussed in section 7.1.1, this risk assessment includes a set of core (higher
3 confidence) risk estimates which are supplemented by sensitivity analyses. The sensitivity
4 analyses explore the potential impact that variation in specific model design elements can have
5 on the core risk estimates. This section specifies which design elements are included in both the
6 core and sensitivity analyses completed for each of the health effect endpoint categories included
7 in the risk assessment. We divided the sensitivity analyses into two categories: (a) those
8 involving air quality characterization and (b) those associated directly with the specification of
9 the C-R functions used in estimating risk. We recognize that there can be overlap between these
10 categories with some modeling elements (e.g., modeling period) affecting both the composite
11 monitor distribution as well as representing an element of C-R function specification. However,
12 we have retained these two categories to aid in the presentation and discussion of sensitivity
9Q
13 analysis results. The sensitivity analyses also included an initial influence analysis designed to
14 evaluate which of the model inputs are primarily responsible for inter-city variability
15 (heterogeneity) in risk. The influence analysis uses estimatedelasticities of risk with respect to
16 the risk function input variables, focusing on the short-term exposure-related mortality endpoint
17 and associated input parameters since this is one of the key risk estimates generated for the REA
18 (additional detail on how the influence analysis was conducted is presented in section 7.5.3).
19 Table 7-5 presents the alternative approaches for adjusting the O^ distributions used in
20 the sensitivity analysis and also identifies the approaches used in the core analysis for each of the
21 study areas. The alternative air quality adjustment approaches examine the differences in
22 changes in air quality and risk when applying NOx-only versus NOx and VOC reductions in the
23 HDDM-adjustment approach. It should be noted that when NOx and VOC reductions were used
24 in the HDDM-adjustment approach in this sensitivity analysis, the same percent reduction for
25 both pollutants was used in the air quality adjustment for meeting the existing and alternative
26 standard in each urban area. More details on these alternative air quality adjustment approaches
27 are discussed in Chapter 4 and appendices.
28 Besides the approach used to adjust the distributions of Os, another fact which has a
29 direct impact on composite monitor composition is the specification of the study area (since this
30 determines the mix of monitors that will be included in constructing the composite monitor). As
31 discussed in section 7.1.1, for the core analysis, we modeled all endpoints (for all study areas)
29 As noted in 7.1.1, in presenting both the core and sensitivity analyses, we include both point estimates and 95th
CIs, 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.
7-46
-------
1 using CBSA-based study areas. For the sensitivity analysis (for the short-term Os-attributable
2 mortality endpoint), we included a smaller study area based on the original study area definition
3 used in the Smith et al., 2009 study.30
4 Table 7-6 presents the model elements included in sensitivity analyses exploring
5 alternative C-R function specifications. These sensitivity analyses were applied both to short-
6 term Os-attributable mortality and long-term Os-attributable mortality. As discussed in section
7 7.1.1, we were not able to differentiate between alternative C-R function specifications for short-
8 term (Vattributable morbidity endpoints and therefore included the full set of alternative C-R
9 function specifications in the core analysis. This results in a distribution of core risk estimates for
10 each endpoint which can be used to gain insights into the impact of different C-R function
11 specifications on risk. Because separate sensitivity analyses were not completed for short-term
12 Os-attributable morbidity endpoints, this category is not included in Table 7-6.
13
14 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, MI
Houston, TX
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
Core simulation
(type of precursor reduced to
adjust Os 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
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
30 We did not include an alternative study area simulation as a sensitivity analysis for long-term exposure related
mortality, since 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 O3-attributable mortality) is likely to introduce less uncertainty.
7-47
-------
1
2
3
4
5
Study Area
St. Louis, MO
Core simulation
(type of precursor reduced to
adjust C>3 distribution)
NOx-only
Sensitivity analysis
Alternative modeling
approach not evaluated
A lower-bound fit of the HDDM-based Oj sensitivities (reflecting a greater increment of Oj 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
Short-term 63-
attributable
mortality
Long-term Os-
attributable
mortality
Modeling elements included
Core analysis
- Full monitoring period
(specific to each study
area), 8hr max metric,
national-Bayes adjusted,
single pollutant model.
- effect estimates obtained
from: Smith et al., 2009
study
- Single national estimate,
two -pollutant model
(PM2.5), long-term peak
trend metric (based on
daily Ihr max values),
CBSA-based study area.
- effect obtained from
Jerrett et al., 2009 study
Sensitivity analysis
- summer (warm month),
8hr mean, regional-bayes
adjusted, multi-pollutant
(withPM10).
- effect estimates obtained
from Zanobetti and
Schwartz, 2008 and Smith
et al., 2009
- Regional-differentiated
effect estimates, single
pollutant model.
- National-level effect
estimate, single pollutant
model.
- effect estimates also
obtained from Jerrett et al.,
2009 study
7 7.5 URBAN STUDY AREA RESULTS
8 This section discusses risk estimates generated for the set of 12 urban study areas,
9 including both the core risk estimates and accompanying sensitivity analyses. In summarizing
10 risk estimates, this discussion focuses on results most relevant to two policy-related questions:
11 (a) to what extent is the existing O?, standard protective of public health , and, (b) what is the
12 nature and magnitude of additional public health protection provided by the suite of alternative
13 standards under consideration? Consequently, we focus on two types of risk estimates including
14 the magnitude of Os-attributable risk after simulation of just meeting the existing standard and
7-48
-------
1 the degree of risk reduction potentially provided by each of the alternative standards relative to
2 just meeting the existing standard.31
3 This section is organized as follows. We begin by presenting the core risk estimates in
4 both tabular and graphical format at the end of this section. We then present key observations
5 about the risk estimates for just meeting the existing standard (for core risk) in section 7.5.1. Key
6 observations related to risk estimates for just meeting alternative standard levels, and for
7 estimates of risk changes comparing alternative standards to just meeting the existing standard
8 (again, for core risk) are presented in section 7.5.2. After presenting key observations related to
9 the core risk estimates, we then present key observations resulting from the sensitivity analyses
10 (section 7.5.3).
11 A number of details regarding the design of the core risk assessment should be kept in
12 mind when reviewing the core risk estimates presented in this section (see section 7.1.1 for
13 additional detail on these design elements):
14 • All risk estimates reflect application of a CBSA-based study area.
15 • Estimates are presented for two simulation years (2007 and 2009):
16 • Short-term Os-attributable mortality estimates are generated for all 12 urban study
17 areas, while most short-term Os-attributable morbidity estimates (depending on the
18 specific health endpoint) are generated for only a subset of urban study areas. Long-
19 term Os-attributable mortality is modeled for all 12 urban study areas.
20 • For all health effect endpoints, we model risk down to zero 63 and do not include
21 either consideration for LML or alternative threshold levels.
22 There are several categories of risk metrics generated for the core mortality and
23 morbidity endpoints modeled in this analysis. Below we describe both the types of risk metrics
24 generated for the core analysis and the specific types of tables and figures used in presenting
25 those metrics.
26
27 I. Core short-term Ch-attributable mortality estimates
28 • Table presenting estimates of Os-attributable mortality incidence after just
29 meeting the existing standard and the estimated change in incidence associated
30 with meeting each of the alternative standard levels relative to the existing
31 standard (Table 7-7): These estimates include point estimates and 95th percentile
32 confidence intervals representing uncertainty associated with the statistical fit of the
33 effect estimates.
31 As part of this draft of the risk assessment, 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 the
simulated attainment of the existing standard. See Appendix 7B.
7-49
-------
1 • Table presenting estimates of the percent of total mortality attributable to Os
2 after just meeting the existing standard and the percent reduction in 63-
3 attributable risk associated with each alternative standard (Table 7-8).
4 • Heat maps for mortality illustrating distribution across daily Os levels of total
5 Os-attributable risk after just meeting the existing standard and risk reductions
6 after meeting alternative standards (Figures 7-2 and 7-3): Heat maps are provided
7 for each of the 12 urban areas. The color gradient in each figure reflects the
8 distribution of mortality (or the change in mortality) across the range of daily 8-hour
9 Os levels and provides a visual tool to explore trends in mortality counts across daily
10 Os levels and between cities. Visual patterns in the figures presenting total risk and
11 risk reduction are interpreted differently:
12 o For figures depicting total Os-attributable risk, colors range from blue (lower
13 mortality count) to red (higher mortality count). As an example, with Figure 7-2,
14 top heat map (which presents total Os-attributable risk for the existing standard in
15 2007, based on Smith et al., 2009 C-R functions), if we focus on the first row
16 (Atlanta, GA), we see a value of 38 under the column 55-60 ppb. This value
17 reflects the fact that 38 of the 270 Os-attributable deaths estimated for Atlanta
18 after just meeting the existing 75 ppb standard in 2007 occurred on days with
19 composite monitor Os levels between 55 and 60 ppb. Similarly, in the same row,
20 we see that only 3 Os attributable deaths occurred on days when the composite
21 monitor value was between 20 and 25 ppb. We also include the total Os-
22 attributable incidence (for each study area) in the final column marked "Total".
23 o For figures depicting changes in risk associated with simulation of existing and
24 alternative standard levels, we see that the pattern is more complex since we can
25 have a combination of increases and decreases in risk in the heat maps, with
26 increases in risk identified as red to yellow and decreases in risk identified as
27 yellow to blue. Increases in risk are negative numbers, decreases are positive. In
28 addition, in the final three columns of each map, we provide estimates of the total
29 Os-attributable incidence, as well as the total broken down into the subtotals
30 across days with increases (negative) and days with decreases (positive) in that
31 incidence. The increase and decrease for a given study area should sum
32 (accounting for rounding in these subtotals) to the overall total for Os-attributable
33 deaths for that study area. Several factors can contribute to the patterns of changes
34 in Os-attributable risk reflected in these maps. For example, non-linearities in Os
7-50
-------
1 formation can result in increases in O?, on some days, even when simulating
2 attainment of a lower alternative standard (see section 7.1.1). In addition,
3 simulation of alternative standard levels can result in a change in the overall
4 distribution of the composite monitor ambient O?, distribution. Often, this change
5 will take the form of a shift in the upper tail of the distribution towards the mean,
6 given that simulated attainment of alternative standard levels targets higher Os
7 days. If we look at figure 7-2 at the second map (Decrease 75 to 70) and
8 specifically at the row for Houston, we see that there is a -4 increase in deaths
9 distributed across 20-35 ppb days and a decrease in deaths of 9, primarily
10 distributed across 40-60 ppb days.
11 • Graphic plots of Os-attributable deaths per 100,000 population for just meeting
12 the existing and alternative standards (Figures 7-4): OsThis plot provides
13 estimates that are adjusted for the size of the underlying urban population, thereby
14 allowing the mortality estimates and associated trends to be more readily compared
15 across urban study areas (consideration of absolute O^ mortality is complicated by the
16 role that underlying urban population plays in driving total (Vattributable mortality -
17 larger study areas like Los Angeles and New York having substantially larger
18 mortality estimates primarily due to their higher underlying populations). These
19 figures allow us to evaluate the overall magnitude of risk reductions across standard
20 levels and determine the degree to which those trends differ for different study areas.
21 Tables summarizing incidence, percent of baseline incidence and percent reduction in 63-
22 attributable risk for short-term Os-attributable morbidity (Tables 7-9 through 7-11): Three
23 categories of short-term Os-attributable mortality effects were modeled for the analysis
24 (respiratory related HA, respiratory-related ER visits and asthma exacerbations). As discussed in
25 section 7.1.1, these morbidity effects were modeled for a combination of all 12 urban study areas
26 and a subset of those study areas depending on the endpoint (see below). The C-R functions
27 available for modeling many of these morbidity endpoints included consideration for a number
28 of design elements (e.g., copollutants and lag structure). However, as noted earlier in section
29 7.1.1, for short-term exposure morbidity endpoints with multiple C-R functions, we were not
30 able to differentiate between C-R functions in terms of overall confidence and consequently we
31 could not identify a single core model. Therefore, when we have multiple C-R functions
32 reflecting different treatments of key design elements such as lag structure, we consider the risk
33 estimates that result from the full set of C-R functions to represent a core range of risk. Each of
34 the tables summarizing short-term Os-attributable morbidity risk present several risk metrics
7-51
-------
1 including: (a) total Os-attributable incidence (after just meeting the existing standard), (b)
2 reductions in (Vattributable incidence (for each of the alternative standard levels relative to just
3 meeting the existing standard), (c) percent of baseline incidence attributable to Os (after just
4 meeting the existing standard) and (d) percent reductions in Os-attributable risk (for each of the
5 alternative standard levels). In presenting these morbidity risk estimates, we do not include 95th
6 percentile confidence intervals in order to conserve space. Specific tables summarizing these
7 morbidity incidence estimates include:
8 o HA visits (for respiratory symptoms including asthma): Table 7-9 presents
9 estimates of the incidence of HA (for respiratory symptoms, chronic lung disease
10 and asthma). Risk estimates are generated for a subset of the urban study areas for
11 some of the health endpoints (e.g., New York City for HA [chronic lung disease
12 and asthma]), while HA (respiratory-related) estimates cover all 12 urban study
13 areas.
14 o ER visits (for respiratory symptoms including asthma): Table 7-10 presents
15 estimates of the incidence of ER visits for respiratory symptoms and asthma)
16 specifically for New York City and Atlanta based on C-R functions obtained from
17 several epidemiological studies.
18 o Asthma exacerbations: Table 7-11 presents estimates of the incidence of asthma
19 exacerbations (including estimates for a range of symptoms) for Boston, the only
20 urban study area with C-R functions supporting modeling for this endpoint.
21 • Graphic plots of Cb-attributable respiratory-related HA per 100,000 residents
22 for the existing and alternative standard levels (Figures 7-5): This figure is
23 intended to complement Figure 7-4 which presents the same type of risk information
24 for short-term (Vattributable mortality. By plotting respiratory HA per 100,000, we
25 adjust for the underlying population which makes trends in risk more comparable
26 across urban study areas. We have only created this graphic for respiratory HA (based
27 on application of Medina-Ramon et al., 2006) since that is the only morbidity
28 endpoint modeled for all 12 urban study areas. As with the mortality figure, this
29 figure allows us to evaluate the overall magnitude of risk reductions across standard
30 levels and determine the degree to which those trends differ for different study areas.
31 III. Core long-term O^-attributable mortality estimates
32
7-52
-------
1 • Table presenting estimates of long-term Os-attributable mortality incidence
2 including total risk after just meeting the existing standard and risk reductions
3 based on comparing risks after meeting alternative standards to risks after
4 meeting the existing standard (Table 7-12): Estimates presented in Table 7-12
5 reflect respiratory mortality and include 95th percentile confidence intervals
6 representing uncertainty associated with the statistical fit of the effect estimates used.
7 Estimates presented in these tables allow for consideration for the magnitude of risk
8 associated with just meeting the existing standard and the pattern of risk reduction (in
9 incidence) in meeting alternative standards relative to the existing standard.
10 • Table presenting estimates of the percent of respiratory mortality attributable to
11 63 and percent reductions in (^-attributable risk for long-term (^-attributable
12 mortality (Table 7-13).
13 • Graphic plots of Os-attributable deaths per 100,000 population for just meeting
14 the existing and alternative standards (Figures 7-6): This plot provides estimates
15 that are adjusted for the size of the underlying urban population, thereby allowing the
16 mortality estimates and associated trends to be more readily compared across urban
17 study areas (consideration of absolute 63 mortality is complicated by the role that
18 underlying urban population plays in driving total Os-attributable mortality - larger
19 study areas like Los Angeles and New York having substantially larger mortality
20 estimates primarily due to their higher underlying populations). These figures allow
21 us to evaluate the overall magnitude of risk reductions across standard levels and
22 determine the degree to which those trends differ for different study areas.
7-53
-------
1
2
Table 7-7 Short-Term o3-attributable All Cause Mortality Incidence (2007 and 2009)
(Smith et al., 2009 C-R Functions) (O3 season, CBSA-based study area, no threshold)
4
5
Study Area
Air Qualtiy Scenario
Absolute Incidence
75ppb
Change in Incidence
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
270
(-370-890)
440
(-250-1100)
350
(-500-1200)
430
(-41-890)
86
(-280-440)
660
(32-1300)
680
(130-1200)
1300
(-530 - 3000)
2800
(1700-3900)
1200
(270-2200)
370
(-390 - 1100)
430
(-110-950)
12
(-16-39)
13
(-7-33)
7
(-10-24)
14
(-1-28)
2
(-6-10)
23
(1-44)
5
(1-9)
43
(-18-100)
130
(80-190)
35
(8-62)
7
(-7-20)
18
(-5-41)
21
(-30-72)
27
(-15-68)
20
(-28-67)
32
(-3-67)
4
(-14-23)
42
(2-81)
11
(2-20)
87
(-36-210)
640
(380-890)
76
(17-140)
13
(-13-39)
39
(-10-86)
34
(-47-110)
45
(-25-110)
32
(-45-110)
64
(-6-130)
8
(-25-40)
69
(3-130)
24
(4-43)
160
(-66-380)
NA
NA
120
(25-210)
23
(-24-70)
60
(-15-130)
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
240
(-340 - 800)
400
(-220 - 1000)
320
(-450 - 1100)
400
(-38-830)
83
(-270-420)
580
(28-1100)
700
(130-1200)
1300
(-540-3100)
2600
(1600-3700)
1100
(240-2000)
370
(-390-1100)
380
(-96-840)
9
(-12-28)
7
(-4-19)
-1
(2 --4)
12
(-1-24)
0
(-1-2)
-21
(-1--41)
-1
(0--1)
41
(-17-100)
84
(50-120)
19
(4-34)
6
(-7-19)
8
(-2-18)
16
(-22-54)
17
(-10-44)
5
(-7-17)
29
(-3-60)
2
(-6-10)
-6
(0--12)
4
(1-7)
89
(-37-210)
440
(260-610)
44
(10-78)
12
(-13-38)
21
(-5-46)
23
(-32-77)
28
(-15-71)
14
(-19-47)
49
(-5-100)
7
(-23-37)
15
(1-30)
14
(3-26)
160
(-68-390)
NA
NA
69
(15-120)
21
(-22-64)
37
(-9-83)
NA: for NYC, the model-based adjustment methodology was unable to adjust QT, distributions such that they would meet the
lower alternative standard level of 60 ppb.
7-54
-------
1
2
3
4
5
6
7
Table 7-8 Percent of Total All-Cause Mortality Attributable to O3 and Percent Change in
o3-Attributable Risk (2007 and 2009) (Smith et al., 2009 C-R functions) (O3 season,
CBSA-based study area, no threshold)
9
10
11
12
13
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
12
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
3.9
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 NYC, the model-based adjustment methodology was unable to adjust
lower alternative standard level of 60 ppb.
distributions such that they would meet the
7-55
-------
1
2
Figure 7-2 Heat Maps for Short Term o3-attributable Mortality (Just meeting existing
standard and risk reductions from just meeting alternative standards) (2007) (Smith
et al., 2009 C-R functions) (see Key at bottom of figure)
Current Standard (75)
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
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
1
0
1
0
0
0
0
0
2
0
1
20-25
3
1
4
5
0
2
17
0
21
0
1
3
25-30
5
11
20
14
0
7
49
0
98
34
18
7
30-35
18
22
45
40
1
42
126
0
297
62
53
18
35-40
24
43
50
65
5
72
146
17
544
156
98
65
40-45
41
84
58
89
6
123
148
340
741
213
67
66
45-50
52
71
57
81
13
147
95
445
475
236
65
76
50-55
63
69
35
43
17
75
50
388
364
209
40
74
55-60
73
20
40
23
52
49
44
233
165
20
47
60-65
44
30
31
15
56
3
13
39
101
5
29
65-70
12
9
12
4
20
0
5
0
42
2
29
70-75
9
13
10
2
43
0
0
0
9
0
12
>75
0
3
11
0
0
17
0
0
0
10
0
3
Total
267
443
353
431
86
655
683
1,253
2,812
1,238
367
430
Decrease 75 to 70
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
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
-1
0
-1
0
0
0
25-30
0
0
0
0
0
0
-1
0
-2
-1
-1
0
30-35
0
0
0
0
0
-1
-2
0
0
0
-1
0
35-40
1
0
0
0
0
0
0
0
12
0
1
1
40-45
1
1
1
2
0
2
2
6
27
3
2
2
45-50
2
2
1
3
0
5
2
16
32
7
2
3
50-55
3
3
1
2
0
3
2
17
35
8
2
4
55-60
2
3
1
2
1
3
2
2
25
9
1
3
60-65
1
2
1
2
1
4
0
1
5
6
0
2
65-70
0
1
0
1
0
2
0
0
0
3
0
2
70-75
0
1
1
1
0
3
0
0
0
1
0
1
>75
0
0
1
0
0
1
0
0
0
1
0
0
Total
12
13
7
14
2
23
5
43
134
35
7
18
Change in risk
Inc.
0
0
0
-1
0
Dec.
12
14
8
15
2
24
9
43
146
38
8
19
Decrease 75 to 65
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
Daily Shr 1
0-5
0
0
0
0
0
0
0
0
0
0
0
0
.flax Ozone
5-10
0
0
0
0
0
0
0
0
0
0
0
0
Level (pp
10-15
0
0
0
0
0
0
0
0
0
0
0
0
l)
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
-1
0
0
0
-3
0
2
-1
-1
0
30-35
1
0
0
0
0
-
0
2
0
35-40
1
0
1
2
0
1
0
0
85
0
2
3
40-45
3
9
45-50
4
4
3
7
0
8
5
33
136
15
4
7
50-55
6
5
3
5
1
6
4
35
136
17
3
8
55-60
4
7
2
5
2
6
5
4
90
18
2
6
60-65
2
5
3
4
1
7
0
1
19
12
0
4
65-70
1
1
1
2
0
3
0
1
0
6
0
4
70-75
1
1
2
2
0
6
0
0
0
1
0
2
>75
0
0
2
0
0
3
0
0
0
2
0
1
Total
21
27
20
32
4
42
11
87
640
76
13
39
Change
Inc.
0
-2
-2
-3
-1
-3
-9
0
-6
-5
-3
-1
in risk
Dec.
22
28
22
34
5
45
20
87
646
81
15
39
Decrease 75 to 60
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
Daily Shr 1
0-5
0
0
0
0
0
0
0
0
0
0
0
.flax Ozone
5-10
0
0
0
0
0
0
0
0
0
0
0
Level (pp
10-15
0
0
0
0
0
0
0
0
0
0
0
1)
15-20
0
0
0
0
0
0
0
0
0
0
0
20-25
0
0
0
-1
0
0
-3
0
0
0
0
25-30
0
0
-1
0
0
0
-5
0
-2
-2
0
30-35
1
0
1
0
0
-1
-4
0
-1
-2
0
35-40
2
1
2
5
0
2
2
2
1
4
5
40-45
4
5
4
12
0
8
8
41
11
6
7
45-50
N
50-55
9
9
5
10
2
10
7
48
26
5
12
55-60
6
11
3
10
3
9
8
6
27
3
9
60-65
3
8
5
8
2
11
1
2
17
1
6
65-70
1
2
2
4
1
5
0
1
8
0
7
70-75
1
2
3
3
0
9
0
0
2
0
3
>75
0
1
2
0
0
4
0
0
2
0
1
Total
34
45
32
64
8
69
24
159
116
23
60
Change
Inc.
0
-1
0
in risk
Dec.
34
47
34
67
8
73
37
159
123
27
61
4
5
6
7
8
9
10
11
12
NA: for NYC, the model-based adjustment methodology was unable to adjust
lower alternative standard level of 60 ppb.
distributions such that they would meet the
Key: For current standard (75) which is an absolute risk metric expressed as incidence of mortality, color gradient ranges from
blue (smallest O3-related mortality count) to red (highest O3-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
-------
1
2
Figure 7-3 Heat Maps for Short Term o3-attributable Mortality (Just meeting existing
standard and risk reductions from just meeting alternative standards) (2009) (Smith
et al., 2009 C-R functions) (see Key at bottom of figure)
Current Standard (75)
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
DailyShrl
0-5
0
0
0
0
0
0
0
0
0
0
0
0
Aax Ozone
5-10
0
0
0
0
0
0
0
0
0
0
0
0
Level (pp
10-15
1
0
1
0
0
1
0
0
0
0
0
1
l)
15-20
2
1
0
0
0
9
6
0
6
2
0
6
20-25
8
11
5
0
7
28
0
36
16
2
7
25-30
16
25
25
1
26
51
0
215
51
23
17
30-35
18
45
46
2
46
122
2
427
159
64
29
35-40
33
71
57
68
4
67
123
17
356
126
69
54
40-45
49
64
50
75
9
114
114
281
632
219
73
52
45-50
44
92
53
81
17
148
90
328
469
175
56
77
50-55
29
64
48
57
22
38
84
496
274
198
42
66
55-60
30
45
7
28
20
51
36
152
175
97
33
53
60-65
9
11
3
11
6
46
27
9
56
68
6
13
65-70
2
0
6
6
2
0
7
0
0
0
0
8
70-75
0
0
9
0
1
21
4
0
0
0
0
0
>75
0
0
3
0
0
6
4
0
0
0
0
0
Total
241
404
319
401
83
579
695
1,285
2,645
1,112
367
383
Decrease 75 to 70
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
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
0
0
-1
0
0
-2
-2
0
-3
-1
0
-1
25-30
0
0
-1
-1
0
-6
-3
0
-14
-2
-1
-1
30-35
0
-1
-1
0
0
-5
-3
0
-8
-2
-1
0
35-40
1
0
0
1
0
-5
-1
0
8
-1
1
1
40-45
2
1
0
2
0
-5
1
4
22
5
2
1
45-50
3
3
1
3
0
-2
2
11
31
5
2
3
50-55
2
2
1
3
0
0
3
20
22
8
2
3
55-60
2
2
0
2
1
2
2
6
19
5
2
3
60-65
1
1
0
1
0
2
2
0
6
4
0
1
65-70
0
0
0
0
0
0
0
0
0
0
0
1
70-75
0
0
0
0
0
2
0
0
0
0
0
0
>75
0
0
0
0
0
1
0
0
0
0
0
0
Total
8
7
-1
12
0
-21
-1
41
84
19
6
8
Change in risk
Inc.
-2
-2
-6
-2
-1
-29
-10
0
-38
-9
-2
-4
Dec.
10
9
5
14
1
8
10
41
121
28
8
12
Decrease 75 to 65
Study area
Atlanta, GA
Baltimore, MD
Boston, MA
Cleveland, OH
Denver, CO
Detroit, Ml
Los Angeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
Daily Shr T
0-5
0
0
0
0
0
0
0
0
0
0
0
Aax Ozone
5-10
0
0
0
0
0
0
0
0
0
0
0
Level (pp
10-15
0
0
0
0
0
-1
0
0
0
0
-1
1)
15-20
0
0
0
0
0
-4
0
-1
-1
0
-2
20-25
-1
0
-1
-1
0
-2
nt
-5
-2
0
-1
25-30
-1
-1
-1
-1
0
-7
-f-
-17
-4
-2
-1
30-35
0
-1
-1
0
0
-5
zfc
16
-4
-1
0
35-40
1
0
0
4
0
-4
0
52
-1
2
2
40-45
3
3
1
5
0
-2
10
107
10
3
3
45-50
5
6
2
8
0
3
23
120
11
4
6
50-55
3
5
3
7
1
2
42
81
17
3
6
55-60
4
3
0
60-65
2
1
0
2
1
5
1
21
8
1
2
65-70
0
0
0
1
0
0
0
0
0
0
1
70-75
0
0
1
0
0
3
0
0
0
0
0
>75
0
0
0
0
0
1
0
0
0
0
0
Total
16
17
5
29
2
-6
89
437
44
12
21
Change
Inc.
-3
-3
-6
-3
-1
-27
0
-43
-15
-4
-5
in risk
Dec.
19
21
11
32
3
21
89
479
59
16
26
Decrease 75 to 60
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
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
0
-1
0
0
0
0
-1
15-20
0
0
0
0
0
-4
-2
0
-1
0
-2
20-25
-1
-1
-2
-1
0
-2
-7
0
-3
-1
-1
25-30
-1
-1
-1
-1
0
-7
-7
0
-6
-3
-2
30-35
0
-1
-1
1
0
-5
-7
0
-4
-1
0
35-40
2
1
2
7
0
-2
0
2
0
3
3
40-45
5
5
2
9
0
2
5
32
16
6
5
45-50
6
10
5
13
1
10
8
44
NA
17
6
10
50-55
5
8
5
11
2
4
11
63
25
5
9
55-60
5
6
1
6
3
8
6
22
14
5
9
60-65
2
2
1
2
1
8
5
1
11
1
3
65-70
0
0
1
1
0
0
1
0
0
0
2
70-75
0
0
2
0
0
4
1
0
0
0
0
>75
0
0
1
0
0
2
1
0
0
0
0
Total
23
28
14
49
7
15
14
164
69
21
37
Change in risk
Inc.
-3
-4
-6
-3
-1
-26
-25
0
-20
-6
-6
Dec.
26
32
20
53
8
41
40
164
88
27
43
5
6
7
8
9
10
11
12
NA: for NYC, the model-based adjustment methodology was unable to adjust
lower alternative standard level of 60 ppb.
distributions such that they would meet the
Key: For current standard (75) which is an absolute risk metric, color gradient ranges from blue (smallest Os-related mortality
count) to red (highest O3-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).
7-57
-------
Figure 7-4 Plots of Short-Term o.,-attributable All-Cause Mortality for Meeting Existing
standard and Alternative Standards (Smith et al., 2009) (Simulation year 2007 and
2009)
2
2
.007 Simulation year
Total ozone- re bted mortality per 100.000 residents
ofo-t-cnrooMEcnroO
Trend in ozone-related mortality across standard
levels (deaths per 100,000)
—^Atlanta. GA
_ ^""^«*^ ^^^^^ ^"LJJIUII.UI L, IvlU
***"••. "*— — .^^^ -^-Boston. MA
1- ^Cleveland, OH
t • — ~~T •! 111171'^
^ 1 Houston. TX
*~ • Lo=Angele5,CA
T i - zzriTir,,.
Hi^SBcrarnento. CA
-*— St. Louis, MO
75ppb 70ppb 65ppb 60ppb
.009 Simulation year
Total ozone- re bted mortality per 100,000 residents
orojiCTicoow-ticriooo
Trend in ozone-related mortality across standard
levels (deaths per 100,000)
< Atlanta, GA
^*^Xs^ * Boston, MA
^ )t Cleveland, OH
^^^^fs^^^i^es—-— ^!rz'i°
^^^ Houston, TX
• • • • LosAngeles.CA
•1 Sacramento, CA
-*-St. Louis, MO
75ppb 70ppb 65ppb 60ppb
7-58
-------
Table 7-9 Short-Term -attributable Morbidity Incidence, Percent of Baseline and
Reduction in -attributable Risk - Respiratory-Related Hospital Admissions (2007
and 2009)
Endpoint/Study Area/Descriptor
Air Quality Scenario
Absolute
Incidence
75ppb
Change in Incidence
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 etal., 2009)
Ihr max, penalized splines
Ihr max, natural splines
230
230
13
12
23
22
37
36
HA (respiratory); NYC (Silverman and Ito, 2010; Lin et al., 2008)
HA Chronic Lung Disease (Lin)
HA Asthma (Silverman)
HA Asthma, PM2. 5 (Silverman)
120
420
310
6.7
28
20
29
120
84
NA
2.8
2.7
3.3
27.6
20.1
5
5
10
10
15
15
5
5
5
23
21
22
NA
HA (respiratory); LA (Linn etal., 2000)
|lhr max penalized splines
790
19
38
60
HA (COPD less asthma); all 12 study areas (Medina-Ramon, etal., 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
67
77
100
61
27
90
68
180
180
130
34
53
4
3
2
2
1
3
1
8
11
4
1
3
6
6
6
5
2
6
2
16
50
10
2
5
10
9
10
10
3
9
4
25
NA
15
4
8
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 etal., 2009)
Ihr max, penalized splines
Ihr max, natural splines
220
210
3.6
3.4
13
12
25
24
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)
120
410
310
respiratory); LA (Linn et al., 2000)
Ihr max penalized splines
640
5.1
24
17
21
96
68
NA
18
39
62
HA (COPD less asthma); all 12 study areas (Medina-Ramon, etal., 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
65
74
92
58
27
81
71
200
170
120
41
51
3
2
0
2
0
-3
1
8
7
2
1
2
5
4
1
5
1
-2
2
16
35
6
3
4
8
6
4
8
3
1
4
26
NA
9
4
6
2.5
2.4
3.2
27.2
19.8
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 NYC, the model-based adjustment methodology was unable to adjust QT, distributions such that they would meet the
lower alternative standard level of 60 ppb.
7-59
-------
Table 7-10 Short-Term -attributable Morbidity Incidence, Percent of Baseline and
Reduction in -attributable Risk - Emergency Room Visits (2007 and 2009)
Endpoint/Study Area/Descriptor
Air Quality Scenario
Absolute
Incidence
75ppb
Change in Incidence
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
7,500
4,500
410
230
740
420
1,200
670
ER-visits (respiratory); Atlanta (Tolbert et al., 2007, Darrow et al., 2011)
Tolbert
Tolbert-CO
Tolbert-NO2
Tolbert-PMlO
Tolbert-PMlO, NO2
Darrow
8,100
7,200
6,500
5,100
4,900
4,400
ER-visits (asthma); NYC (Ito et al, 2007)
single pollutant model
PM2.5
NO2
CO
SO2
9,000
7,100
5,800
9,500
7,300
360
320
290
230
220
190
670
590
530
420
400
360
1,100
940
840
660
640
560
530
410
330
570
420
2,300
1,800
1,500
2,500
1,900
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
16.0
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
6,800
4,000
310
170
570
320
800
460
ER-visits (respiratory); Atlanta (Tolbert et al., 2007, Darrow et al., 2011)
Tolbert (single pollutant
Tolbert-CO
Tolbert-NO2
Tolbert-PMlO
Tolbert-PMlO, NO2
Darrow (single pollutant
7,400
6,600
6,000
4,700
4,500
4,000
ER-visits (asthma); NYC (Ito et al, 2007)
single pollutant model
PM2.5
NO2
CO
SO2
8,800
6,900
5,700
9,300
7,100
270
240
210
170
160
140
500
450
400
310
300
270
720
640
580
450
430
380
400
310
250
430
320
1,800
1,400
1,100
1,900
1,400
NA
17.2
10.1
5.1
4.5
4.1
3.2
3.1
2.8
19.3
15.0
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 NYC, the model-based adjustment methodology was unable to adjust
lower alternative standard level of 60 ppb.
distributions such that they would meet the
7-60
-------
Table 7-11 Short-Term o3-attributable Morbidity Incidence, Percent of Baseline and
Reduction in o3-attributable Risk - Asthma Exacerbations (2007 and 2009)
Endpoint/Study Area/Descriptor
Air Quality Scenario
Absolute
Incidence
75ppb
Change in Incidence
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 (Gent et al., 2003, 2004)
Chest Tightness
Shortness of Breath
Chest Tightness (Ihr max)
Shortness of Breath (Ihr max)
Chest Tightness (PM2.5)
Chest Tightness (PM2.5)
Wheeze (PM2.5)
69,000
49,000
51,000
59,000
69,000
64,000
130,000
2,100
1,400
1,200
1,300
2,100
1,900
3,800
5,600
3,700
3,200
3,600
5,600
5,100
10,000
8,600
5,700
5,000
5,800
8,700
8,000
16,000
28.8
16.2
21.2
19.6
29.1
26.8
23.2
2
2
2
2
2
2
2
5
6
5
5
5
5
6
9
10
8
8
9
9
9
2009 Simulation Year
Asthma exacerbation (wheeze); Boston (Gent et al., 2003, 2004)
Chest Tightness
Shortness of Breath
Chest Tightness (Ihr max)
Shortness of Breath (Ihr max)
Chest Tightness (PM2.5)
Chest Tightness (PM2.5)
Wheeze (PM2.5)
63,000
45,000
47,000
54,000
64,000
59,000
120,000
490
330
-180
-210
500
450
900
2,400
1,600
790
910
2,400
2,200
4,300
4,800
3,200
2,200
2,500
4,800
4,400
8,700
27.0
15.1
19.8
18.3
27.2
25.1
21.7
0.4
1
-0.4
-0.4
0.4
0.4
0.5
2
3
1
1
2
3
3
5
6
3
4
5
5
6
7-61
-------
Figure 7-5 Plots of Short-Term orattributable Respiratory HA for Meeting Existing
standard and Alternative Standards (Medina-Ramon, et al., 2006) (Simulation year
2007 and 2009)
2007 Simulation year
Trend in ozone-related HA across standard levels
(HA per 100,000)
,
I 8
-Atlanta, GA
-Baltimore, MD
-Boston, MA
-Cleveland. OH
-Denver, CO
- Detroit, Ml
-Houston,!*
-LosAngeles, 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, MO
-Boston, MA
-Cleveland, OH
-Denver, CO
-Detroit, Ml
-Houston, TX
-LosAngeles, CA
-NewYork, NY
-Philadelphia, PA
-Sacramento, CA
•St Louis. MO
75ppb
70ppb
65ppb
60ppb
7-62
-------
Table 7-12 Long-Term -attributable Respiratory Mortality Incidence (2007 and 2009)
(Jerrett et al., 2009 C-R Functions) (CBSA-based study area, no threshold)
Study Area
Air Qualtiy Scenario
Absolute Incidence
75ppb
Change in Incidence
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
710
(260-1100)
750
(270-1200)
1,100
(400-1800)
530
(190-820)
480
(170-740)
760
(270-1200)
550
(190-860)
2,600
(930-4000)
1,800
(660-2900)
1,300
(450-1900)
680
(250-1100)
600
(210-930)
43
(15-71)
33
(11-55)
35
(12-58)
26
(9-43)
19
(6-31)
35
(12-59)
9.5
(3-16)
140
(46-230)
120
(41-200)
56
(19-93)
31
(10-51)
34
(230-1000)
78
(26-130)
67
(23-110)
93
(32-150)
56
(19-93)
39
(13-64)
63
(21-100)
19
(6-31)
260
(89-430)
480
(160-790)
120
(40-190)
60
(20-98)
69
(23-110)
120
(41 - 200)
110
(37-180)
140
(49 - 240)
100
(35-170)
64
(22-100)
99
(33-160)
32
(11-53)
410
(140-670)
NA
170
(59-290)
100
(34-170)
100
(35-170)
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
700
(250-1100)
730
(260-1100)
1,100
(380-1700)
510
(180-800)
490
(180-770)
720
(260-1100)
610
(220-950)
2,800
(1000-4300)
1,900
(670-2900)
1,200
(430-1900)
730
(260-1100)
580
(210-900)
41
(14-68)
25
(8-41)
6.8
(2-11)
24
(8-41)
9.0
(3-15)
-8.9
(-3 --15)
14
(5-23)
130
(45 - 220)
110
(37-180)
44
(15-73)
34
(12-57)
24
(210-910)
76
(26-120)
54
(18-90)
42
(14-69)
54
(18-89)
28
(10-47)
18
(6-30)
30
(10-49)
280
(94 - 460)
390
(130-630)
94
(32-160)
66
(22-110)
53
(18-87)
100
(36-170)
84
(28-140)
85
(29-140)
84
(29-140)
70
(24-120)
51
(17-85)
49
(17-82)
430
(150-710)
NA
140
(47-230)
110
(36-170)
86
(29-140)
NA: for NYC, the model-based adjustment methodology was unable to adjust
lower alternative standard level of 60 ppb.
distributions such that they would meet the
7-63
-------
Table 7-13 Long-Term -attributable Respiratory Mortality Percent of Baseline
Incidence and Percent Reduction in -attributable Risk (simulation years 2007 and
2009) (Jerrett et al., 2009 C-R Functions) (CBSA-based study area, no threshold)
Study Area
Air Quality Scenario
% of Baseline
Attributable to
Ozone
70ppb
Change in OS-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
17.7
18.1
16.7
16.9
20.1
17.7
16.1
19.6
15.9
17.7
17.1
17.9
5
4
3
4
3
4
1
4
6
4
4
5
9
8
7
9
7
7
3
9
24
8
7
10
15
12
11
17
11
11
5
13
NA
12
13
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
16.1
16.9
15.9
16.1
19.7
17.1
16.6
19.9
15.9
16.7
17.3
17.1
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
15
12
6
7
13
NA
10
12
13
NA: for NYC, the model-based adjustment methodology was unable to adjust
lower alternative standard level of 60 ppb.
distributions such that they would meet the
7-64
-------
Figure 7-6 Plots of Long-Term G3-attributable Respiratory Mortality for Meeting Existing
standard and Alternative Standards (Jerrett et al., 2009) (Simulation year 2007 and
2009)
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
-NewYork, NY
-Philadelphia, PA
-Sacramento, CA
-St. Louis, MO
75ppb
70ppb S5ppb
6Qppb
2009 Simulation Year
Trend in ozone-related mortality across standard
levels (deaths per 100,000)
75ppb 70ppb 65ppb SOppb
-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
7-65
-------
1 The presentation of key observations drawn from review of the core risk estimates is
2 divided into two sections: 1) the assessment of health risks associated with just meeting the
3 existing standard (section 7.5.1) and 2) the assessment of risk changes from meeting alternative
4 standards relative to meeting the existing standardfsection 7.5.2). The presentation of key
5 observations in each of these two sections is further separated into those associated with (a)
6 short-term Os-attributable mortality, (b) short-term Os-attributable morbidity and (c) long-term
7 Os-attributable mortality. Unless otherwise noted, all risk estimates discussed in these three
8 sections are core risk estimates. In some cases we refer to the confidence intervals around risk
9 estimates. When an effect estimate is drawn from a study with low statistical power, confidence
10 intervals can be wide, and can include negative values because of the assumptions of normality
11 in the distribution of the effect estimate. Negative lower-confidence bounds do not imply that
12 additional exposure to Os has a beneficial effect, but rather that the estimated O^ effect estimate
13 in the C-R function was not statistically significantly different from zero, and thus has a higher
14 degree of uncertainty as to the magnitude of the estimated risk. As noted earlier, presentation of
15 sensitivity analysis results and their use in interpreting the core risk estimates is covered in
16 section 7.5.3.
17 7.5.1 Assessment of Health Risk After Just Meeting the Existing 75 ppb standard
18 The analysis of risk after simulating just meeting the existing standard focuses on
19 absolute risk, since this is of greatest relevance in evaluating the adequacy of the existing
20 standard.
21
22 Short-term Ch-attributable mortality
23
24 • After meeting the existing standard, estimates of Ch-related all-cause mortality range
25 across urban areas from 86 to 2,800 deaths (for simulation year 2007) and from 83 to
26 2,600 deaths (for simulation year 2009) (see Table 7-7). This translates into from 0.8
27 to 4.1% of baseline all-cause mortality (for simulation year 2007) and from 0.8 to
28 3.9% (for simulation year 2009) (see Table 7-8) in these study areas. Generally, Os-
29 attributable all-cause mortality risks continue to be lower for the 2009 simulation year
30 as compared with the 2007 simulation year (with the exception of Houston),
31 reflecting the generally lower ambient O^ levels associated with 2009 for most of the
32 study areas (see Tables 7-7 and 7-8).
33 • Confidence intervals (CIs) reflecting the statistical fit of the effect estimates used in
34 modeling risk demonstrate substantial variability across the 12 urban study areas. In
7-66
-------
1 general, the upper 95* percentile CI tends to be from 2-3 times larger than the point
2 estimate for the 12 urban study areas (see Table 7-7). However, some cities have
3 markedly wider confidence intervals (e.g., Denver where the upper CI is ~5 times the
4 point estimate), while others have tighter relative CIs (e.g., New York, where the
5 upper CI is ~1.4 times larger than the point estimate). This variation in the CIs
6 associated with risk estimates can reflect a number of factors including the statistical
7 power of the underlying epidemiological study, which is based on the population size,
8 and differences in the magnitude of such factors as exposure measurement error and
9 correlations between O?, and other pollutants.
10 After just meeting the existing Oj, standard, all-cause mortality estimates based on C-
11 R functions from Smith et al., 2009 (for simulation year 2007) continue to be driven
12 largely by days with total O?, levels falling in the range of 30 to 70 ppb, with 87 to
13 99% of the mortality estimate across the 12 urban study areas associated with days in
14 this range. A smaller, but still significant fraction (9 to 24%) of the mortality risk is
15 associated with days above 60 ppb (see Figure 7-2, "Existing standard (75)" plot).32
16 For 2009, this trend continues although risk distributions are shifted down somewhat
17 (reflecting the lower ambient O?, levels generally seen in this simulation year
18 compared with 2007) (see Figure 7-3, "current standard (75)" plot). For 2009, a
19 substantial portion (2% to 24%) of Cb-attributable mortality risk is now associated
20 with days having O^ measurements 55-60 ppb or higher. A relatively smaller fraction
21 (-0% to 2%) of total mortality estimates for the existing standard are associated with
22 days having ambient Os levels of 20 ppb or less.33
23 • Estimates of O^- attributable respiratory-related HA range from 10's to 100's of cases
24 (after just meeting the existing standard) depending on the type of respiratory HA
25 endpoint modeled and the specific urban study areas evaluated (see Table 7-9). All 12
26 urban study areas were modeled for one of more respiratory-related HA endpoints.
32 Houston has a significantly smaller percentage (<1) of its mortality signal associated with days above 60ppb.
33 In the first draft O3 REA, 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 REA, U.S. EPA, 2012). For the 8hr max monitoring season LML (applicable to the Smith et al.., 2009-
based core risk estimate generated for this second draft REA), 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 B) we see that the vast majority of ozone-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 very little
change in estimates of risk.
7-67
-------
1 • Os- attributable ER (for respiratory symptoms) ranged into the thousands for both
2 New York and Atlanta under simulated attainment of the existing standard (these
3 were the only two study areas modeled for this health endpoint) (see Table 7-10).
4 • Estimates of Cb-attributable asthma exacerbation (wheeze) in Boston are in the tens
5 of thousands to over 100,000 (see Table 7-11). The percent of baseline for this
6 endpoint (after just meeting the existing standard) is generally in the 20-30% range
7 which is markedly higher than other short-term morbidity endpoints modeled for this
8 analysis (see Table 7-11 and compare to values in 7-9 and 7-10).
10 Long-term Ch-attributable mortality
11 • After simulating just meeting the existing standard, estimates of O^-related respiratory
12 mortality range across urban areas from 480 to 2,600 deaths (for 2007) and from 490
13 to 2,800 deaths (for 2009) (see Table 7-12). This translates into from 16.3 to 20.8% of
14 baseline across the 12 urban study areas (for 2007) and from 15.9 to 20.7% (for 2009)
15 using the single Jerrett et al., 2009 C-R national-scale function applied to each urban
16 area (see Table 7-13). As discussed in section 7.3.2, because of the long-term
17 exposure metric (seasonal mean of daily 8-hour maximum) employed in risk
18 modeling, there is the potential for some degree of overlap between short-term and
19 long-term exposure-related mortality estimates. For that reason, these two categories
20 of mortality estimates cannot be considered distinct and should not be added to
21 estimate total mortality.
22 • 95* percentile CIs for long-term Os-attributable respiratory mortality suggest greater
23 power (and potential less heterogeneity) associated with modeling this health
24 endpoint, compared with short-term Os-attributable mortality. None of the CIs for
25 long-term Os-attributable mortality include negative estimates as lower bounds (see
26 Table 7-12).
27 7.5.2 Assessment of Health Risk Associated with Simulating Meeting Potential
28 Alternative Standards of 70, 65, and 60 ppb
29 As discussed earlier, we have considered three alternative standard levels (70, 65 and 60
30 ppb), each evaluated using the form and averaging time of the existing standard. In presenting
31 risk estimates associated with the simulated attainment of each of these alternative standard
7-68
-------
1 levels, we focus on the change in risk associated with a comparison of Os levels after simulation
2 of the existing standard with levels after simulation of each of the alternative standard levels.
3 This is of greatest relevance in comparing the potential public health benefit associated with each
4 of the alternative standards relative to the level of protection afforded by just meeting the
5 existing standard.
6 In reviewing these risk estimates, it is important to keep in mind that simulation of
7 alternative standard levels is based on a reaching a peak-based attainment metric. Based on the
8 simulated air quality information for the 12 urban study areas, there is a tendency for Os to
9 increase on lower concentration days and decrease on higher concentration days.34 Therefore, it
10 is not immediately clear that we would expect risk reductions when applying C-R functions that
11 are based on the full distribution of daily 8-hour max values. Specifically, risk reductions are
12 only expected to the extent that the composite monitor daily 8-hour max values decrease as
13 lower alternative standards are simulated. As discussed in Chapter 4 (section ???), after
14 adjustment to alternative standard levels, decreases in Os typically occur on higher Os days
15 which tend to occur during warmer (summer) months and are concentrated in suburban areas.
16 Conversely, increases in Os, ttypically occur lower Os days which tend to occur in the cooler
17 portions of the year and are focused in core urban areas. In general, variability in predicted daily
18 63 concentrations decreases when meeting lower standard levels.
19
20 Short-term Os-attributable mortality
21
22 • In our analysis, the mortality risk metric is generally not responsive to meeting the
23 existing and alternative standard levels. This reflects a number of factors all related to
24 1) how Os concentrations respond to reductions in NOx emissions used to meet the
25 standards, and 2) how the risk metrics are associated with temporal and spatial
26 patterns of 03. As discussed in section 7.1.1, mortality risk is modeled using
27 composite monitor values (i.e., averages of 63 measurements across monitors in an
28 urban study area) which removes spatial variability in measured Os within an urban
29 study area (also removing variability in changes in 63 across an urban area resulting
30 from NOx reductions). Furthermore, in modeling total mortality risk for the core
31 analysis, we add the risk changes occurring across all days within the monitored Os
32 season, including days with low values of Os as well as days with high values of Os.
33 This means that we include both decreases in risk on those days when Os is estimated
34 to decrease (generally occurring on days with higher values of Os) and increases in
34 This relationship is also observed in ambient air quality measurements as discussed in Chapter 4 and appendices.
7-69
-------
1 risk when O^ is simulated to increase (generally associated with lower values of
2 The dampened response of short-term mortality risk can be contrasted with clinical
3 study-based risk estimates. The clinical study-based estimates primarily reflect
4 changes in the upper end of the O?, distribution where we tend to see more uniform
5 reductions under simulation of alternative standard levels. In addition, clinical-based
6 estimates of risk are based on detailed micro-environmental exposure modeling which
7 uses individual monitor values instead of composite monitor values, thereby resulting
8 is less dampening of spatial variability in 63 within a given urban study area.
9 • Generally, the magnitude of risk reduction increases as lower alternative standard
10 levels are simulated. For example, for the lowest alternative standard we evaluated,
11 60 ppb, across the 12 urban study areas, we predict from 8 to 160 fewer 63-
12 attributable deaths for simulation year 2007 (relative to risk after just meeting the
13 existing standard) (see Table 7-7). This range is from 7 to 160 deaths for simulation
14 year 2009. These ranges (for the 60 ppb standard level) translate into a 3 to 14%
15 reduction in Os-attributable risk relative to risk after just meeting the existing
16 standard (see Table 7-8).
17 • As noted in section 7.1.1, some of the urban study areas are projected to experience
18 increases in O?, (and hence risk) when attainment with the existing standard and some
19 of the alternative standard levels is simulated. Focusing specifically on the alternative
20 standard levels, we see that, for the core analysis, this potential increase in risk only
21 occurs for the 2009 simulation year and specifically for three of the urban study areas
22 (Boston, Detroit and Houston - see Table 7-7). For example, Detroit is predicted to
23 have an increase of 21 Os-attributable deaths after meeting the 70 ppb standard (when
24 compared to risk remaining after meeting the existing standard). However, we
25 estimate a net reduction of 15 (Vattributable deaths after meeting the 60 ppb level
26 (again based on comparison to risk after meeting the existing standard). Furthermore,
27 for all three urban study areas with initial risk increases (based on comparing meeting
28 the existing standard to meeting alternative standards), we see that these increases are
29 offset after meeting the lowest alternative standard simulated (60 ppb) (see Table 7-
30 7). The potential for risk increases is increased somewhat for several of the urban
31 study areas when we simulate how the Os distribution shifts from recent conditions to
32 just meeting the existing standard (see Appendix 7B, Tables 7B-1 and 7B-2).
33 Specifically, in simulating estimated risk from moving from recent conditions to
34 attaining the existing standard, we see that for the 2007 simulation year, two of the
7-70
-------
1 study areas (Houston and Los Angeles) have risk increases after meeting the existing
2 standard compared to recent conditions while half of the twelve urban study areas
3 have risk increases for the 2009 simulation year in adjusting air quality to meetthe
4 existing standard relative to recent conditions. It is also important to keep in mind
5 that, for the urban areas of New York and Los Angeles, there are additional
6 uncertainties in the simulation of existing and alternative standards given the
7 limitations in the application of the adjustment methodology to very large emissions
8 perturbations and the fact that the 95th percent confidence interval lower bound
9 estimate of hourly 03 concentrations was used to capture a scenario in which these
10 cities could meet lower standard levels (65 ppb for New York and 60 ppb for Los
11 Angeles). In five of these eight cases, the initial risk increases (including the increase
12 in going from recent conditions to the existing standard) is fully offset after meeting
13 the lowest alternative standard level (60 ppb).35
14 • Figure 7-4 provides plots of short-term mortality risk for the existing and alternative
15 standards adjusting for total exposed population (i.e., Os-attributable deaths per
16 100,000 exposed). From this figure it can be seen that total (Vattributable risk, even
17 when adjusted for population, varies substantially across the 12 urban study areas,
18 with New York and Philadelphia having the highest risk and Boston and Denver the
19 lowest. This spread in risk (adjusted for population) reflects, to a great extent,
20 differences in the effect estimates used in modeling this endpoint for each study area,
21 which can in turn reflect a number of factors (e.g., differences in behavior such as
22 outdoor activity across cities and differences in exposure measurement error).
23 However, despite considerable variability in absolute Os-attributable risk, Figure 7-4
24 also suggests that most of the study areas display relatively limited reduction in 63-
25 attributable risk across the three alternative standards (with the exception of New
26 York, which has a notable decrease in risk for the 70 to 65 ppb standard level).36 This
27 suggests that a substantial fraction of (Vattributable risk would still remain, even
28 after simulated attainment of the lowest alternative standard considered.
29 • Heat map plots of risk reductions for 2007 suggest that most of the risk reductions
30 associated with simulation of all three alternative standards occur on days with
35 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).
36 With the New York City 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) falling within that larger urban study area (see discussion in section 7.6.1).
7-71
-------
1 composite Os level between 35 and 60 ppb (see Figure 7-4). By contrast, most of the
2 risk increases occur on days with composite 63 levels between 20 and 35ppb (see
3 Figure 7-4). This is expected given that most of the increases in urban core Os are
4 associated with lower 03 days where NOx titration is prevalent (see Appendix D,
5 section 4.6 Figures 40-54). Very little of the projected change in risk (increases, or
6 decreases) for any of the alternative standards considered occurred on days with Os
7 levels below 20 ppb CbSimilar observations hold for risk results generated for
8 simulation year 2009.
9 Short-term Ch-attributable morbidity
10 • Generally, because the short-term O^ exposure-related morbidity endpoints use the
11 same air metrics as used in modeling short-term Os-attributable mortality (i.e., 8hr
12 maximum and 8hr mean) the pattern of risk reduction seen for these morbidity
13 endpoints are similar to those seen with short-term mortality (see Tables 7-9 though
14 7-11 and Figure 7-5). However, New York, as mentioned with regard to short-term
15 Os-attributable mortality, has substantially higher percent reductions (for O^-
16 attributable risk) compared with the other study areas. For example, with ER visits
17 (asthma), under the lowest alternative standard in simulation year 2007, New York is
18 estimated to have a 22 to 23% reduction in the number of ER-visits associated with
19 63 exposure (see Table 7-10).
20 Long-term Ch-attributable mortality
21 • Although long-term Os-attributable mortality is modeled using a different O?, metric
22 (essentially a long-term trend in the Ihr maximum for the hottest two seasons - see
23 section 7.3.2) the overall magnitude and pattern of reduction in Os-related risk is
24 similar to that seen with short-term exposure related mortality. Specifically, for the
25 2007 simulation year, for most urban study areas risk reductions range from 11 to
26 15% (for the 60 ppb standard) (see Table 7-13). Risk reductions are generally slightly
27 smaller across alternative standard levels for simulation year 2009. For the 2009
28 simulation year, for Detroit, we see a relatively small risk increase for the 70 ppb
29 alternative standard (compared to risk under the existing standard). However that
30 initial increase is offset by risk reductions for the other (lower) alternative standard
31 levels simulated (see Table 7-13).
7-72
-------
1 7.5.3 Sensitivity Analyses Designed to Enhance Understanding of the Core Risk Estimates
2 We have completed a number of sensitivity analyses intended to support interpretation of
3 the core risk estimates. These sensitivity analyses, which are described in section 7.4.3, can be
4 divided into two categories: (a) sensitivity analyses exploring factors impacting air quality
5 characterization (specifically composite monitor composition) and (b) sensitivity analyses
6 exploring the impact of alternative C-R function specifications. As noted in section 7.4.3, we
7 also completed an initial influence analysis designed to identify which of the input factors to the
8 risk model (for short-term exposure-related mortality) are primarily responsible for inter-city
9 variability in that risk metric. This section summarizes the results of these sensitivity analyses
10 and presents key observations related to those analyses, beginning with the influence analysis
11 and then proceeding to sensitivity analyses focused on air quality characterization and alternative
12 C-R function specification.
13
14 Influence analysis
15 The influence analysis considered three factors involved in modeling risk for the short-
16 term exposure-related mortality endpoint including: baseline incidence, composite monitor 63
17 levels and Bayes-adjusted city-specific effect estimates (recall that the core risk estimate is based
18 on effect estimates derived as part of analyses published in Smith et al., 2009). Each of these
19 input factors displays inter-study area variation and are responsible, collectively, for
20 heterogeneity in risk estimates.37 In completing the analysis, we first calculated a central
21 tendency estimate of risk based on the mean of each input factor across the 12 urban study areas
22 for the 2009 simulation year (i.e., using the average of the city-specific values for each of the
23 input factors). We then systematically varied each of the three heterogeneity-related factors
24 (effect estimate, composite monitor-based 63 level and baseline incidence) to one standard
25 deviation (SD) above its mean value (reflecting variance across the 12 urban study area values)
26 and noted the percent increase produced by that perturbation over the initial mean risk estimate.
27 This influence analysis allowed us to explore the impact of both model form - specifically,
28 potential non-linearities in the model - as well as the relative magnitude of variability in each of
29 the three heterogeneity-related input factors on risk. The influence analysis generated the
30 following results: baseline incidence (23%), composite monitor-based Os level (8%), and effect
31 estimate (58%). In other words, the 58% result for effect estimate means that use of a value 1 SD
32 over the mean (for the effect estimate) in generating risk, resulted in a risk estimate that was 58%
37 Note, that the demographic count input factor also varies across the study areas and is an important factor in
determining total incidence. 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.
7-73
-------
1 larger than the risk estimate based on the mean of all input factors. These results clearly show
2 that, of the three input factors considered, the effect estimate is primarily responsible for inter-
3 city variability in short-term exposure-related risk.
4 Interestingly, when we look at the coefficient of variation (CV) for these three
5 heterogeneity-related input factors we see values almost identical to the influence analysis results
6 in terms of relative magnitude to each other (i.e., 0.232, 0.084, and 0.527 for baseline incidence,
7 composite monitor-based 63 levels and effect estimate, respectively). Given that the CV values
8 only reflect variability in each input factor and not model form (i.e., do not reflect potential non-
9 linearities in the model), the fact that the CV values almost exactly match the influence analysis
10 results in terms of relative magnitude suggest that there is very little if any non-linearity in the
11 model calculations involving these three input factors. Had non-linearity existed to a significant
12 extent, then the influence anlaysis results would have differed substantially from the CV results.
13 The fact that both analyses suggest a primary role for the effect estimate in driving inter-city
14 variability in risk emphasizes the importance of the sensitivity analyses exploring alternative C-R
15 functions specifications that were completed for the REA (see below).
16
17 Air quality-related sensitivity analyses
18
19 This category of sensitivity analysis covers (a) the use of a smaller study area (the Smith
20 et al., 2009 study areas) as contrasted with the CBSA-based study areas used in the core analysis,
21 and (b) the use of alternative approaches to simulate attainment of the existing and alternative
22 standards (for a subset of the study areas) (see section 7.4.3 for additional detail). This category
23 of sensitivity analysis was applied to short-term (Vattributable mortality given the importance of
24 the endpoint in the policy-context.38
25 To allow for easier visual comparisons, we have presented the results of this sensitivity
26 analysis category in graphical form (see Figure 7-7, numerical results are presented in Appendix
27 7C). This figure presents point estimates and 95* percentile confidence ranges for the core model
28 and for two sensitivity analyses: (a) SA1 (use of the smaller Smith et al., 2009 based study area)
29 and (b) SA2 (use of the alternative approach to simulating attainment). SA2 is not presented for
30 all of the study areas, only for the subset included in these alternative simulations (see section
31 7.4.3). The sensitivity analyses results presented in Figure 7-7 are the changes in Os-related risk
32 that result from meeting the three alternative standards relative to meeting the existing standard.
33 Furthermore, these changes reflect deaths per 100,000, which standardizes the estimates on
38 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.
7-74
-------
1 population. This removes variation in the size of the underlying exposed population as a factor to
2 consider in interpreting these results.
3 For the sensitivity analysis examining use of the smaller Smith et al., 2009 study area, we
4 have also included heat maps similar to those used in conveying core estimates for short-term
5 exposure related mortality (see section 7.5 for a description of the heat maps used in the core
6 analysis). These heat maps (included in Appendix C - see Figure 7C-1) allow us to consider how
7 changes in risk, including both reductions in risk and increases in risk are distributed across the
8 63 air quality distributions for each study area.
9 Key observations related to the air quality-related sensitivity analyses include:
10
11 • Use of smaller study area reduces magnitude of risk reduction: For most of the
12 study areas, use of the smaller Smith et al., 2009-based study area resulted in smaller
13 risk reductions (again expressed in terms of changes in deaths per 100,000). For
14 example, in Figure 7-7 (Baltimore plot), we see that estimated change in risk for SA1
15 (the smaller study area) are lower than estimated change in risk for the core scenario.
16 This likely reflects the mix of monitors in the smaller study areas which results in a
17 smaller change in the composite monitor value (for the existing standard versus
18 alternative standard levels) as compared with composite monitor values based on the
19 larger CBSA study area. However, it is important to keep the relative small
20 magnitude of these risk reductions in mind when considering these sensitivity
21 analysis results. Most of these differences in risk reductions are less than 1 individual
22 per 100,000 which reflects the fact that total risk reduction (for short-term 63-
23 attributable mortality) across the urban study areas is relatively small (see Table 7-7).
24 o Reductions in risk are focused on higher 63 days while increases
25 are focused on lower OT, days: Figure 7C-1 allows us to consider patterns
26 in risk reductions and increases when using the smaller Smith et al., 2009-
27 based study areas in modeling risk. Figure 7C-1 (particularly the plots of
28 risk decreases) suggests that decreases in risk tend to occur on days with
29 composite monitor Os concentrations ranging from 40-70ppb, while
30 increases in risk tend to occur on days with composite monitor values in
31 the range at or below 30-40 ppb (with most risk increases falling in the
32 range of 15ppb to 40ppb). As noted in 7.1.1, there is less confidence in
33 specifying the nature of the C-R function (and therefore less confidence in
34 specifying risk) in the range below 20 ppb.
7-75
-------
1 • Application of effect estimates derived for smaller study areas to larger CBSA-
2 based study areas: As noted in section 7.3.2, in those instances where an
3 epidemiological study provides effect estimates for multiple subareas within a larger
4 CBSA-based study area, we are selecting the effect estimates that represent the
5 largest number of individuals to model that CBSA-based study areas. There is
6 uncertainty associated with this approach. Specifically, as illustrated in Table 7-3,
7 effect estimates within some of the CBSA-based study areas can display considerable
8 heterogeneity. For example, consider the Smith et al., 2009-based effect estimates
9 that fall within the CBSA-based New York study areas (these vary from 0.0001 to
10 0.0009 - almost a 10 fold factor, see Table 7-3). Furthermore, with the CBSA-based
11 New York study area, Smith et al., 2009-based effect estimates only cover about half
12 of the total population, with 8.3 million residents living within portions of the CBSA
13 not covered by Smith et al., 2009-based effect estimates. As noted in section 7.3.2, in
14 these types of situations, we have decided to use the single effect estimates
15 representing the largest number of residents in modeling the larger CBSA-study area.
16 This reflects the observation that, in the case of the New York CBSA, one of the
17 available effect estimates (for the New York study area), represents ~7 times the
18 population of the other effect estimates (see Table 7-3). In the case of the Los
19 Angeles CBSA, there is significantly less difference between the available effect
20 estimates, making the issue of heterogeneity (and the specification of a single effect
21 estimate for this study area) less important. Never the less, we recognize that the issue
22 of heterogeneity does complicate extrapolation of effect estimates for smaller study
23 areas to the larger CBSA study areas modeled in this analysis and does introduce a
24 degree of uncertainty that is difficult to characterize.
25 • Use of alternative approach for simulating attainment of existing and alternative
26 standard levels: Use of an alternative approach to simulate attainment of the existing
27 and alternative standard levels did not produce a consistent trend in terms of changes
28 in risk between existing and alternative standards relative to the core analysis. For
29 example, if we look at Figure 7-7 (plot for Houston), we see that SA2 (reflecting
30 application of the alternative simulation approach) has a larger risk reduction than the
31 core estimate. By contrast, if we look at the plot for Los Angeles, we see that the SA2
32 risk change is lower than the core estimate. Again, as with the sensitivity analysis
33 results looking at study area size, it is important to keep in mind that the magnitude of
34 these differences is relatively small, reflecting the small magnitude of mortality risk
35 associated with these analyses in general (see Table 7-7). It is also important to note
36 that in the alternative simulation approach, the HDDM-adjustment approach assumed
7-76
-------
1 the same percent reductions of NOx and VOC and did not examine if a different air
2 quality distribution could have been obtained with a different combination of NOx
3 versus VOC reductions. For most of the urban areas, the percent NOx and VOC
4 reductions were very similar to the NOx-only percent reductions. The similarity in the
5 NOx reductions between the two approaches could be the reason for there being little
6 difference in the risk estimates between the core and the alternative approach.
7
8 Sensitivity analyses related to specification of C-R functions
9
10 This category of sensitivity analysis covers a number of factors related to the
11 specification of C-R functions for both short-term Os-attributable and long-term Os-attributable
12 mortality. In the case of short-term Os-attributable mortality, we consider (a) the use of Bayes
13 adjusted effect estimates using regional priors (as contrasted with the Bayes adjusted values
14 using a national prior applied in the core analysis), (b) the use of a copollutants model
15 considering PMio (as contrasted with the single pollutant model used in the core analysis) and (c)
16 application of effect estimates from Zanobetti and Schwartz 2008 reflecting a summer focused
17 analysis (as contrasted with the Smith et al., 2009-based analysis reflecting the entire monitoring
18 period in each study area, which is used in the core analysis). For long-term Os-attributable
19 mortality, we consider the use of regionally-differentiated single pollutant effect estimates
20 obtained from Jerrett et al., 2009, as contrasted with the single national copollutants model used
21 in the core analysis (see section 7.1.1). We also present estimates for long-term Oj attributable
22 mortality based on application of results from a national level single pollutant model.
23 For sensitivity analyses examining alternative specification of the C-R function for short-
24 term Os-attributable mortality, we have used the same graphical approach as used in presenting
25 results of the sensitivity analyses examining air quality characterization (i.e., plots of point
26 estimates with 95th percentile C.I.s for the core and sensitivity analyses for each of the study
27 areas - see Figure 7-8). Here we also plot estimates of risk changes using deaths per 100,000 to
28 standardize in terms of total exposed population. For the sensitivity analysis considering
29 alternative C-R functions for long-term Os-attributable mortality, we present results in tabular
30 form (Table 7-14). Specifically, for both the core and sensitivity analysis, we present (a) the
31 percent of baseline mortality attributable to Os (under simulated attainment of the existing
32 standard) and (b) the percent reduction in Os-attributablerisk for each of the alternative standard
33 levels. Key observations related to sensitivity analyses examining alternative C-R functions
34 specifications include:
35 • Use of regional Bayes-adjusted effect estimates in modeling short term 03-
36 attributable mortality: The use of Bayes-adjusted effect estimates with regional
7-77
-------
1 priors in modeling short-term Os-attributable mortality, had a mixed impact across the
2 urban study areas, with some study areas having increased changes in risk and others
3 having smaller changes, relative to the core analysis. For example, in Figure 7-8 (plot
4 for Baltimore), SA1 had a larger change in risk compared with the core analysis.
5 However, as with the sensitivity analyses examining air quality-related factors
6 (discussed above), it is important to keep in mind that the overall magnitude of the
7 Os-attributable mortality risk is relatively small and that these differences in changes
8 in risk (comparing SA1 to the core analysis) are generally in the fraction of a person
9 per 100,000 exposed population.
10 • Use of a copollutants model (with PMio) in modeling short-term Os-attributable
11 mortality: The use of the PMio copollutant model in modeling short-term 63-
12 attributable mortality (as contrasted with the single pollutant model used in the core
13 analysis) tended to have a relatively small effect on estimates of risk changes for the
14 alternative standards considered. For example, in Figure 7-8 (plot for Boston), we see
15 that the estimates of risk changes for SA2 (reflecting application of the PMio
16 copollutant model) is essentially the same as the core risk estimate. It is important to
17 keep in mind that the PMio copollutant model suffers from significantly reduced
18 power due to the 1/3 to 1/6 day sampling frequency used in measuring PMio (this
19 reduces the number of observations available to support epidemiological analysis).
20 This has the impact of greatly increasing the confidence intervals on the SA2 risk
21 estimates relative to the core estimates.
22 • Use of Zanobetti and Schwartz 2008 effect estimates in modeling short-term 63-
23 attributable mortality: The use of Zanobetti and Schwartz, 2008 effect estimates
24 (reflecting a focus on the warmer summer months) produces a mixed set of results
25 when compared to the core risk estimates. If we look at Figure 7-8 we see that, for
26 Boston, estimates of risk changes for SA3 (reflecting application of the Zanobetti and
27 Schwartz 2008 effect estimates) are significantly larger than core estimates. By
28 contrast, SA3 estimates of risk changes for Houston are significantly smaller than the
29 core estimates. It is important, however to keep in mind that the Zanobetti and
30 Schwartz 2008 effect estimates will tend to under-estimate total risk since they only
31 model impacts during the summer months (while the Smith et al., 2009 effect
32 estimates allow us to model impacts for the entire Os monitoring season in each study
33 area). Note that if the 63 effect were only occurring during the summer months, then
34 the total risk estimated using effect estimates from the two studies would be similar.
35 However, because the risks in many locations are smaller (using the Zanobetti and
7-78
-------
1 Schwartz, 2008 based effect estimates), this suggests that the O^ effect occurs outside
2 of the summer months evaluated in this study.
3 • Use of regional-differentiated effect estimates in modeling long-term 63-
4 attributable mortality: Risk estimates generated using regional-specific effect
5 estimates for long-term (Vattributable mortality differ substantially from the core
6 estimates based on a single national-level effect estimate (see Table 7-14).
7 Furthermore, the risk estimates generated using the regional effect estimates display
8 considerable variability (see Table 7-14) reflecting the significant variability in the
9 underlying effect estimates (see Jerrett et al., 2009, Table 4). The regional effect
10 estimates range from 0.99 (for the Northeast) to 1.21 (for the Southwest) and include
11 1.00 (no Os effect for the Industrial Midwest). As noted earlier in section 7.5,
12 negative risk estimates should not be interpreted as suggesting that 03 exposure is
13 beneficial. Rather, these suggest that there may be instability in the underlying
14 estimates or that potential confounding has not been fully addressed. Regional effect
15 estimates used in this analysis have considerably larger confidence intervals than the
16 national estimate (compare values in Jerrett et al., 2009 Table 3 with values in Table
17 4). This suggests that the regional estimates are less stable than the national estimates
18 and are subject to considerably greater uncertainty. For this reason, while the results
19 of this sensitivity analysis point to the potential for regional heterogeneity in the long-
20 term Os-attributable mortality effect estimate, we do not have significant confidence
21 in the regionally-based risk estimates themselves given the relatively large confidence
22 intervals associated with those estimates.
23 • Use of national-based single pollutant model in modeling long-term Os-
24 attributable mortality: Risk estimates generated using the national-level Cb-only
25 effect estimate were significantly lower (-30%) than the core risk estimates which
26 utilize a copollutants model (which includes PM^.s) (see Table 7-15). In this case,
27 control for another pollutant results in a stronger 63 signal, possibly due to an
28 association between PM2.5 and a confounder or effect modifier associated with the
29 O3-related effect.
7-79
-------
Figure 7-7 Sensitivity Analysis: Short-Term o3-attributable Mortality (air quality-related factors including study area size
and method used to simulate attainment of existing and alternative standard levels) (2009) SAl-smaller (Smith-based)
study area, SA2-alternative method for simulating standards.
Standard levels(delta)
75-70 75-65 75-60
\\
a 0 -
g. u
£ -1 -
I'1
4
|s
*• ^
a. x
I-1
hs per 100,000
3 O Q M M
£•
a ° -
2 -|
§2
r '
Q
Atlanta, GA
ll
f • T f T T
1
Q v> < 3 % < ^j * <
Core and Sensitivity Anatysis Simulatons
Cleveland, OH
ll
f T
* * t T IT
1 < 3 | 5 2 J S Sj
Core and Sensitivity Analysis Simutatons
Houston, TX
. . Lt
• "
Core and Sensitivity Analysis Simulatons
Philadelphia, PA
,,
T ,
t.t Lt U
I S a |aa I a a
Core and Sensitivity Analysis Simulatons
2 -i
&
I-1-
|i-
§ i-
r a
a.
is per 100,000
is u -
£ -i
2 -
I 0 -
Baltimore, MD
ll
11 ' •
f* T t T
i
5 5 < 5 i/i < 5^9
Core and Sensitivity Analysis Simulatons
Denver, CO
1
• i— •— * " • "
1
Core and Sensitivity Analysis Simulatons
Los Angeles, CA
SrfS SrfS ^^rf
Core and Sensitivity Analysts SimuEatons
Sacramento, CA
1 1
t t t T T T t
1
Core and Sensitivity Analysis Simuiatons
li:
&
i-
OOO'OOT
-i -
^ -i -
4 -i
§3 -
5 2 -
1
*
3 -
li :
8.-i
1 -3
a
Boston, (JA
rS Pnri
,1 M T i
• 1 f i i
T
5*8 s*< s S <
Core and Sensitivity Analysis Simulatons
Detroit, Ml
1 * * i *
* T
Core and Sensitivity Analysis Simuiatons
New York, NY
• m
Core and Sensitivrty Analysis Simulatons
St. Louis, MO
*, h
1 I
-------
Figure 7-8 Sensitivity Analysis: Short-Term o3-attributable Mortality (C-R function specification) (2009) SAl-regional Bayes-
based adjustment; SA2-copollutant model (PMIO); SA3-Zanobetti and Schwartz-based effect estimates.
Standard levels (delta)
75-70 75-65 75-60
3 -
I'
&1" 0 -
4 -
S. o
= -i -
a-2 -
i
S.o -
|.J
KT 100.000
£ 1
Atlanta, GA
' I
Core and Sensitivity Analysis Simulatons
Cleveland, OH
rill
* • i . tl_M
Core and Sensfttvrty Analysis Simutatons
Houston, TX
1
ma** * * •'.*
Core and Sensitivity Analysis Simulatons
Philadelphia, PA
| * * |
5 < < < 5 < < < §<<<
Core and Sensitivity Analysis Simulatons
2 -
S. 0 -
2 -
g. j
£ „
I-
a i -
L.
1 'I '
Baltimore, MD
•M' tM* }*'
•3 -1 'S> 'SI ..^i/ll/li/l s5 VI i
Core and Sensitivity Analysis Simubtons
Denver, CO
L 1 f 1
' 1
3 % % 5 3^53 35tt
Core and Sensitivity Analysts Simulatons
LosAngeles, CA
— * . * f i* H
Core and Sensitwity Analysis Simutatons
Sacramento, CA
!<<< 5<<< 5<-=
Core and Sensitivity Analysts Simutatons
•}
!
S
.1
1 1
* a
It
|
it
c <
|i:
s i-
i 0 -
|-i-
i -
j
«9
5 -
E. x "
|.2-
4 -
[ii
g. 0
1-1 -
i-2 -
1 Boston, MA 1
rS rhrS
ftit
T
Core and Sensitivitv Analysis Simulatons
Detroit, Ml
i I * * ' *
r t T
3 «l @ ^ ;3^^;/i g 3 M W
Core and Sensitrvrty Analysis SimuEatons
New York, NY
Mu
11 t • T
1
Core and Sensitivity Analysis Simulatons
St. Louis, MO
• . t * u_U \ t ' t
£<<« £<<< 5<<<
Core and Senshtivity Analysis Simulatons
7-81
-------
1
2
3
Table 7-14 Sensitivity Analysis for Long-Term -attributable Respiratory Mortality -
Alternative C-R Function Specification (regional effect estimates) % of baseline all-
cause mortality and change in -attribuable risk (2009) (Smith et al., 2009, season))
4
5
6
7
Study Area
Air Quality Scenario
Baseline
Incidence
Attributable to
Ozone Change in O3-Attributable Risk
75ppb
75-70
75-65
75-60
Core Simulation
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
16.6
17.4
15.9
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
4
10
6
3
9
5
2
4
8
18
7
8
8
14
10
7
15
12
6
7
13
18
10
12
13
Sensitivity analysis
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
41.21
-7.01
-6.19
0.00
27.38
0.00
41.15
4.46
-6.61
-6.89
24.79
0.00
4
4
1
0
1
0
2
3
7
4
4
0
8
9
5
0
4
0
3
6
24
9
7
0
11
13
9
0
11
0
6
10
NA
13
11
0
NA: for NYC, the model-based adjustment methodology was unable to adjust
lower alternative standard level of 60 ppb.
, distributions such that they would meet the
7-82
-------
1
2
3
4
Table 7-15 Sensitivity Analysis for Long-Term -attributable Respiratory Mortality -
Alternative C-R Function Specification (national -only effect estimates) % of
baseline all-cause mortality and change in -attribuable risk (2009) (Smith et al.,
2009, season))
5
6
7
Study Area
Air Quality Scenario
Percent of
Baseline
Incidence
75ppb
Change in O3-Attributable Risk
75-70
75-65
75-60
Core Simulation
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
16.6
17.4
15.9
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
4
10
6
3
9
5
2
4
8
18
7
8
8
14
10
7
15
12
6
7
13
18
10
12
13
Sensitivity analysis
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
11.1
11.8
14.1
11.9
11.9
14.6
11.7
12.0
12.6
12.4
5
3
1
4
2
-1
2
4
5
3
4
4
10
7
3
10
5
2
4
9
19
7
8
8
14
10
7
15
12
6
7
14
19
10
13
13
NA: for NYC, the model-based adjustment methodology was unable to adjust
lower alternative standard level of 60 ppb.
, distributions such that they would meet the
7-83
-------
1 7.6 KEY OBSERVATIONS REGARDING OVERALL CONFIDENCE IN THE RISK
2 ASSESSMENT AND RISK ESTIMATES
3 This section discusses our overall confidence associated with risk estimates presented in
4 this draft of the REA. We begin by presenting a set of key observations related to overall
5 confidence in the risk assessment. These observations are drawn largely from (a) consideration
6 for the systematic approach used in designing the risk assessment, (b) our assessment of the
7 degree to which we have captured key sources of variability in the analysis (section 7.4.1) (c) our
8 qualitative assessment of uncertainty in the risk assessment (section 7.4.2), and (d) the results of
9 the sensitivity analyses completed (section 7.5.3). Once we present these observations, we
10 provide a synthesis statement reflecting our overall degree of confidence in the risk estimates (at
11 the end of this section). Key observations addressing overall confidence in the analysis include:
12 • A deliberative process was used in specifying each of the analytical elements
13 comprising the risk model. This is in line with recommendations made by the
14 National Research Council in Science and Decisions, Advancing Risk Assessment
15 (NRC, 2009. P. 89-90) for improving risk assessment as applied in the regulatory
16 context. This deliberative process included first identifying specific goals for the
17 analysis, and then designing the analysis to meet those goals, given available
18 information and methods. Specific analytical elements reflected in the design include:
19 selection of urban study areas, characterization of ambient air Os levels, selection of
20 health endpoints to model and selection of epidemiological studies (and specification
21 of C-R functions) (see sections 7.1.1 and 7.3). In addition, the design of this draft of
22 the REA reflects consideration for comments provided by the public and by CASAC
23 in their review of the 1st draft REA in letter form (Frey, H. C. 2012.).
24 • Review of available literature (as specified in the 63 ISA, U.S. EPA. 2013a), resulted
25 in a decision not to incorporate a true (no effect) threshold into our risk modeling.
26 Conversely, the studies used to develop the C-R functions indicate a range of ambient
27 63 (area-wide daily levels, based on averaging across monitors in locations with
28 multiple monitors, of < 20ppb) below which there is reduced confidence in specifying
29 the nature of the concentration-response relationship, based on less data in the studies,
30 specifically for short-term O3-attributable respiratory mortality and morbidity (see
31 section 7.1.1). In any case, only a relatively small fraction of short-term 03-
32 attributable mortality reflected in the risk estimates is associated with days in this
33 range with the vast majority of the risk estimates reflecting days with peak Os
34 measurements well above this level (see section 7.5.1 and 7.5.2). 03
35 • Modeling of short-term O3-attributable mortality utilized Bayes-adjusted city-specific
36 effect estimates (see section 7.1.1 and section 7.3.2). These effect estimates are
7-84
-------
1 considered to have increased overall confidence since they combine elements of the
2 local city-specific signal with a broader scale (national) signal.
3 • Use of CBSA-based study areas in modeling all health endpoints in order to address
4 known bias associated with using smaller study areas. As discussed in 7.1.1, we have
5 used larger CBSA-based study areas to avoid focusing the risk assessment only on
6 core urban areas (often used in the epidemiological studies providing effect estimates)
7 which can experiences increases in O^ based on simulated attainment of both existing
8 and alternative standard levels. There is uncertainty in using effect estimates based on
9 smaller study areas to represent larger CBSA-based study areas (see section 7.4.2 and
10 7.5.3). A key concern is heterogeneity in the effect estimates which may suggest
11 increased uncertainty in applying effect estimates to larger study areas (since larger
12 study areas may display heterogeneity in the nature of the relationship between O^
13 exposure and risk). It is possible also that this heterogeneity varies across urban areas,
14 or regionally. For both categories of mortality endpoints (short-term and long-term
15 Os-attributable), potential heterogeneity in the mortality effect even within larger
16 urban areas remains a potentially important source of uncertainty.
17 • Specifically in relation to short-term exposure-related mortality and morbidity which
18 depend on time-series studies, there is uncertainty in applying effect estimates derived
19 based on evaluating the longitudinal (in terms of time) relationship between ambient
20 OT, and a particular health effect to the modeling of a discrete shift in the entire
21 distribution that occurs when you simulate an alternative standard. Specifically, the
22 time-series studies relate unit changes in day to day 63 with a degree of impact on
23 baseline health effect rates. In the risk assessment, we use this effect estimate to
24 predict risk for a unit shift in daily composite monitor value. There is uncertainty in
25 this application of the effect estimates, although it is not possible at this time to
26 characterize either qualitatively, or quantitatively the magnitude of this uncertainty
27 and the degree of any potential bias that could be introduced into the simulation of
28 risk.
29 • Use of HDDM-adjustment approach to simulate attainment of both the existing and
30 alternative standard levels provides more refined estimates of ambient 63
31 distributions given its ability to characterize the physical and chemical processes of
32 Os formation in the atmosphere However, in the case of both the New York and Los
33 Angeles study areas, given the limitations in the application of the adjustment
34 methodology to very large emissions perturbations and the need to use the 95th
35 percent confidence interval lower bound estimate to simulate attainment of these
36 standard levels, we have reduced overall confidence in the simulation of the 63
7-85
-------
1 concentrations for these study areas and consequently all health endpoints modeled
2 for risk for these two study areas (see section 7.4.2 and 7.5.3).
3 • Sensitivity analyses exploring alternative C-R functions for modeling short-term Ch-
4 attributable mortality (e.g., Bayes regional prior based estimates, copollutants
5 models) suggested that alternative models can have a moderate impact on risk (see
6 section 7.5.3). This modest impact reflects primarily the relatively small magnitude of
7 short-term Os-attributable mortality reductions simulated for the alternative standard
8 levels.
9 • The use of alternative C-R functions for modeling long-term Ch-attributable mortality
10 (specifically the regional-based estimates referenced earlier) was shown to have a
11 significant impact on risk (see section 7.5.3). However, concerns over the power and
12 hence stability of the regional effect estimates used in this simulation limit our ability
13 to draw firm conclusions regarding the potential magnitude of that regional
14 heterogeneity.
15
16 Based on the key observations regarding confidence presented above, we draw the
17 following conclusions regarding overall confidence in the risk estimates generated for this draft
18 of the REA. We have a reasonable degree of confidence in short-term Ch-attributable mortality
19 and morbidity estimates for ten of the twelve study areas. This confidence is tempered somewhat
20 by concerns over potential heterogeneity in effect estimates for mortality which can impact the
21 risk assessment given our use of larger CBSA-based study areas. Our confidence in risk
22 estimates generated for both New York and Los Angeles is considerably lower than for the
23 remaining ten study areas due to (a) concerns over air quality modeling (specifically the use of
24 lower-bound fits to the DDM model) and (b) specifically in the case of New York, evidence for
25 significant heterogeneity in the mortality effect estimates for subareas within the CBS A. For
26 long-term O^-attributable mortality, we also have a reasonable degree of confidence in our risk
27 estimates. However, as with short-term (Vattributable mortality, this confidence is also
28 tempered by concerns over regional heterogeneity in the Os effect. If we had regionally-
29 differentiated effect estimates for this endpoint that had sufficient power and stability, we would
30 consider using these as the basis for generating core risk estimates (rather than the national-level
31 effect estimate used in the current analysis).
32
7-86
-------
1 7.7 REFERENCES
2
3 Abt Associates Inc. 1996. A Paniculate Matter Risk Assessment for Philadelphia and Los Angeles.
4 Prepared for Office of Air Quality Planning and Standards. Research Triangle Park, NC: EPA
5 Office of Air and Radiation, OAQPS, July 1996. Available at:
6 .
7 Abt Associates, Inc. 2010. Environmental Benefits and Mapping Program (Version 4.0). Prepared for
8 U.S. Environmental Protection Agency, Bethesda, MD. Research Triangle Park, NC: EPA Office
9 of Air Quality Planning and Standards. Available on the Internet at:
10 .
11 Akinbami, LJ; C. D. Lynch; J. D. Parker and T. J. Woodruff. 2010. "The Association Between Childhood
12 Asthma Prevalence and Monitored Air Pollutants in Metropolitan Areas, United States, 2001-
13 2004." Environmental Research, 110: 294-301.
14 Bell, M. L. and F. Dominici. 2008. "Effect Modification by Community Characteristics on the Short-term
15 Effects of O3 Exposure and Mortality in 98 U.S. Communities." American Journal of
16 Epidemiology, 167: 986-997.
17 Bell, M. L.; A. McDermott; S. L. Zeger; J. M. Samet; F. Dominici. 2004. "O3 and Short-term Mortality in
18 95 U.S. Urban Communities, 1987-2000." JAMA, 292: 2372-2378.
19 Darrow, L. A.; M. Klein; J. A. Sarnat; J. A. Mulholland; M. J. Strickland; S. E. Sarnat, et al. 2011. "The
20 Use of Alternative Pollutant Metrics in Time-series Studies of Ambient Air Pollution and
21 Respiratory Eemergency Department Visits." Journal of Exposure Science and Environmental
22 Epidemiology, 21,10-19.
23 Frey, H. C. 2012. "Letter from Clean Air Scientific Advisory Committee to the Honorable Lisa P.
24 Jackson, Administrator, US EPA. CASAC Review of the EPA's Health Risk and Exposure
25 Assessment for O3 (First External Review Draft - Updated August 2012) and Welfare Risk and
26 Exposure Assessment for O3 (First External Review Draft - Updated August 2012)," November,
27 19,2012.
28 Gent, J. F.; E. W. Triche; T. R. Holford; K. Belanger; M. B. Bracken; W. S. Beckett, et al. 2003.
29 "Association of Low-level O3 and Fine Particles with Respiratory Symptoms in Children with
30 Asthma." Journal of the American Medical Association, 290(14), 1859-1867.
31 Ito, K.; G. D. Thurston and R. A. Silverman. 2007. "Characterization of PM2 5, Gaseous Pollutants, and
32 Meteorological Interactions in the Context of Time-series Health Effects Models." Journal of
33 Exposure Science and Environmental Epidemiology, 17 Suppl 2, S45-60.
34 Jerrett, M.; R. T. Burnett; C. A. Pope, III; K. Ito; G. Thurston; D. Krewski; Y. Shi and E. Calle; M. Thun.
35 2009. "Long-term O3 Exposure and Mortality." New England Journal of Medicine, 360:1085-
36 1095.
37 Katsouyanni, K.; J. M. Samet; H. R. Anderson; R. Atkinson; A. L. Tertre; S. Medina, et al. 2009. "Air
38 Pollution and Health: A European and North American Approach (APHENA)," Health Effects
39 Institute.
40 Lin, S.; E. M. Bell; W. Liu; R. J. Walker; N. K. Kim and S. A. Hwang. 2008a. "Ambient O3
41 Concentration and Hospital Admissions Due to Childhood Respiratory Diseases in New York
42 State, 1991- 2001." Environmental Research, 108, 42-47.
43 Lin, S; X. Liu; L. H. Le; S. A. Hwang. 2008. "Chronic Exposure to Ambient O3 and Asthma Hospital
44 Admissions Among Children." Environonmental Health Perspective, 116: 1725-1730.
7-87
-------
1 Linn, W. S.; Y. Szlachcic; H. Gong, Jr.; P. L. Kinney and K. T. Berhane. 2000. "Air Pollution and Daily
2 Hospital Admissions in Metropolitan Los Angeles." Environmental Health Perspective, 108(5),
3 427-434.
4 Medina-Ramon, M.; A. Zanobetti and J. Schwartz. 2006. "The Effect of O3 and PM10 on Hospital
5 Admissions for Pneumonia and Chronic Obstructive Pulmonary Disease: A National Multicity
6 Study." American Journal of Epidemiology, 163(6), 579-588.
7 Meng, Y. Y.; R. P. Rull; M. Wilhelm; C. Lombardi; J. Balmes and B. Ritz. 2010. "Outdoor Air Pollution
8 and Uncontrolled Asthma in the San Joaquin Valley, California." Journal of Epidemiology
9 Community Health, 64: 142-147.
10 Moore, K; R. Neugebauer; F. Lurmann; J. Hall; V. Brajer; S. Alcorn; I. Tager. 2008. "Ambient O3
11 Concentrations Cause Increased Hospitalizations for Asthma in Children: An 18-year Study in
12 Southern California." Environmental Health Perspective, 116: 1063-1070.
13 National Research Council (NRC). 2009. Science and Decisions, Advancing Risk Assessment. Committee
14 on Improving Risk Analysis Approaches. Washington, DC: The National Academies Press.
15 Silverman, R. A.; and K. Ito. 2010. "Age-related Association of Fine Particles and O3 with Severe Acute
16 Asthma in New York City." Journal of Allergy Clinical Immunology, 125(2), 367-373 e365.
17 Smith, R.L.; B. Xu and P. Switzer. 2009. "Reassessing the Relationship Between O3 and Short- term
18 Mortality in U.S. Urban Communities." Inhalation Toxicology, 21: 37-61.
19 Strickland, M. J.; L. A. Darrow; M. Klein; W. D. Flanders; J. A. Sarnat; L. A. Waller, et al. 2010. "Short-
20 term Associations between Ambient Air Pollutants and Pediatric Asthma Emergency Department
21 Visits." American Journal of Respiratory Critical Care Medicine, 182, 307-316.
22 Tolbert, P. E.; M. Klein; J. L. Peel; S. E. Sarnat and J. A. Sarnat. 2007. "Multipollutant Modeling Issues
23 in a Study of Ambient Air Quality and Emergency Department Visits in Atlanta." Journal of
24 Exposure Science and Environmental Epidemiology, 17 Suppl 2, S29-35.
25 U.S. Center for Disease Control. 2010. "Centers for Disease Control and Prevention, Behavioral Risk
26 Factor Surveillance System (BRFSS), 2010, Table "Table Cl Adult Self-Reported Current
27 Asthma Prevalence Rate (Percent) and Prevalence (Number) by State or Territory." Available at:
28 .
29 U.S. Environmental Protection Agency. 2001. Risk Assessment Guidance for Superfund. Vol. Ill, Part A.
30 Process for Conducting Probabilistic Risk Assessment (RAGS 3A). Washington, DC: EPA. (EPA
31 document number EPA 540-R-02-002; OSWER 9285.7-45; PB2002 963302). Available at:
32 .
33 U.S. EPA. 2004. EPA 's Risk Assessment Process for Air Toxics: History and Overview. In: Air Toxics
34 Risk Assessment Reference Library, Technical Resource Manual, Vol. 1., pp. 3-1 - 3-30. (EPA
35 document number EPA-453-K-04-001A). Washington, DC: EPA. Available at:
36 .
37 U.S. EPA. 2007. O3 Health Risk Assessment for Selected Urban Areas. Research Triangle Park, NC: EPA
38 Office of Air Quality Planning and Standards. (EPA document number EPA 452/R-07-009).
39 Available at: .
40 U.S. EPA. 2011. O3 National Ambient Air Quality Standards: Scope and Methods Plan for Health Risk
41 and Exposure Assessment. Research Triangle Park, NC: EPA. (EPA document number EPA-
42 452/P-l 1-001).
7-88
-------
1 U.S. EPA. 2012. Health Risk and Exposure Assessment for O3 First External Review Draft. Research
2 Triangle Park, NC: U.S. Environmental Protection Agency, Research Triangle Park, NC. (EPA
3 document number EPA 452/P-12-001).
4 U.S. EPA. 2013a. Integrated Science Assessment for O3 and Related Photochemical Oxidants: Final.
5 Research Triangle Park, NC: U.S. Environmental Protection Agency. (EPA document number
6 EPA/600/R-10/076F).
7 U.S. EPA. 2013b. Environmental Benefits Mapping and Analysis Program—Community Edition (Version
8 0.63), 2013a. Research Triangle Park, NC: U. S. Environmental Protection Agency. Available at:
9 .
10 World Health Organization. 2008. Part 1: Guidance Document on Characterizing and Communicating
11 Uncertainty in Exposure Assessment, Harmonization Project Document No. 6. Published under
12 joint sponsorship of the World Health Organization, the International Labour Organization and
13 the United Nations Environment Programme. WHO Press, World Health Organization, 20
14 Avenue Appia, 1211 Geneva 27, Switzerland (tel.: +41 22 791 2476).
15 Zanobetti, A; J. Schwartz. 2008. "Mortality Displacement in the Association of O3 with Mortality: An
16 Analysis of 48 Cities in the United States." American Journal of Respiratory and Critical Care
17 Medicine, 111: 184-189.
7-89
-------
1 8 NATIONAL SCALE MORTALITY RISK BURDEN BASED ON
2 APPLICATION OF RESULTS FROM EPIDEMIOLOGICAL STUDIES
3 As described in Chapter 2, the O3 ISA (U.S. EPA 2013) concluded that there is likely to
4 be a causal relationship between short-term 63 exposure and all-cause mortality and that there is
5 likely to be a causal relationship between long-term Os exposure and respiratory effects,
6 including respiratory mortality. Chapter 7 estimated health risks associated with recent 63
7 concentrations and meeting the current and alternative 63 standards in 12 selected urban study
8 areas. In this chapter we estimate nationwide premature mortality attributable to recent short-
9 term and long-term exposures to ambient 63 (Section 8.1); and assess the degree to which the
10 selected urban case study areas represent the full national distribution of risk-related attributes
11 and air quality dynamics (Section 8.2). Compared with the urban scale analysis in Chapter 7, this
12 analysis includes full spatial coverage across the U.S. but has less geographic specificity in the
13 concentration-response functions that are used to calculate Os.attributable mortality. The national
14 scale analysis is therefore intended as a complement to the urban scale analysis, providing both a
15 broader assessment of (Vrelated health risks across the U.S. as well as an evaluation of how
16 well the urban study areas examined in Chapter 7 represent the full distribution of Cb-related
17 health risks and air quality dynamics in the U.S.
18
19 8.1 NATIONAL-SCALE ASSESSMENT OF MORTALITY RELATED TO O3
20 EXPOSURE
21 This section estimates the total annual deaths for 2007 populations associated with
22 average 2006-2008 63 levels across the continental U.S. We first describe the methods and
23 inputs used to estimate Cb-attributable risk across the continental U.S., including OT, exposure
24 estimates, population and baseline mortality rate estimates, and epidemiologically derived 03-
25 mortality effect estimates. Results for the estimation of (Vattributable risk are then discussed in
26 terms of the magnitude and percent of total mortality attributable to Os exposure. We provide
27 two analyses to give perspective on the confidence in the estimates of Ch-related mortality: (1)
28 risk estimated only within the urban areas for which 63 mortality effect estimates are available;
29 and (2) the distribution of Os-related deaths across the range of observed 2006-2008 average 03
30 concentrations fused with modeled 2007 concentrations. These results are then synthesized and
31 compared with previous estimates of the burden of Os exposure on mortality in the U.S. from the
32 literature in a discussion section.
33
8-1
-------
1
2
3
4
5
6
1
8
9
10
11
12
13
14
15
16
Air Quality Inputs (Chapter 4)
National ambient ozone
12 km x12km
gridded spatial field
for recent year
sponse Functions
Identify
Relative Risk
(RR) or slope
coefficents (li)
I
Convert
RRto B
\
^f Nationwide set of city V
~~\ specificC-Rfunctions j~
•-—
Population Information
Daily county-specific
baseline health incidence
Population allocated
to 12 kmx12 km grid
BenMAP
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 V
[ estimates of % of total mortality I
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 (Vattributable risk. As shown in the conceptual diagram in Figure 8-1, we
conduct this analysis using the BenMAP software, which uses projections of the size and
geographic distribution of the potentially exposed population along with estimates of the ambient
63 concentrations to estimate (Vattributable 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 (Vattributable risk in the selected urban case 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 Os Concentrations
1 Air quality inputs to this analysis are described in detail in Chapter 4. In contrast to the
2 urban study areas analysis in Chapter 7, the national-scale analysis employs a data fusion
3 approach that takes advantage of the accuracy of monitor observations and the comprehensive
4 spatial information of the CMAQ modeling system to create national-scale "fused" spatial
5 surfaces of seasonal average Oi. Measured Os concentrations from 2006-2008 were fused with
6 modeled concentrations from a 2007 CMAQ model simulation, run for a 12 km domain covering
7 the contiguous U.S. In the first draft of the REA, the spatial surfaces were created using the
8 enhanced Voronoi Neighbor Averaging (eVNA) technique (Timin et al, 2010), using the EPA's
9 Model Attainment Test Software (MATS; Abt Associates, 201 Ob). In this draft, the spatial
10 surfaces are created using EPA's Downscaler software (Berrocal et al, 2012). More details on
11 the ambient measurements, the 2007 CMAQ model simulation, the Downscaler fusion technique,
12 and a technical justification for changing from eVNA to Downscaler can be found in Chapter 4.
13 Three "fused" spatial surfaces were created for: (1) the May-September mean of the 8-hr
14 daily maximum (consistent with the metric used by Smith et al. 2009); (2) the June-August mean
15 of the 8-hr daily mean from 10am to 6pm (consistent with the metric used by Zanobetti and
16 Schwartz 2008); and (3) the April-September mean of the 1-hr daily maximum (consistent with
17 the metric used by Jerrett et al. 2009) Os concentrations across the continental U.S. These fused
18 spatial surfaces each represent one seasonal average across 2006-2008, rather than three separate
19 years of concentrations. Section 4.3.2 presents maps, distributions, and statistical
20 characterizations of these O^ concentrations metrics across the U.S., including how they compare
21 to 2006-2008 design values.
22
8.1.1.2 Concentration-Response Functions
23 While Chapter 7 assessed both mortality and morbidity risks associated with Os
24 concentrations, due to limitations in baseline morbidity incidence rates, the national scale
25 assessment focuses on mortality risks only. To quantify the impact of 63 concentrations on
26 mortality, we apply risk estimates drawn from two major short-term epidemiological studies and
27 one long-term epidemiological study. These studies are consistent with those used in the analysis
28 of Os-related risk in selected urban areas (Section 7.2) and those mortality endpoints concluded
29 to have a causal or suggestive causal relationship with O^ exposure by the 2013 Integrated
30 Science Assessment for Oi and Related Photochemical Oxidants (U.S. EPA 2013).
31 For short-term mortality, we use city-specific and national average risk estimates drawn
32 from the Smith et al. (2009) study of Os and mortality in 98 U.S. urban communities between
33 1987 and 2000 as our main results, and the Zanobetti and Schwartz (2008) study of 63 and
8-3
-------
1 mortality in 48 U.S. cities between 1989 and 2000 as a sensitivity analysis, consistent with the
2 urban case study analysis in Chapter 7. City-specific effect estimates for both studies are
3 provided in Appendix 4-A.
4 Smith et al. (2009) found that the average non-accidental mortality increase across all 98
5 urban areas was 0.32% ± 0.08 (95% posterior interval [PI], 0.41%-0.86%) for a 10 ppb increase
6 in the 8-hr daily maximum Os concentration, based on April to October Os observations. Since
7 the national-scale analysis requires a single modeling period definition but some monitors only
8 collect data from May to September, the corresponding city-specific effect estimates are applied
9 to each day from May to September in BenMAP using May to September average 8-hr daily
10 maximum Os concentration based on 2006-2008 observed concentrations fused with 2007
11 modeled concentrations. The length of the Os season can affect the magnitude of mortality effect
12 estimates - a longer season may yield higher effect estimates per unit O?, concentration since Os
13 concentrations over the longer season may be lower than the Oi concentrations over the warmest
14 months only. Conversely, if the longer period captures periods of lower Os-related mortality
15 incidence, the effect estimates may be lower than effect estimates for the warmest months only.
16 Our application of the Smith et al. (2009) April to October effect estimates to May to September
17 Os concentrations likely introduces some bias in the results, but it is unclear in which direction.
18 Zanobetti and Schwartz (2008) found that the average total mortality increase across all
19 48 cities was 0.53% (95% confidence interval, 0.28%-0.77%) for a 10 ppb increase in June-
20 August 8-hr daily mean O?, concentration from 10 am to 6 pm, using a 0-3 day lag. We apply the
21 city-specific effect estimates that correspond to this national average effect estimate each day
22 from June to August in BenMAP using the June to August, mean 8-hr daily mean Os
23 concentration based on 2006-2008 observed concentrations fused with 2007 modeled
24 concentrations. Consistent with Chapter 7, these results are presented as a sensitivity analysis.
25 As in Chapter 7, we use city-specific risk estimates from the short-term epidemiology
26 studies, but apply them here only to the counties that were included in the epidemiology studies
27 rather than to the entire core-based statistical area (CBSA). Chapter 7 estimated risk across entire
28 CBS As to more completely capture expected O3 changes across broader areas and avoid bias
29 resulting from including only those areas where O3 is expected to increase under alternative
30 standards. The inclusion of the entire CBSA in that analysis required the application of a single
31 effect estimate to the entire CBSA. However, the national-scale assessment is a gridded analysis,
32 which allows greater spatial resolution in the application of effect estimates. In addition, eight
33 CBS As nationwide included multiple cities defined separately by Smith et al. (2009), some of
34 which showed considerable heterogeneity in effect estimates within the same CBSA.
35 Heterogeneity among effect estimates within a single CBSA implies that effect estimates from
36 one county may not be accurate representations of effect estimates in nearby
8-4
-------
1 counties. However, since city-specific effect estimates often have low power due to small
2 population size, we are unable to draw a strong conclusion regarding how well one county's
3 effect estimates represents those in nearby counties. For this national-scale assessment, we
4 apply effect estimates from each city as defined in the epidemiology studies to retain the full set
5 of information available from those studies. In addition, for counties not included by the
6 epidemiology studies, we apply the average effect estimate derived from all the urban areas
7 included in each of the studies ("national average") as it takes advantage of a wider and more
8 diverse population.
9 Since both national average estimates from these studies are based on urban areas only,
10 we have higher confidence in their application to other U.S. urban areas than to rural areas. To
11 demonstrate the magnitude of the results for which we have the highest confidence, we present
12 the percentage of estimated deaths occurring within the urban areas included in the
13 epidemiological studies and within all urban areas across the U.S. Lower confidence in the
14 results for rural areas does not indicate that the mortality risk among populations living in such
15 areas is unaffected by 63 pollution. Rather, the level of understanding for the (Vmortality
16 relationship in these areas is simply lower due to a lack of available epidemiological data at these
17 levels. We also examine the effect of varying the effect estimate applied between the cities
18 included by the epidemiology studies in a sensitivity analysis.
19 We quantify long-term Os-related respiratory mortality in this REA since the Integrated
20 Science Assessment for O^ and Related Photochemical Oxidants (Os ISA) concluded that the
21 evidence supports a likely to be causal relationship between long-term 63 exposure and
22 respiratory effects, including respiratory morbidity and respiratory-related mortality (U.S. EPA,
23 2013). As detailed in Chapter 7, we quantify long-term Os-related mortality using the respiratory
24 mortality effect estimates from the Jerrett et al. (2009) two-pollutant model that controlled for
25 PM2.5 concentrations, applied to each gridcell across the entire United States. This model found
26 that a 10 ppb increase in the April-September average of the 1-hr daily maximum 63
27 concentration was associated with a 4% (95% confidence interval, 1.0%-6.7%) increase in
28 respiratory mortality.
8.1.1.3 Demographic Inputs
29 This analysis uses the same baseline mortality rates and population estimates as were
30 used in the urban case study area analysis in Chapter 7. We derive baseline incidence rates for
31 mortality by age, cause, and county from the CDC Wonder database (CDC, 2004-2006). As this
32 database only provides baseline incidence rates in 5-year increments, we use data for the year
33 2005, the closest year to the analysis year 2007 used for the population and air quality modeling.
8-5
-------
1 We use 2007 population because it matches both the year of the emissions inventory and
2 meteorology used for the air quality modeling.
3 The starting point for estimating the size and demographics of the potentially exposed
4 population is the 2010 census-block level population, which BenMAP aggregates up to the same
5 grid resolution as the air quality model. BenMAP back-casts this 2010 population to the analysis
6 year of 2007 using county-level growth factors based on economic projections (Woods and
7 Poole Inc., 2012).
8
9 8.1.2 Results
10 Table 8.1 summarizes the estimated Cb-related premature mortality associated with 2006-
11 2008 average O^ concentrations under various assumptions for the health impact function.
12 Applying Smith et al. (2009) effect estimates for May-September, we estimate 15,000 (95% CI,
13 1,400-28,000) premature Os-related non-accidental deaths annually for 2007. As a sensitivity
14 analysis, we apply Zanobetti and Schwartz (2008) effect estimates for June-August, finding
15 16,000 (95% CI, 6,000-25,000) premature O3-related all-cause deaths annually for 2007. Figure
16 8.2 Figure 8.4 show that estimated Os-related mortality is most concentrated in highly populated
17 counties or those counties with urban areas found to have high effect estimates by Smith et al.
18 (2009) or Zanobetti and Schwartz (2008). For the application of Jerrett et al. (2009) national
19 average effect estimate for April-September, we estimate 45,000 (95% CI, 17,000-70,000)
20 premature Os-related respiratory deaths among adults age 30 and older.
21 Because the epidemiological studies included only selected urban areas, we are more
22 confident in the magnitude of the estimated Ch-related deaths occurring within those urban areas.
23 As shown in Table 8.1, approximately 43% of the (Vrelated deaths estimated using Smith et al.
24 (2009) effect estimates occur in the 98 urban locations included in that study, and 30% of the O^-
25 related deaths estimated using Zanobetti and Schwartz (2008) effect estimates occur in the 48
26 urban areas included in that study. We are also more confident in extrapolating the national
27 average effect estimates to other urban areas than we are to rural areas, as the national average
28 estimates are based on all urban areas included by the study. To estimate the percentage of total
29 Os-attributable deaths occurring within all urban areas across the continental U.S., we sum the
30 results for the 12km gridcells that have a total population greater than 12,000 (approximately
31 equal to the 95th percentile of gridcell populations across the continental U.S.). The percentage of
32 Os-attributable deaths occurring within urban areas defined in this way is 65% for results based
33 on Smith et al. (2009) effect estimates and 64% for results based on Zanobetti and Schwartz
34 (2008) effect estimates. While our confidence is lower when the national average effect estimates
8-6
-------
1
2
3
4
5
are extrapolated to rural areas, less certainty in the magnitude of Os-related deaths in rural areas
does not imply that 63 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 (95th percentile confidence interval)
Source of risk estimate and modeling period
Smith et al. (2009), May-September
95% confidence interval
% occurring within the 98 cities
Zanobetti and Schwartz (2008), June-August
95% confidence interval
% occurring within the 48 cities
Jerrett et al. (2009), April-September
95% confidence interval
Exposure
duration
Short-term
Short-term
Long-term
Age
>0
>0
>30
years
City-specific
effect
estimates1
15,000
(1,400-28,000)
43%
16,000
(6,000-25,000)
30%
-
National
average effect
estimate2
16,000
(7,200-22,000)
15,000
(8,300-22,000)
45,000
(17,000-70,000)
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
City-specific effect estimates are applied to the gridcells 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 gridcells. For the application of Smith et al. (2009) effect estimates, city-specific effect
estimates were applied to 2,227 gridcells and the national average to 44,064 gridcells. For the application of
Zanobetti and Schwartz (2008) effect estimates, city-specific effect estimates were applied to 925 gridcells and
the national average to 45,366 gridcells.
2 National average effect estimates are based on the average of all cities included in the epidemiological studies
applied to all 12km gridcells nationally.
Table 8.1 also shows Cb-related deaths estimated by applying the national average risk
estimate from the epidemiological studies to all gridcells in the U.S. Compared with applying
city-specific effect estimates to the gridcells corresponding to each urban area, using the national
average effect estimate for all gridcells 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, 2.5 percentile and 97.5 percentile of the estimated
percentage of mortality attributable to ambient Os across all counties in the U.S. Using Smith et
al. (2009) effect estimates, (Vattributable 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
8-7
-------
1 and Schwartz (2008) effect estimates, Os-attributable mortality contributes an average of 2.5%
2 (95% confidence interval, 1.7%-3.0%) to county-level June-August all-cause mortality (all ages)
3 and 0.6% (0.4%-0.8%) to all year all-cause mortality (all ages). For the results using Jerrett et al.
4 (2009) effect estimates, Os-attributable mortality contributes an average of 18.5% (95%
5 confidence interval, 15.2%-21.5%) to county-level April-September adult (age 30+) respiratory
6 mortality and 1.9% (1.3%-2.6%) to all year all-cause mortality (all ages). Figure 8.5 through
7 Figure 8.7 show that the counties with the highest percentage of mortality attributable to 63 are
8 typically those with the highest 63 levels.
9 Figure 8.8 displays the cumulative distribution of the percent of county-level all-cause,
10 all-age, and all-year mortality attributable to ambient 63 using effect estimates from all three
11 epidemiological studies. For the results based on Smith et al. (2009) and Zanobetti and Schwartz
12 (2008) effect estimates, 0.8% of all-cause, all-age, and all-year mortality is attributable to 03 for
13 approximately 99% of U.S. counties. For the results based on Jerrett et al. (2009) effect
14 estimates, 2.8% of all-cause, all-age, and all-year mortality is attributable to OT, for
15 approximately 99% of U.S. counties.
16
------- An error occurred while trying to OCR this image.
-------
1
2
S**
Figure 8.2 Estimated annual non-accidental premature deaths (individuals) in
2007 associated with average 2006-2008 May-September average 8-hr daily
maximum Os levels by county using Smith et al. (2009) effect estimates
5
6
7
8
Figure 8.3 Estimated annual all-cause premature deaths (individuals) in 2007
associated with average 2006-2008 June-August average 8-hr daily mean (10am-
6pm) Os levels by county using Zanobetti and Schwartz (2008) effect estimates
8-10
-------
I
2
3
4
5
#*
Figure 8.4 Estimated annual adult (age 30+) respiratory premature deaths
(individuals) in 2007 associated with average 2006-2008 April-September
average 1-hr daily max Os levels by county using Jerrett et al. (2009) effect
estimates
c*
-------
i
2
3
4
Figure 8.6 Estimated percentage of June-August total all-cause mortality (all ages)
attributable to 2006-2008 average Os levels by county using Zanobetti and Schwartz
(2008) effect estimates
6
7
8
9
10
11
Figure 8.7 Estimated percentage of April-September respiratory mortality among
adults age 30+ attributable to 2006-2008 average Os levels by county using Jerrett et
al. (2009) effect estimates
8-12
------- An error occurred while trying to OCR this image.
-------
1 Figure 8.9 shows the cumulative distribution of the county-level percent of total Os-
2 related deaths by 63 concentration. The mortality results based on Smith et al. (2009)
3 concentration-response functions are compared with the May-September average of the 8-hr
4 daily maximum O^ concentration, those based on Zanobetti and Schwartz (2008) concentration-
5 response functions are compared with the June-August average of the 8-hr mean 63
6 concentration from 10am to 6pm, and those based on Jerrett et al. (2009) concentration-response
7 functions are compared with the April-September average of the 1-hr daily maximum 63
8 concentration, consistent with the 63 concentration metrics used in each study. The mortality
9 results based on Zanobetti and Schwartz (2008) effect estimates are shifted to the right of the
10 mortality results based on the Smith et al. (2009) concentration response functions because the
11 seasonal averaging time for the results based on Zanobetti and Schwartz (2008) is limited to the
12 summer months when O^ tends to be highest. Similarly, the mortality results based on Jerrett et
13 al. (2009) effect estimates are shifted to the right of the mortality results based on Zanobetti and
14 Schwartz (2008) and Smith et al. (2009) because Jerrett et al. (2009) results use the seasonal
15 average of the 1-hr daily maximum, which tends to be higher than the seasonal average of 8-hr
16 daily maximum and seasonal average of 8-hr daily mean metrics (see Figure 4-18). For all three
17 epidemiology studies, we find that 90-95% of Os-related deaths occur in locations where the
18 May to September average 8-hr daily maximum, June to August average 8-hr daily mean (10am-
19 6pm), or April to September average 1-hr daily maximum Os concentrations are greater than 40
20 ppb. A seasonal average concentration of 40 ppb corresponds to 2006-2008 design values
21 ranging from approximately 50 to 90 ppb, depending on the seasonal average concentration
22 metric (see Figure 4-19).
23
24
8-14
------- An error occurred while trying to OCR this image.
------- An error occurred while trying to OCR this image.
-------
1 8.10. Estimated Os-attributable deaths using the regional prior city-specific effect estimates and
2 the regional average effect estimates between the 98 cities included by Smith et al. (2009) are
3 approximately 20% larger than the main results, with 38% of estimated deaths occurring in the
4 98 cities rather than 43%. The 95% confidence interval for the results using the regional prior
5 spans zero, whereas the 95% confidence interval for the results using the national prior does not.
6 Since the regional average effect estimates are all based on fewer data points (in some regions,
7 the regional average is based on only seven cities; see Appendix 8-A) than is the national
8 average, the confidence interval for each regional average effect estimate is large and sometimes
9 spans zero. The large confidence intervals for the regional average effect estimates drive the
10 confidence interval that spans zero for (Vattributable mortality estimated using regional prior
11 effect estimates. Confidence intervals that span zero do not imply that higher Os is associated
12 with decreased mortality, as there is no biologically plausible mechanism for such an effect, and
13 in no case do we see a significant negative central estimate. Rather, confidence intervals
14 spanning zero indicate a lack of statistical power to precisely determine the magnitude of an
15 effect.
16 Figure 8.11 shows estimated (Vattributable deaths by region using the national average
17 prior compared with using the regional average priors from Smith et al. (2009). Results generally
18 follow conclusions made by Smith et al. (2009) based on the magnitude of the regional effect
19 estimates. For example, using the national average effect estimate may substantially
20 underestimate Os-attributable deaths in the North East and Industrial Midwest where regional
21 effect estimates are large. Using the national average effect estimate may also overestimate 63-
22 attributable deaths in the Upper Midwest, Southern California, and South West, which were
23 found to have small effect estimates. However, these three regions have very large confidence
24 intervals which all span zero, since these regional averages are based on few cities (7, 7, and 9,
25 respectively, compared with 26 in the South East, 19 in Industrial Midwest, 16 in North East, and
26 12 in North West; see Appendix 8-A).
27
8-17
-------
1
2
3
4
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 in the main results.
Risk estimate
Os-attributable premature
deaths
Percent Os-
attributable deaths in
98 cities
City-specific, national prior
with national average
City-specific, regional prior
with regional averages
15,000
(1,400-28,300)
18,000
(-2,000 - 24,000)
43%
38%
5
6
7
9
10
11
12
13
Northwest
Southern
California
Upper
Midwest
Industrial
Midwest
Northeast
Southwest
Figure 8.10 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).
8-18
-------
i
2
3
4
5
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 Os-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 in the main results
7 8.1.4 Discussion
8 We estimated the total all-cause deaths associated with short-term exposure to recent Os
9 levels across the continental U.S., using average 2006-2008 observations from the Os monitoring
10 network fused with a 2007 CMAQ simulation and city-specific (Vmortality effect estimates
1 1 from two short-term epidemiology studies. Applying Smith et al. (2009) effect estimates for
12 May-September, we estimate 15,000 (95% CI, 1,400-28,000) premature O3-related non-
13 accidental deaths (all ages) annually for 2007. Using Smith et al. (2009) effect estimates, Os-
14 attributable mortality contributes an average of 1.5% (95% confidence interval, 1 . 1%-1 .8%) to
15 county-level May- September non-accidental mortality (all ages) and 0.6% (0.4%-0.7%) to all
16 year all-cause mortality (all ages). As a sensitivity, we apply Zanobetti and Schwartz (2008)
17 effect estimates for June- August, finding 16,000 (95% CI, 6,000-25,000) premature O3-related
18 all-cause deaths (all ages) annually for 2007. For results using Zanobetti and Schwartz (2008)
19 effect estimates, Os-attributable mortality contributes an average of 2.5% (95% confidence
20 interval, 1.7%-3.0%) to county-level June-August all-cause mortality (all ages) and 0.6% (0.4%-
21 0.8%) to all year all-cause mortality (all ages). For the application of Jerrett et al. (2009) effect
22 estimates for April-September, we estimate 45,000 (95% CI, 17,000-70,000) premature O3-
23 related adult (age 30 and older) respiratory deaths. For the results using Jerrett et al. (2009) effect
8-19
-------
1 estimates, Os-attributable mortality contributes an average of 18.5% (95% confidence interval,
2 15.2%-21.5%) to county-level April-September adult (age 30+) respiratory mortality and 1.9%
3 (1.3%-2.6%) to all year all-cause mortality (all ages). For all three epidemiology studies, we find
4 that 90-95% of Os-related deaths occur in locations where the May to September average 8-hr
5 daily maximum, June to August average 8-hr daily mean (10am-6pm), or April to September
6 average 1-hr daily maximum Os concentrations are greater than 40 ppb. A seasonal average
7 concentration of 40 ppb corresponds to 2006-2008 design values ranging from approximately 50
8 to 90 ppb, depending on the seasonal average concentration metric.
9 A previous analysis estimated that short-term 03 exposure was associated with 4,700
10 (95% CI, 1,800-7,500) premature deaths nationwide annually, based on 2005 Oj, concentrations
11 and Bell et al. (2004) national average effect estimates (Fann et al., 2012). The results estimated
12 here are higher, resulting mainly from two important differences. First, Fann et al. (2012)
13 estimated risk only above North American background, simulated 63 concentrations in the
14 absence of North American anthropogenic emissions, which was set to 22 ppb in the east and 30
15 ppb in the west. Fann et al. (2012) also used a national average mortality effect estimate for 8-hr
16 daily maximum 63 during the warm season only, calculated using ratios of 24-hr mean
17 concentrations to 8-hr daily maximum concentrations (see Abt Associates 2010). The Smith et
18 al. (2009) national average beta used here, 0.000322, is based on April-October 63 data and is
19 approximately 23% larger than that used by Fann et al. (2012), 0.000261. Since the risk
20 modeling period (and the seasonal definition for the seasonal average 8-hr daily maximum
21 concentration) was May to September for both studies, the higher beta used here yields a larger
22 Os mortality estimate. These two differences in methods explain the larger OT, mortality estimates
23 of this analysis compared with the previous estimate by Fann et al. (2012).
24 Estimated (Vattributable premature deaths based on Jerrett et al. (2009) effect estimates
25 are approximately three times larger than results based on Smith et al. (2009) and Zanobetti and
26 Schwartz (2008) effect estimates. The mean estimated county-level percent of all-cause, all-year,
27 and all-age mortality is also three times larger for results based on Jerrett et al. (2009) effect
28 estimates, indicating that the larger estimate does not simply result from a longer modeling
29 period or different population subset (e.g. adult respiratory disease for Jerrett et al. (2009) effect
30 estimates versus all-age non-accidental or all-cause mortality for Smith et al. (2009) and
31 Zanobetti and Schwartz (2008) effect estimates). Recent studies using long-term (Vmortality
32 relationships found by Jerrett et al. (2009) to quantify the burden of mortality due to
33 anthropogenic O?, globally (Anenberg et al. 2010, 2011) and for the U.S. specifically (Fann et al.
34 2012) have also found that using Jerrett et al. (2009) long-term effect estimates yields Ch-related
35 mortality burden estimates that are approximately two to four times larger than estimates based
36 on short-term effect estimates. Since long-term mortality relationships include both acute and
8-20
-------
1 chronic exposure effects, the significantly larger mortality estimates calculated using long-term
2 concentration-mortality relationships suggest that considering only short-term mortality may
3 exclude a substantial portion of Cb-related risk. However, since the short-term mortality
4 relationships include a larger population (all ages versus adults ages 30 and older only) and all
5 mortality causes, the short-term mortality relationships may capture some 63 effects that are not
6 captured by Jerrett et al. (2009). It is likely that some portion of the estimated premature deaths
7 attributable to short-term 63 exposure is captured by estimated premature deaths attributable to
8 long-term 63 exposure, but the extent of the overlap between these estimates is unknown.
10 8.2 EVALUATING THE REPRESENTATIVENESS OF THE URBAN STUDY AREAS
11 IN THE NATIONAL CONTEXT
12 To further support interpretation of risk estimates generated in Section 7.2, this section
13 presents three analyses that assess the representativeness of the 12 urban study areas in the
14 national context. First, we assess the degree to which the urban study areas represent the range of
15 air quality levels and key 63 risk-related attributes that vary spatially across the nation. We have
16 partially addressed this issue by selecting urban study areas in different geographical regions of
17 the country (see Section 7.2). In this section, we evaluate how well the selected urban areas
18 represent the overall U.S. for a set of spatially-distributed 63 risk related variables (e.g. weather,
19 demographics including socioeconomic status, baseline health incidence rates; Section 8.2.1).
20 Section 8.2.2 identifies where our 12 urban study areas fall along the distribution of 03-
21 attributable mortality risk across the U.S. This analysis allows us to assess the degree to which
22 the 12 urban study areas capture locations within the U.S. likely to experience elevated levels of
23 risk related to ambient 63. Finally, we give a national context to the estimated 63 responses to
24 emission changes in the urban study areas by assessing how well these 12 areas and the 3
25 additional exposure areas represent air quality trends and responses to emissions across the entire
26 U.S. (Section 8.2.3).
27 We do not attempt to assess the representativeness of the 15 urban study areas considered
28 in the exposure assessment for O^ related risk because data limitations preclude us from being
29 able to characterize individual-level exposure across the U.S. However, the urban study areas
30 considered in both the exposure and risk assessments shared common selection criteria,
31 including consideration of 63 concentrations, availability of adequate monitoring data,
32 demographics, and exposure factors. Therefore, conclusions from this analysis of the
33 representativeness of the 12 urban study areas for risk would also apply to those areas for
34 exposure.
8-21
-------
1 8.2.1 Analysis Based on Consideration of National Distributions of Risk-Related
2 Attributes
3 This section evaluates how well the urban study areas reflect national-level variability in
4 a series of 63 risk-related variables. For this analysis, we first generate distributions for risk-
5 related variables across the U.S. and for the specific urban study areas considered in Section 7.2
6 from generally available data (e.g. from the 2000 Census, Centers for Disease Control (CDC), or
7 other sources). We then plot the specific values of these variables for the selected urban study
8 areas on these distributions, and evaluate how representative the selected study areas are of the
9 national distributions for these individual variables.
10 Estimates of risk (either relative or absolute, e.g. number of cases) within our risk
11 assessment framework are based on four elements: population, baseline incidence rates, air
12 quality, and the coefficient relating air quality and the health outcome (i.e. the 63 effect
13 estimates). Each of these elements can contribute to heterogeneity in risk across urban locations,
14 and each is variable across locations. In addition, there may be other identifiable factors that
15 contribute to the variability of the four elements across locations. In this assessment, we examine
16 the representativeness of the selected urban area locations for the four main elements, as well as
17 factors that have been identified as influential in determining the magnitude of the C-R function
18 across locations.
19 While personal exposure is not incorporated directly into Os epidemiology studies, city-
20 specific 63 effect estimates are affected by differing levels of exposure which in turn are related
21 to variability in exposure determinants. The correlation between monitored Os and personal Os
22 exposure also varies between cities. The O?, ISA has comprehensively reviewed epidemiological
23 and toxicological studies to identify variables which may affect the 63 effect estimates used in
24 the city-specific risk analysis in Section 7.2 and the national-scale risk analysis in Section 8.1
25 (U.S. EPA 2013). Determinants of the 63 effect estimates used in risk assessment can be grouped
26 into four broad areas:
27 • Demographics: education, income, age, unemployment rates, race, body mass index and
28 physical conditioning, public transportation use, and time spent outdoors.
29 • Baseline health conditions: asthma, chronic obstructive pulmonary disease,
30 cardiovascular disease (atherosclerosis, congestive heart disease, atrial fibrillation,
31 stroke), diabetes, inflammatory diseases, and smoking prevalence.
32 • Climate and air quality: O?, levels, co-pollutant levels (annual mean PM^.s), temperatures
33 (days above 90 degrees, mean summer temp, 98th percentile temp).
34 • Exposure determinants: air conditioning prevalence.
35 Although data limitations preclude our ability to conduct a national-scale exposure assessment as
36 we have done for (Vattributable risk in Section 8.1, we assess the representativeness of the
8-22
-------
1 urban study areas across the national distribution of climate, air quality, and air conditioning
2 prevalence, factors which influence individual exposure. As discussed in detail in Chapter 5, no
3 available data base is sufficient to assess the national representativeness of time spent outdoors,
4 another important personal exposure determinant, among persons residing in each of the urban
5 case study areas. However, previous analyses suggest that children's time spent outdoors varies
6 little across U.S. regions (section 8.10.2 of U.S. EPA, 2009). In addition, as discussed in Section
7 5.1.1, time spent outdoors and the percent of person-days having at least one minute outdoors
8 (participation rate) does not appear to vary much over the past few decades based on analyses
9 using the CHAD database, nor does there appear to be a temporal trend over the past decade
10 based on analyses using the American Time Use Survey (ATUS). In considering that many of
11 the activity pattern studies in CHAD were from national surveys conducted in metropolitan areas
12 and that the evaluation results indicate little difference in time expenditure over broad
13 geographic areas and survey collection years, it is likely that the distribution of time spent
14 outdoors generated for the simulated persons in the 15 urban study areas (Chapter 5) reasonably
15 reflects the most important elements of a national distribution of time spent outdoors.
16 Based on these identified potential risk determinants, we identify datasets that could be
17 used to generate nationally representative distributions for each parameter. We are not able to
18 identify readily available national datasets for all variables. In these cases, if we are able to
19 identify a broad enough dataset covering a large enough portion of the U.S., we use that dataset
20 to generate the parameter distribution. In addition, we are not able to find exact matches for all of
21 the variables identified through our review of the literature. In cases where an exact match is not
22 available, we identify proxy variables to serve as surrogates. For each parameter, we report the
23 source of the dataset, its degree of coverage, and whether it is a direct measure of the parameter
24 or a proxy measure (Table 8.5). Summary statistics for the most relevant variables are provided
25 in Table 8.6.
26 Figure 8.12 through Figure 8.18 show the cumulative distribution functions (CDF)
27 plotted for the nation for the four critical risk function elements (population, air quality, baseline
28 incidence, and the O^ effect estimate), as well as where the urban study areas fall on the
29 distribution. While the urban-scale analysis in Chapter 7 includes the full core-based statistical
30 area for the selected cities, we consider here only the counties included in each city as defined by
31 the epidemiological studies, since we only have information on Oj, effect estimates for these
32 counties. This approach is consistent with the national-scale assessment of (Vattributable risk in
33 Section 8.1, from which we draw county-level Os-attributable risk estimates for the
34 representativeness analysis in Section 8.2.3. These figures focus on critical variables representing
35 each type of risk determinant, e.g. we focus on all-cause and non-accidental mortality rates, but
36 we also have conducted analyses for cardiovascular and respiratory mortality separately. The
8-23
-------
1 vertical black lines in each graph show the values of the variables for the individual urban study
2 areas. The city-specific values that comprise the national CDF for mortality risks found by
3 Zanobetti and Schwartz (2008) are also displayed on the graphs of those attributes, as the number
4 of cities included in that study is smaller (48 cities). The complete set of analyses is provided in
5 Appendix 4-A.
6 These figures show that the selected urban study areas represent the upper percentiles of
7 the distributions of population and do not represent the locations with lower populations (urban
8 study areas are all above the 90th percentile of U.S. county populations). This is consistent with
9 the objectives of our case study selection process, e.g. we are characterizing risk in areas that are
10 likely to be experiencing excess risk due to 63 levels above alternative standards. The urban
11 study areas span the full range of seasonal average 8-hr daily maximum Os concentrations in
12 monitored U.S. counties and the full distribution of Os risk coefficients across the cities included
13 by Smith et al. (2009) and Zanobetti and Schwartz (2008). The urban study area analysis
14 includes the two cities with the highest risk coefficients found by Smith et al. (2009) - New York
15 City and Philadelphia - as well as the two highest found by Zanobetti and Schwartz (2008) -
16 New York City and Detroit. In Table 8.6, respiratory and cardiovascular mortality have higher
17 concentration-response relationships than non-accidental and all-cause mortality because they
18 are based on a smaller baseline population and are the diseases most affected by 63 exposure.
19 The urban study areas do not capture the upper end of the distribution of baseline mortality,
20 including all-cause (Figure 8.15) and non-accidental mortality (Figure 8.16), as well as
21 cardiovascular and respiratory mortality (see Appendix 8-B). The interpretation of this is that the
22 case study risk estimates may not capture the additional risk that may exist in locations that have
23 the highest baseline mortality rates.
8-24
-------
Table - 8.5 Data Sources for Os risk-related Attributes
Potential risk
determinant
Metric
Year
Source
Degree of
national
coverage
Demographics
Age
Age
Age
Education
Unemployment
Income
Race
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
2005 County Characteristics, 2000-2007 Inter- All counties
university Consortium for Political and
Social Research
2005 County Characteristics, 2000-2007 Inter- All counties
university Consortium for Political and
Social Research
2005 County Characteristics, 2000-2007 Inter- All counties
university Consortium for Political and
Social Research
2000 USDA/ERS, All counties
http://www.ers.usda.gov/Data/Education/
2005 County Characteristics, 2000-2007 Inter- All counties
university Consortium for Political and
Social Research
2005 County Characteristics, 2000-2007 Inter- All counties
university Consortium for Political and
Social Research
2006 County Characteristics, 2000-2007 Inter- All counties
university Consortium for Political and
Social Research
8-25
-------
Potential risk
determinant
Metric
Year
Source
Degree of
national
coverage
Population
Total population
2008
Population density Population/square mile
2008
Urbanicity
ERS Classification Code
2003
Cumulative Estimates of Resident All counties
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 All counties
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- All counties
university Consortium for Political and
Social Research
Climate and Air Quality
63 levels
63 levels
03 levels
Monitored 4th high 8-hr
daily maximum
Seasonal mean 8-hr daily
maximum
Seasonal mean 1-hr daily
2007
EPA Air Quality System (AQS)
Avg. 2006-2008 AQS
Avg. 2006-2008 AQS
725 Monitored
counties
671 Monitored
counties
671 Monitored
8-26
-------
Potential risk
determinant Metric
maximum
63 levels Seasonal mean
PM2.5 levels Monitored annual mean
Temperature Mean July temp
Relative Humidity Mean July RH
Year Source
Avg. 2006-2008 AQS
2007 AQS
1 94 1 - 1 970 County Characteristics, 2000-2007 Inter-
university Consortium for Political and
Social Research
1 94 1 - 1 970 County Characteristics, 2000-2007 Inter-
university Consortium for Political and
Social Research
Degree of
national
coverage
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 All Cause
Baseline mortality Non Accidental
Baseline mortality Cardiovascular
Baseline mortality Respiratory
Baseline morbidity Acute myocardial
infarction prevalence
CDC Wonder 1999-2005
CDC Wonder 1999-2006
CDC Wonder 1999-2007
CDC Wonder 1999-2008
2007 Behavioral Risk Factor Surveillance System
(BRFSS)
All counties
All counties
All counties
All counties
1 84 metropolitan
statistical areas
8-27
-------
Potential risk
determinant
Baseline morbidity
Baseline morbidity
Baseline morbidity
Obesity
Level of exercise
Level of exercise
Respiratory risk
factors
Smoking
Metric
Diabetes prevalence
Stroke prevalence
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
2007
2007
Source
BRFSS
BRFSS
BRFSS
BRFSS
BRFSS
BRFSS
BRFSS
BRFSS
Degree of
national
coverage
(MSA)
184 MS A
184 MS A
184 MS A
184 MS A
184 MS A
184 MS A
184 MS A
184 MS A
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
8-28
-------
Table - 8.6 Summary Statistics for Selected O3 Risk-related Attributes
Risk 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 AMI (%) *
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
8-29
-------
Risk 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
Os 4th high maximum 8-hr average
(ppm)
Os seasonal mean (ppb)
Os seasonal mean of maximum 8-hr
average (ppb)
Os seasonal mean of 1-hr daily
maximum (ppb)
PM25 annual mean (ug/m3)
PM2 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
8-30
-------
* Attribute for which only city-specific data were available.
8-31
-------
1
2
100%
Urban case study areas
are all above the 90th
percentile of county
populations
1000
10000 100000
Population, 2008
1000000
10000000
•All Counties CDF
Case Study Counties
4 Figure 8.12 Comparison of county-level populations of urban case study area counties to
5 the frequency distribution of population in 3,143 U.S. counties.
8-32
-------
1
2
3
4
5
1/1
.0)
C
3
O
u
0)
O
'E
O
^
'o
*
J.UU/O
90%
80%
70%
60%
50%
40%
30%
20%
10%
no/
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 mean 8-hr daily maximum Os
concentrations in urban case study area counties to the frequency distribution
of seasonal mean 8-hr daily maximum Os concentrations in 671 U.S. counties
with Os monitors.
8-33
-------
1
2
3
4
5
40
50 60 70 80 90 100
4th High 8-hr Daily Maximum Ozone, 2007 (ppb)
110
•All Counties CDF
Case Study Counties
Figure 8.14 Comparison of 2007 county-level 4 high 8-hr daily maximum Os
concentrations in urban case study area counties to the frequency
distribution of 2007 4th high 8-hr daily maximum Os concentrations in 725
U.S. counties with Ch monitors.
8-34
-------
1
2
3
4
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, 1999-2005
1500
•All Counties CDF
Case Study Counties
Figure 8.15 Comparison of county-level all-cause mortality in urban case study area
counties to the frequency distribution of all-cause mortality in 3,137 U.S.
counties.
8-35
-------
1
2
3
4
300
Urban case study
counties are all
below the 80th
percentile of non
accidental
mortality
500 700 900 1100 1300
Non Accidental Mortality per 100,000 Population, 1999-2005
1500
•All Counties CDF
Case Study Counties
Figure 8.16 Comparison of county-level non-accidental mortality in urban case study area
counties to the frequency distribution of non-accidental mortality in 3,135
U.S. counties.
8-36
-------
0%
0.0002
0.0004 0.0006 0.0008 0.001
All Cause Mortality Risk Coefficient ((J)
0.0012
All cities
•All Cities CDF
Case Study Cities
2 Figure 8.17 Comparison of city-level all-cause mortality risk coefficients from Zanobetti
3 and Schwartz (2008) in urban case study areas to the frequency distribution
4 of all-cause mortality risk coefficients from Zanobetti and Schwartz (2008) in
5 48 U.S. cities.
8-37
-------
100%
0.0002 0.0004 0.0006 0.0008
Non Accidental Mortality Risk Coefficient ((J)
0.001
•All Cities CDF
Case Study Cities
2 Figure 8.18 Comparison of city-level national prior Bayes-shrunken non-accidental
3 mortality risk coefficients from Smith et al. (2009) in urban case study areas
4 to the frequency distribution of national prior Bayes-shrunken non-accidental
5 mortality risk coefficients from Smith et al. (2009) in 98 U.S. cities.
6
7 Figure 8.19 through Figure 8.24 show national CDFs and the urban study area values for
8 several selected potential risk attributes. These potential risk attributes do not directly enter the
9 risk equations, but have been identified in the literature as potentially affecting the magnitude of
10 the 63 C-R functions reported in the epidemiological literature. Comparison graphs for other risk
11 attributes are provided in Appendix 4-A. The selected urban study areas do not capture the
12 higher end percentiles of several risk characteristics, including populations 65 years and older,
13 baseline cardiovascular disease prevalence, baseline respiratory disease prevalence, and smoking
14 prevalence. Summarizing the analyses of the other risk attributes, we conclude that the urban
15 study areas provide adequate coverage across population, population density, 63 levels (seasonal
16 mean, seasonal mean 8-hr daily maximum, and seasonal mean 1-hr daily maximum), PM2.5 co-
17 pollutant levels, temperature and relative humidity, unemployment rates, percent non-white
18 population, asthma prevalence obesity prevalence, income, and less than high school education.
8-38
-------
1 We also conclude that while the urban study areas cover a wide portion of the distributions, they
2 do not provide coverage for the upper end of the distributions of percent of population 65 and
3 older (below 60th percentile), percent of population 85 years and older (below 75* percentile),
4 prevalence of angina/coronary heart disease (below 70th percentile), prevalence of diabetes
5 (below 85th percentile), stroke prevalence (below 90th percentile), prevalence of heart attack
6 (below 80th percentile), prevalence of smoking (below 85th percentile), all-cause mortality rates
7 (below 85th percentile), non-accidental mortality rates (below 80th percentile), cardiovascular
8 mortality rates (below 75th percentile) and respiratory mortality rates (below 50th percentile), and
9 percent of residences without air conditioning (below 90* percentile). In addition, the urban
10 study areas do not capture the highest or lowest ends of the distribution of exercise prevalence
11 and do not capture the low end of the distribution of public transportation use (above the 65*
12 percentile).
-Hi
13
14
15
16
16
18 20 22 24
% Younger than 15 Years Old, 2005
26
28
30
•All Counties CDF
Case Study Counties
Figure 8.19 Comparison of county-level percent of population 0 to 14 years old in urban
case study area counties to the frequency distribution of percent of population
0 to 14 years old in 3,141 U.S. counties.
8-39
-------
1
2
3
4
Urban case study
counties are all
below the 60th
percentile of county
%of population 65
years and older
11 13 15 17 19 21
% 65 Years and Older, 2005
23 25
•All Counties CDF
Case Study Counties
27
Figure 8.20 Comparison of county-level percent of population age 65 years old and older
in urban case study area counties to the frequency distribution of percent of
population age 65 and older in 3,141 U.S. counties.
8-40
-------
100%
90%
80%
.2 70%
4-1
§ 60%
i/5 50%
% 40%
° 30%
20%
10%
0%
10000 20000 30000 40000 50000 60000 70000 80000 90000 100000
Income per capita, 2005 ($)
•All Counties CDF
Case Study Counties
2 Figure 8.21 Comparison of county-level income per capita in urban case study areas to
3 the frequency distribution of income per capita in 3,141 U.S. counties.
8-41
-------
66
68
70 72 74 76 78 80 82
Mean Temperature for July, 1941-1970 (F)
84
86
•All Counties CDF
Case Study Counties
2 Figure 8.22 Comparison of county-level July temperature in urban case study area
3 counties to the frequency distribution of July temperature in all U.S. counties.
8-42
-------
Comparison of Urban Case Study Area with U.S. Distribution (184 BFRSS
Cities) - Asthma Prevalence
8 10 12
Asthma Prevalence, 2007 (%)
14
•All Cities CDF
Case Study Cities
2 Figure 8.23 Comparison of city-level asthma prevalence in urban case study areas to the
3 frequency distribution of asthma prevalence in 184 U.S. cities.
8-43
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Urban study areas
are all below the
90th percentile of
percent of
residences with no
air conditioning
10
20
30 40 50 60 70
No air conditioning, 2004 (%)
80
90
100
•All Cities CDF
Case Study Cities
Figure 8.24 Comparison of city-level air conditioning prevalence in urban case 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 case studies. First, the case studies represent urban areas that are
among the most populated in the U.S. Second, they represent areas with relatively high levels of
Os (4* high 8-hr daily maximum, seasonal mean 8-hr daily maximum, seasonal mean 1-hr daily
maximum, 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 03. 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
8-44
-------
1 of certain urban areas with higher percentages of older populations, for example, cities in
2 Florida, may lead to underrepresentation of high risk populations. However, with the exception
3 of areas in Florida, most locations with high percentages of older populations have low overall
4 populations, less than 50,000 people in a county. And even in Florida, the counties with the
5 highest 63 levels do not have a high percent of older populations. This suggests that while the
6 risk per exposed person per ppb of Os may be higher in these locations, the overall risk to the
7 population is likely to be within the range of risks represented by the urban case study locations.
8 Due to data limitations, we were only able to assess the representativeness of the urban
9 study areas in terms of one exposure-related attribute, air conditioning prevalence. Assessing the
10 representativeness of the urban study areas in terms of air conditioning prevalence, we found that
11 the urban study areas do not capture the highest end of percent of residences without air
12 conditioning. If the cities with the lowest air conditioning prevalence also have high O^ levels,
13 we could be missing a high risk portion of the population that is exposed to 63 indoors as air
14 infiltrates indoors from outdoors. However, 4* highest 8-hr daily maximum Os levels in the
15 cities in the top 10th percentile of percentage of residences without air conditioning (mainly in
16 northern California and Washington) are approximately average (0.08 ppm) or lower than
17 average. Since these concentrations are not the highest found across the U.S., we are likely not
18 excluding a high risk population that has both low air conditioning prevalence and high 63
19 concentrations, and the overall risk to the population is likely to be within the range of risks
20 represented by the urban case study locations.
21
22 8.2.2 Analysis Based on Consideration of National Distribution of Os-Related Mortality
23 Risk
24 In this section we discuss the second representativeness analysis which identifies where
25 the 12 urban study areas examined in Chapter 7 fall along the distribution of estimated national-
26 scale mortality risk. This assessment reveals whether the baseline Os mortality risks in the 12
27 urban case study areas represent more typical or higher end risk relative to the national risk
28 distribution presented in Section 8.1. For consistency, we compare the national OT, mortality risk
29 distribution to the O^ mortality risk results for the urban study areas that were generated from the
30 national-scale assessment in Section 8.1, rather than the results from the urban study area
31 analysis in Chapter 7 which uses different methods. To be consistent with the national-scale
32 assessment, we define the urban study areas here as they were defined in the epidemiology
33 studies, rather than including full core-based statistical areas as in Chapter 7. The results of this
34 representativeness analysis are presented graphically in Figure 8.25 through Figure 8.27, which
35 display the cumulative distribution of total mortality attributable to ambient 63 at the county
36 level developed as part of the national-scale analysis. Values for the 23 counties included in the
8-45
-------
1 urban case study areas as defined in the epidemiology studies are then superimposed on top of
2 the cumulative distribution to assess the representativeness of the urban case study areas.
3 For the results based on Smith et al. (2009) effect estimates, New York City and
4 Philadelphia have the highest percentage of May-September non-accidental mortality attributable
5 to ambient 63 of the 12 urban study areas and are located at the highest end of the distribution of
6 U.S. Os-related mortality risk (Figure 8.25). Of the 12 urban study areas, these two cities had the
7 highest effect estimates found by Smith et al. (2009; See Appendix 4-A). Boston and Los
8 Angeles had the lowest (Vrelated mortality risk of the 12 urban study areas and are located at
9 the lowest end of the U.S. distribution. Overall, Os mortality risk in the 12 urban study areas are
10 representative of the full distribution of U.S. Os-related mortality risk, with the mean percentage
11 of May-September non-accidental mortality for all ages of 1.5% (95% confidence interval, 1.1-
12 1.8%).
13 For the results based on Zanobetti and Schwartz (2008) effect estimates, Detroit and New
14 York City are at the very highest end of the U.S. distribution of county-level risk of June-August
15 all-cause mortality due to ambient Os (Figure 8.26). These two cities had the highest effect
16 estimates of the 48 cities included in the study (see Appendix 4-A). The high effect estimates in
17 Detroit and New York City could be due to high rates of public transportation use (for New York
18 City), low air conditioning prevalence, high smoking prevalence (in Detroit), high incidence of
19 mortality and other adverse health outcomes (e.g. diabetes, stroke, acute myocardial infarction,
20 etc.), and high unemployment rates. Houston and Los Angeles had the lowest risk and were
21 located at the very lowest end of the U.S. distribution of county-level risk of mortality due to
22 ambient Os. These two cities had the lowest effect estimates found by Zanobetti and Schwartz
23 (2008), possibly because they cover a large spatial extent and have high rates of time spent
24 driving, which could lead to exposure misclassification in the underlying epidemiologic study.
25 Houston also has a very high rate of air conditioning use (nearly 100% of residences) and Los
26 Angeles has been shown to have high rates of adaptive behavior on high ambient Os days (i.e.
27 more time spent indoors as a result of high ambient Os concentrations; Neidell 2009, 2010), both
28 of which would lead to lower personal O^ exposure relative to other cities. Overall, O^ mortality
29 risk in the 12 urban study areas are representative of the full distribution of U.S. Os-related
30 mortality risk, with the mean percentage of June-August all-cause mortality for all ages of 2.5%
31 (95% confidence interval, 1.7-3.0%).
32 For the results based on Jerrett et al. (2009) effect estimates, the 12 urban study areas are
33 centered more in the middle of the distribution of U.S. county-level risk of adult (ages 30 and
34 older) respiratory mortality due to ambient Os exposure. These results are based on the
35 application of a single national average effect estimate to all gridcells across the U.S., rather than
36 city-specific effect estimates as were applied for the results based on Smith et al. (2009) and
8-46
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
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 63 than these four cities. Overall, 63 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 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 (Vrelated risk
based on Jerrett et al. (2009) effect estimates in the 12 urban study areas.
100%
0%
0.5 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 study area
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.
8-47
-------
1/1
Q)
S
u
^
0)
00
ro
0)
u
QJ
Q.
0)
>
S
ro
3
E
3
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
1 1.5 2 2.5 3 3.5 4 4.5
Percentage of baseline mortality attributable to ozone
1
2
3
4
5
0.5
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.
8-48
-------
1
2
3
4
5
6
Figure 8.27
14 16 18 20 22 24 26 28
Percentage of baseline mortality attributable to ozone
^—Results based on Jerrett et al. (2009) effect estimate
• Selected urban study areas
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 8.2.3 Analysis Based on Consideration of National Responsiveness of Os Concentrations
9 to Emissions Changes
10 Estimates of Os response to precursor emissions reductions (NOx and VOC) are
11 important inputs to estimation of risk for scenarios of just meeting existing and alternative 63
12 standards. To evaluate the national representativeness of Os responses to decreases in precursor
13 emissions in the 15 urban study areas, we examine two different types of air quality data. In
14 section 8.2.3.1 we examine ambient 63 trends that have been measured at monitor locations
15 across the country over a recent period of decreasing NOx emissions. This analysis provides
16 real-world observations but does not isolate the effects of emissions changes alone and can only
17 characterize past phenomena. In section 8.2.3.2, we look at air quality model predictions of
18 temporal and spatial patterns of Os changes in response to further NOx reductions from 2007
8-49
-------
1 levels. This analysis is subject to typical model limitations but has the advantage of isolating the
2 effects of precursor emissions changes and has the ability to simulate how 63 would change in
3 response to NOx (and VOC) emissions reductions (relative to recent 2007 levels) similar to those
4 used in the HDDM adjustment scenarios for just meeting existing and alternative standards.
5 These two complimentary analyses give qualitatively similar results, building confidence that the
6 overarching conclusions are robust across the US as a whole.
8.2.3.1 Ambient patterns in trends of measured Os concentrations
7 This section describes how annual distributions of Os measurements collected by EPA's
8 national monitoring network have changed between 1998 and 2011. These years were chosen
9 because large reductions in anthropogenic NOx emissions have occurred over this time period
10 especially in the Eastern half of the United States. From 2000 to 2011 nationwide NOx emissions
11 were cut almost in half (from 22.6 to 12.9 million tons per year )3. However, it should be noted
12 that these reductions did not occur uniformly across the country. Improvements in vehicle
13 emissions standards helped reduce NOx emissions in many locations throughout the country. In
14 contrast, EPA rules like the NOx SIP call were focused on controlling emissions from power
15 plants in the Eastern US and consequently there have been relatively larger reductions in NOx
16 emissions in the East. In addition, some urban areas which have traditionally had high Oj levels,
17 like Los Angeles and Houston, have substantially cut local NOx and VOC emissions to improve
18 their air quality. Also, there may be some localized areas in which NOx emissions have
19 increased due to population growth, new sources such as oil and gas development, or increased
20 wildfire activity. Appendix 8-C provides plots of emissions trends by region of the U.S. These
21 plots show that each of nine regions of the U.S. have experienced decreasing NOx emissions
22 ranging from approximately a 20% decrease to a 45% decrease from 2002 to 2011 depending on
23 the region. Conversely, VOC emissions have increased in some regions since 2002 (the South,
24 the Southwest, and the West-North-Central) and decreased in others. Due to non-linear O3
25 formation chemistry and the potential for changes in local chemical regimes resulting from these
26 emissions reductions, past trends may not reflect the ambient changes which will occur from
27 future emissions reductions. Nonetheless, these ambient data provide information on actual Oj
28 changes in response to emissions reductions and can give insight into the types of changes in Os
29 that have occurred both within and outside the urban study areas.
30 First, we look at national maps which show changes in 50th percentile and 95th percentile
31 summer season (April-October)4 8-hour daily maximum Oj, values (Figure 8.28,Figure 8.29).
3 Data were accessed from EPA's emission trend website on August 15, 2013:
http://www.epa.gov/ttn/chief/trends/trends06/national_tierl_caps.xlsx
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
8-50
-------
1 These maps reflect the absolute (ppb) difference between Os percentiles from two three-year
2 periods (2001-2003 and 2008-2010)5. Figure 8.28 shows that increases in median O3
3 concentrations occurred in many large urban areas including both study area locations in Chapter
4 76 and non-study area locations7. Only a few monitors with increasing median Os appear outside
5 of cities, most notably in southwestern Colorado and central Kansas. The increases in urban
6 areas are likely explained by Os "disbenefits" to NOx reductions which were described in
7 Chapter 4, Appendix 4-C and in the following section of this chapter. Widespread decreases of
8 median Os in suburban and rural locations suggest the efficacy of large NOx emissions
9 reductions on reducing Os over large regions of the country. Finally, the less frequently observed
10 cases of median Os increases in rural areas are likely caused by different phenomena. Cooper et
11 al. (2012) suggested that increasing rural Os in the Western US may be due to increasing oil and
12 gas development, wildfires and Os transport from Asia. Conversely, Figure 8.29 shows that 95*
13 percentile 63 values for these two sets of years decrease in almost all urban as well as rural areas
14 of the country. Only a few sites in Colorado, Nevada, and California show any increases in 95*
15 percentile Os between 2001-2003 and 2008-2010. The consistent decreases across most of the
16 United States indicate that the large NOx reductions from power plants and mobile sources have
17 been quite successful in reducing Os on the highest Os days. These results suggest that many of
18 the urban case study areas may show Os responses that are typical of other large urban areas in
19 the U.S. However, decreasing Os in large non-urban portions of the country may not be fully
20 captured in the urban case studies.
use the April-October time period in Chapter 4 for composite monitor distributions to summarize ozone values
relevant to the epidemiology-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-hour ozone standard and
the 2008-2010 period was used to designate areas for the 2008 8-hour ozone 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
8-51
-------
Change in April - October Median Daily Maximum 8-hour Ozone Concentration from 2001 - 2003 to 2008 - 2010
1
2
Increase 10+ppb
•"• Increase 5 ppb
° No Change
T Decrease 5 ppb
T Decrease 10+ppb
Figure 8.28 Change in 50th percentile summer season (April-October) daily 8-hr maximum
O3 concentrations between 2001-2003 and 2008-2010.
8-52
-------
Change in April - October 95th Percentile Daily Maximum S-hour Ozone Concentration from 2001 - 2003 to 2008 - 2010
Increase 10+ppb
•"• Increase 5 ppb
° No Change
T Decrease 5 ppb
T Decrease 10+ppb
2 Figure 8.29 Change in 95 percentile summer season (April-October) daily 8-hr
3 maximum O3 concentrations between 2001-2003 and 2008-2010.
4 To examine these trends further, we evaluate the 1998-2011 data from the 15 case-study
5 areas. Only monitors within the 15 study areas8 were analyzed, and within each study area,
6 monitors were put into three groups based on the degree of urbanization. The degrees of
7 urbanization were determined by the population density of the census tract containing the
8 monitor (plotted in Figure 8.30). Population data were obtained from the U.S. Census Bureau9,
9 and the classes were determined by breaks in the population density calculated from those data:
r\
10 "high population density" (> 1000 people/km ), "medium population density" (between 400 and
9 9
11 1000 people/km ), and "low population density" (< 400 people/km ). Data were additionally split
12 out into three different time periods (all months, warm months: May through September, and
13 cool months: October through April). These warm and cool season categorizations were chosen
14 to isolate effects that are observed at different times of year. The April-October time period
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
8-53
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
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.
Pop Dens
(#/km2)
H >=1000
800
600
400
200
0
Figure 8.30 Population density at each Os monitor.
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
th
-th
lines bordering the dark and light red ribbons in this plot are (from top to bottom) the 95 , 75 ,
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 63 distribution (not just discrete cut points of 5th, 25th, 50th, 75th,
and 95* percentiles) are provided in Appendix 8-C.
These plots show consistent trends over the past 13 years for 63, with high 63 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
8-54
-------
1 low population density areas. Mid-range O?, concentrations at low population density locations
2 within the case study areas (so still relatively close to a major city) have generally decreased over
3 a period of substantial NOx emissions reductions10. This decrease is most pronounced in the
4 summer months and in the Eastern half of the U.S. (low population density monitors in 3 out of
5 the 5 Western11 case-study areas and in only 4 out of the 10 Eastern12 case study areas do not
6 have significant decreases in summertime median Os concentrations). Mid-range Os
7 concentrations in many, but not all, high population density areas have significantly increased in
8 winter months. Wintertime increases were significant in 11 of the 15 areas (only Atlanta, Boston,
9 Houston and Sacramento did not increase significantly). Thirteen out of 15 summertime high
10 population density area trends in median Oj, were not significant13, but combining winter and
11 summer measurements to determine annual trends showed that Denver, Los Angeles, New York
12 and Philadelphia high population density sites had significantly increasing annual median O?,
13 while Boston, Chicago, Dallas and St. Louis had significantly increasing 25th percentile 63 but
14 no significant median trend. These results reflect increasing mid-range Os concentrations mainly
15 confined to urban centers during periods of NOx reductions. One important point to note is that
16 the design value monitor (the monitor with the highest average (over three years) of 4th highest
17 daily maximum value) in most of the case-study locations is located outside of the high
18 population density area (as defined here). Downward trends in medium and low population
19 density areas are therefore generally representative of the behavior at the highest Os monitor in
20 an area, whereas trends in urban centers may be important from an exposure perspective.
21 In summary, any increasing Oj trends occur more in highly populated areas, during cool
22 months, and at the lower end of the Os distribution. Conversely, any decreasing Os trends occur
23 more during warm months, in lower population areas, and at the upper end of the Os distribution.
24 One result of these two phenomena is a narrowing of the range of Oi concentrations over this
25 period of decreasing NOx emissions. For instance, there are many cases where the top and
26 bottom of a single distribution exhibit different trends. For example, the low population density
27 monitors of Dallas, Los Angeles, Philadelphia and Saint Louis and the high population density
28 monitors for Baltimore, Dallas, and Philadelphia for all months show a significant increase in the
29 5th percentile and a simultaneous significant decrease in the 95th percentile. More common is a
10 Denver is an outlier among the case study areas with consistently increasing mid-range ozone trends across
seasons and urban classifications. Denver may be subject to increasing emissions from large wildfires and oil
and gas 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 ozone
8-55
-------
1
2
3
4
5
6
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.
Chicago Cleveland
Houston Los Angeles New York Philadelphia Sacramento Saini Louis Washington
Trend
Chars
— Signif:Neg
Signif:Pos
InsignitNeg
InsignifrPos
Year
8 Figure 8.31 Distributions of Os concentrations for high population density monitors by
9 different subsets of months over a 13-year period. From top to bottom in each
10 ribbon plot, the blue and white lines indicate the spatial mean of the 95th, 75th,
11 50th, 25th, and 5th percentiles for each monitor for every year from 1998-2011.
8-56
-------
Atlanta Boston Chicago Cleveland Dallas Denver DetroH Houston Los Angeles Mew York Philadelphia Sacramento Saint Louis
Trend
Chars
— Signif:Neg
Signif:Pos
InsignifrNeg
lnsignif:Pos
.JIM
i i i
Year
2
3
4
5
6
7
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
th
each ribbon plot, the blue and white lines indicate the spatial mean of the 95 ,
75th, 50th, 25 , and 5th percentiles for each monitor for every year from 1998-
2011.
8-57
-------
Atlanta Baltimore Boston Chicago Cleveland Dallas Denver Detroit Houston Los Angeles New York Philadelphia Sacramento Sain) Louis Washington
| I
iillh
Trend
Chars
— Signif:Neg
f Signif:Pos
InsignifrNeg
lnsignif:Pos
2 Figure 8.33 Distributions of Os concentrations for low population density monitors by
3 different subsets of months over a 13-year period. From top to bottom in each
4 ribbon plot, the blue and white lines indicate the spatial mean of the 95th, 75th,
5 50th, 25 , and 5th percentiles for each monitor for every year from 1998-2011.
6
7 Maps of ambient trends in both New York City and Chicago most clearly show these
8 trends and further illustrate this behavior. Figures 8.34 and 8.35 show trends in daily maximum
9 8-hour 63 values these two cities for May-September. Plots for other case-study areas are
10 provided in Appendix 8-C. For both cities, the fourth highest 8-hr daily maximum 63 value
11 either has a downward trend or no trend at all monitors. In New York (Figure 8.34), mean and
12 median 63 values significantly decrease at downwind locations in New York and Connecticut.
13 Conversely, median Os values significantly increase from 1998 to 2011 at two core urban sites
14 (one at City College of NY in upper Manhattan and one near Queen's college) and at a nearby
15 site on Long Island. Similarly, in Chicago (Figure 8.35), mean and median trends in 63 are
16 downward or insignificant in Indiana and in suburban Illinois locations and show increases near
17 the highly populated urban core.
18
8-58
-------
— ^.Patersonoiry /Y/rirSTs—
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 8-hr daily
maximum Os values, center panel shows trends in annual mean 8-hr daily maximum Os values, and right
panel shows trends in annual median 8-hr daily maximum Os values.
8-59
-------
Max4
Mean
Median
Napervjjle
/ \ Hammond;;—-»Ga
Naperville
/ \ V \7 ^
I \ Hammond^—;
>-*-Joliet
/
/
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 8-hr daily
maximum Os values, center panel shows trends in annual mean 8-hr daily maximum Os values, and right
panel shows trends in annual median 8-hr daily maximum 03 values.
8-60
-------
1
2
O
4
5
6
7
8
9
10
11
12
13
To demonstrate how changes in emissions of NOx and anthropogenic VOCs might be
driving these trends, Table 8-7 shows trends of 63 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 Regions15. There is moderate correspondence between the decreases in NOx
emissions across the regions with the observed decreases in warm season 63 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 8-C.
Table - 8.7 Broad Regional Annual Trends of Concurrent Os Concentrations and
Emissions of NOx and VOCs over the 2000-2011 Time Period
Trend
High Pop Dens, May2Sep O:!
High Pop Dens. Oct2Apr O3
Low Pop Dens. May2S(>p O:(
Low Pop Dens, Oct2Apr O.i
NOj. Emission
VOO Emission
Central
none
up
flowrn
none
down
none
East North
Central
none
up
down
none
down
none
North East
none
up
flown
top %'s up
down
flown
South
down
low %'s up
flown
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
-------
1 "across the board" emissions cuts at four distinct levels and do not represent the exact adjustment
2 cases that were used to estimate 63 concentrations consistent with individual case study areas
3 just meeting various potential levels of the NAAQS standard. However, these four cases
4 generally span the range of emissions perturbations that were applied in the HDDM adjustment
5 methodology described in Chapter 4 and in Appendix 4d.
6 In this analysis we focus on seasonal mean Os and population as proxies for
7 epidemiology based risk estimates in Chapter 7. Since the epidemiology studies used in Chapter
8 7 show relatively linear response of health outcomes to 63 concentrations throughout the entire
9 range of measured O^ values, examining seasonal mean values should provide some
10 understanding of locations where 63 health effects are expected to increase and decrease as a
11 result of precursor emission reductions. By combining population information with these spatial
12 distributions of seasonal 03 responses, we can better understand expected 03 behavior in
13 locations where people live. This is not a detailed risk assessment but can provide information on
14 the representativeness of the case-study areas to the nation as a whole in terms of expected Os-
15 related health outcomes.
16 To begin, we examine maps displaying ratios of mean 63 concentrations in the emissions
17 cut simulations to mean O^ concentrations in the 2007 base simulation. Figure 8.36 and Figure
18 8.37 show the ratio of seasonal (April-October) mean Os in the two NOx emissions reduction
19 simulations to that in the base simulation for the entire model domain. Figure 8.38 and Figure
20 8.39 depict the ratio of January mean 03 for the two NOx cut simulations. Figures showing the
21 ratios based on the May-September seasonal average and figures for the NOx/VOC emissions
22 reductions scenarios are provided in Appendix 8-C. The maps show widespread decreases (i.e.,
23 ratios less than 1) in seasonal mean Oj, across the country. These decreases are especially
24 pronounced in the Eastern U.S. and in California. Oj increases (i.e., ratios greater than 1) are
25 confined to urban core areas except in January. The spatial extent of these 03 increases are
26 generally less for the 90% NOx cut simulation than for the 50% NOx cut simulation although the
27 magnitude is greater over very limited areas in Chicago, Seattle, and San Francisco. The Os
28 increases are most widespread in the cooler months (January, April, and October). For the April-
29 October seasonal average 03 concentrations, VOC in addition to NOx cuts did not substantially
30 change the ratios of Os in the emissions reduction scenarios to Os in the base scenario. In the
31 Northeast and Midwest, increases in seasonal mean Os concentrations were mainly confined to
32 urban case study areas of New York, Detroit, Chicago, and St. Louis. In the Southeastern U.S.,
33 the urban areas which show up as having increased seasonal mean 03 in the 50% NOx cut
34 simulations include Miami, Orlando, Tampa, and New Orleans (only Miami has Os increase in
35 the 90% NOx cut simulation). The only case-study area in the southeast, Atlanta, does not
36 experience increases in seasonal mean 03 in the model simulation (this is consistent with the
8-62
-------
1 changes in risk estimated for Atlanta, which show no increases in total risk as alternative
2 standards are simulated). In the central U.S., seasonal mean 63 in the case study areas of Denver,
3 Houston, and Dallas and non-case-study areas of San Antonio, Duluth, and Minneapolis
4 increased with 50% NOx reductions. Seasonal mean O^ increases were seen only in Houston,
5 Minneapolis, and Duluth with 90% reductions in simulated NOx emissions. The Northwestern
6 U.S. showed some of the most widespread increases in seasonal mean Os in the 50% and 90%
7 NOx cut simulations covering the Seattle and Portland metro areas as well the San Francisco Bay
8 area and in a single model grid cell for Sacramento (50% NOx reduction case only). Sacramento
9 is the only city in the Northwest that was included as a case study area. Finally, Los Angeles (a
10 case study area), San Diego, Phoenix, and Bakersfield were the areas for which CMAQ predicted
11 seasonal mean Os increases with the 50% NOx cut simulation. These Os increases disappeared
12 (or were largely diminished in the case of LA) in the 90% NOx cut case. Based on these maps, it
13 appears that in the Northeast and the Central U.S., the case-study area selection likely
14 oversampled these Os increases on a geographic basis since all locations outside of city centers
15 experienced decreasing seasonal mean Os with the NOx reduction model simulations. However
16 in two regions, the Southeast and the Northwest, the urban case study area did not experience
17 increases in seasonal mean O?, concentrations while other urban areas in the region did. In these
18 two regions, the urban case-study selection likely under-sampled the locations which
19 experienced increases in seasonal mean Os.
8-63
-------
T
0.4
0.6
0.8 1.0 1.2
ratio of seaonal mean ozone
1.4
1
2
3
4
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 O
concentrations in the 2007 base CMAQ simulation.
8-64
-------
-i
T
0.4
0.6
0.8 1.0 1.2
ratio of seaonal mean ozone
1.4
4
5
6
7
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.
8-65
-------
1
2
3
4
* U f
0.4
T
T
T
0.6
0.8 1.0 1.2
ratio of seaonal mean ozone
1.4
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.
8-66
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
T
T
T
0.4
0.6
0.8 1.0 1.2
ratio of seaonal mean ozone
1.4
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.
In order to characterize the representativeness of case study areas in a more quantitative
manner, paired Oj, concentrations and population data16 were extracted from each model grid cell
and categorized in various manners. 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 O^ (June-August), and the bottom panels show data for
seasonal mean Oj, (April-October). Tabulated results and equivalent plots for the combined
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.
8-67
-------
1 NOx/VOC cut simulations are provided in Appendix 8-C. Month by month break-outs for each
2 case study area are also available in Appendix 8-C.
3 The vast majority of the U.S. population lives in areas where the CMAQ simulations
4 predict mean O^ decreases for the June-August and April-October time periods. The majority of
5 population living in case-study areas also lives in locations with decreasing seasonal mean 63
6 concentration under NOx reduction scenarios. As discussed previously, more locations have
7 increasing mean 63 in the cooler months as demonstrated by the fact that almost all of the U.S.
8 population lives in locations where the model predicts increases in mean 63 in January. The case
9 study areas represent 29% of the total U.S. population. These areas account for 20-30% of the
10 U.S. population that experience decreasing seasonal mean 63 for April-October in the NOx cut
11 simulations and 50-60% of the U.S. population that experience increasing seasonal Os for April-
12 October. Consequently, the urban-case study areas over-sample populations living in locations
13 with increasing seasonal mean Oj in response to NOx cuts compared to populations living in
14 locations with decreases in seasonal mean 03. In all panels displayed in Figures 8.28 and 8.29,
15 most of the population lives in locations where increases or decreases in mean Oj, were greater
16 than 1 ppb.
17
8-68
-------
Non-Study Area
Study Area
40-
60-
c
g
OB
Q.
O
Q_
CO
o-
60-
O
§
cr
ppb change
2 Figure 8.40 Histograms of U.S. population living in locations with increasing and
3 decreasing mean Os. Values on the x-axis represent change in mean Os (ppb)
4 from the 2007 base CMAQ simulation to the 50% NOx cut CMAQ simulation.
5 The percentages of the U.S. population living in areas that have changes less
6 than -1 ppb, from -1 to +1 ppb, and greater than 1 ppb are shown on the y-
7 axis. Left plots show population numbers in locations not included in one of the
8 cases study areas while right plots show population numbers in locations
9 included in one of the case study areas. Top plots show changes in January
10 monthly mean Os, middle plots show changes in seasonal mean June-August
11 Os, and bottom plots show changes in seasonal mean April-October Os.
8-69
-------
Non-Study Area
Study Area
20-
60-
c
g
ZJ40 -
Q.
O
Q_
CO
'20
0-
60-
O
§
cr
1
2
3
4
5
6
7
8
9
10
11
12
13
ppb change
Figure 8.41 Histograms of U.S. population living in locations with increasing and
decreasing mean Os. 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 O.?, middle plots show changes in seasonal mean June-
August Os, and bottom plots show changes in seasonal mean April-October
03.
8-70
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
The proportion of the population living in locations of increasing seasonal mean O^ 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 case study areas
have between 5% and 30% of their populations living in these areas with increasing mean Oj,
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.
80%
10^
0%
^^=j
t
&
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.
8-71
-------
50.0%
45.0%
40.0%
35.0%
30.0%
25.0%
20.0%
15.0%
10.0%
5.0%
0.0%
1
2 Figure 8.43 Population (as % of total case-study area population) living in locations of
3 increasing April-October seasonal mean Os in the 90% NOx reduction
4 CMAQ simulation.
5
6 We can further understand these results by looking at them in terms of population density
7 in the case-study areas versus across the U.S. as a whole. As in Section 8.2.3.1, we define census
8 tracts with population density greater than 1000 people/km2 as high population density, but the
9 low-mid population density classification used here is a combination of the low and medium
10 classifications in that section. Figure 8.44 and Figure 8.45 split out the April-October results
11 from Figure 8.40 and Figure 8.41 into high and low-mid sub-categories. Appendix 8-C provides
12 similar breakouts for the other panels in Figure 8.40 and Figure 8.41. First, based on these
13 definitions, we see that 57% of the population in case-study areas lives in high population
14 density locations while only 27% of the U.S. population does. As discussed above, the high
15 population areas are more likely to experience increases in mean Os as a result of NOx emission
16 reductions compared to lower population areas. Therefore, the fact that the case-study areas used
17 in the risk and exposure assessments are more densely populated than the country as a whole
18 means that these analyses may estimate higher risks under emissions reduction scenarios than
19 would be experienced, on average, across the country. Figure 8.44 and Figure 8.45 show
20 generally similar shapes for the high population density histograms in the study-area and non-
21 study area locations. In the 50% NOx cut simulation, 69% of the population living in high
22 density case-study areas would experience increases in mean seasonal Oj compared to 63% of
23 the population the population living in high density areas of the country as a whole. Similarly in
24 the 90% NOx cut simulation, 28% of the population in high density locations both within the
25 study areas and across the U.S. as a whole lives in locations of increasing seasonal mean 03. This
8-72
-------
1 suggests that the selected study areas adequately represent population-weighted changes in mean
2 63 for people living in high density areas. Similarly, less densely populated locations within the
3 case-study areas show Os increases equivalent to those seen in less densely populated areas in
4 the U.S. as a whole. In the 50% NOx cut simulation, 7% of people in low-mid density study area
5 locations live where mean seasonal Oi is increasing, while 5% of people in all low-mid density
6 U.S. locations live where mean seasonal Os is increasing. Similarly, in the 90% NOx cut
7 simulation, the numbers are 2% for both low-mid density study area populations and for low-mid
8 density populations in the U.S. as a whole. Thus the oversampling of populations living in
9 locations of increasing mean seasonal O?, in response to NOx cuts, as shown in Figures 8.28 and
10 8.30, appears to be entirely due to the fact that the study areas oversample populations living in
11 high density areas compared to the U.S. population as a whole.
12
13
14
15
8-73
-------
Non-Study Area
Study Area
40-
"O
o
CD
20-
c
.2
Q.
O
Q_
CO
0-
•a
O
Of
ppb change
2 Figure 8.44 Histograms of U.S. population living in locations with increasing and
3 decreasing mean Os. Values on the x-axis represent the change in seasonal
4 mean (April-October) Os from the 2007 base CMAQ simulation to the 50%
5 NOx cut CMAQ simulation. The percentages of the U.S. population living in
6 areas that have changes less than -1 ppb, from -1 to +1 ppb, and greater than
7 1 ppb are shown on the y-axis. Left plots show population numbers in
8 locations not included in one of the cases study areas while right plots show
9 population numbers in locations included in one of the urban case study
10 areas. Bottom plots show histograms for low-mid population density areas
11 while top plots show histograms for high population density areas.
8-74
-------
Non-Study Area
Study Area
"O
o
(0
c
.2
3 0-
Q.
O
Q_ 60
CO
40-
Q.
3
•a
O
Of
1
2
3
4
5
6
7
8
9
10
11
12
ppb change
Figure 8.45 Histograms of U.S. population living in locations with increasing and
decreasing mean Os. 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. 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 case study
areas. Bottom plots show histograms for low-mid population density areas
while top plots show histograms for high population density areas.
8-75
-------
1 8.2.5 Discussion
2
3 We evaluated two different questions, 1) to what degree are the 15 cities evaluated in the
4 exposure and risk analyses representative of the overall U.S. population with regards to total O^
5 risk?, and 2) to what degree are they representative of the overall U.S. population with regards to
6 the degree of risk reduction that might be observed in response to just meeting the existing and
7 alternative standards.
8 Regarding the first question, we observe that the 12 urban study areas considered in the
9 urban-scale risk assessment presented in Section 7.2 capture urban areas that are among the most
10 populated in the U.S., have relatively high Oj, levels, and represent the range of city-specific
11 effect estimates found by Smith et al. (2009) and Zanobetti and Schwartz (2008). These three
12 factors suggest that the urban study areas capture overall risk for the nation well, with a potential
13 for better characterization of the high end of the risk distribution. We find that the urban study
14 areas are not capturing areas with the highest baseline mortality rates, those with the oldest
15 populations, and those with the lowest air conditioning prevalence. These areas tend to have
16 relatively low 63 concentrations and low total population, suggesting that the urban study areas
17 are not missing high risk populations that have high Os concentrations in addition to greater
18 susceptibility per unit 63. We also find that the 12 urban study areas represent the full range of
19 county-level (Vrelated risk across the entire U.S. We conclude from these analyses that the 12
20 urban study areas adequately represent Os-related risk across the U.S.
21 Concerning the second question, we observe that the 15 urban areas considered in the
22 exposure and risk assessment case study areas over-sample populations living in locations with
23 increasing seasonal mean O^ in response to NOx cuts. This suggests that the selected study areas
24 adequately represent population-weighted changes in mean 63 concentrations for urban
25 populations, but may be under-representing decreasing median Os concentrations in suburban
26 and rural areas. As a result, the risk estimates for populations in the selected urban study areas
27 may understate the risk reductions that might be achieved across the broader U.S. population.
28
29
8-76
-------
1
2 8.3 REFERENCES
3 Abt Associates, Inc. 2010. Model Attainment Test Software (Version 2), prepared for EPA.
4 Research Triangle Park, NC: EPA Office of Air and Radiation, Office of Air Quality
5 Planning and Standards. Research Triangle Park, NC. Available at:
6 .
7 Abt Associates, Inc. 2010. Environmental Benefits and Mapping Program (Version 4.0),
8 prepared for EPA. Research Triangle Park, NC: EPA Office of Air Quality Planning and
9 Standards. Available at .
10 • Anenberg, S.C.; JJ. West; L.W. Horowitz and D.Q. Tong. 2010. "An Estimate of the
11 Global Burden of Anthropogenic Ozone and Fine Particulate Matter on Premature
12 Human Mortality Using Atmospheric Modeling." Environmental Health Perspective.,
13 118:1189-1195.
14 Anenberg, S.C.; JJ. West; L.W. Horowitz and D.Q. Tong. 2011. "The Global Burden of Air
15 Pollution Mortality: Anenberg et al. respond." Environmental Health Perspective.,
16 119:A158-A425.
17 Bell, M.L.; A. McDermott; S.L. Zeger; J.M. Samet and F. Dominici. 2004. "Ozone and Short-
18 term Mortality in 95 U.S. Urban Communities, 1987-2000." Journal of the American
19 Medical Association, 292:2372-2378.
20 Byun, D. and K. L. Schere. 2006. "Review of Governing Equations, Computational Algorithms,
21 and Other Components of the Models-3 Community Multi-scale Air Quality (CMAQ)
22 Modeling System." Applied Mechanics Reviews, 59:51-77.
23 Centers for Disease Control: Wide-ranging Online Data for Epidemiological Research (CDC-
24 Wonder) (data from years 2004-2006), Centers for Disease Control and Prevention
25 (CDC), U.S. Department of Health and Human Services. Available on the Internet at
26 .
27 Cooper, O.R.; R. Gao; D. Tarasick; T. Leblanc and C. Sweeney. 2012. Long-term Ozone Trends
28 at Rural Ozone Monitoring Sites Across the United States, 1990-2010." Journal of
29 Geophysical Research, 117, D22, doi: 10.1029/2012 JDO18261.
8-77
-------
1 Fann N.; A. D. Lamson; S. C. Anenberg; K. Wesson; D. Risley; B. J. Hubbell. 2012.
2 "Estimating the National Public Health Burden Associated with Exposure to Ambient
3 PM2.5 and Ozone." Risk Analysis, 32:81-95.
4 George, BJ. and T. McCurdy. 2009. "Investigating the American Time Use Survey from an
5 exposure modeling perspective." Journal of Exposure Science and Environmental
6 Epidemiology, 21:92-105.
7 Graham, S. and T. McCurdy. 2004. "Developing Meaningful Cohorts for Human Exposure
8 Models." Journal of Exposure Analysis and Environmental Epidemiology, 14:23-43.
9 Hollman, F.W.; T. J. Mulder and J. E. Kalian. 2000. "Methodology and Assumptions for the
10 Population Projections of the United States: 1999 to 2100." Population Division Working
11 Paper No. 38, Population Projections Branch, Population Division, U.S. Census Bureau,
12 Department of Commerce.
13 Jerrett, M.; R.T. Burnett; C. A. Pope, III; K. Ito; G. Thurston; D. Krewski; Y. Shi; E. Calle; M.
14 Thun. 2009. "Long-term 03 Exposure and Mortality." New England Journal of Medicine,
15 360:1085-1095.
16 Neidell, M. 2009. Information, avoidance behavior and health. J Human Res. 44:450-478.
17 Neidell, M. 2010. Air quality warnings and outdoor activities: evidence from Southern California
18 using a regression discontinuity approach design. J Epidemiol Community Health.
19 64:921-926.
20 Samet, J. M.; S. L. Zeger; F. Dominici; F. Curriero; I. Coursac; D.W. Dockery; J. Schwartz and
21 A. Zanobetti. (2000). "The National Morbidity, Mortality, and Air Pollution Study Part
22 II: Morbidity and Mortality from Air Pollution in the United States." Health Effects
23 Institute, Boston, MA, Number 94, Part II.
24 Smith R.L.; B. Xu and P. Switzer. 2009. "Reassessing the Relationship Between Ozone and
25 Short-term Mortality in U.S. Urban Communities." Inhalation Toxicology, 21:37-61.
26 Timin, B.; K. Wesson and J. Thurman. "Application of Model and Ambient Data Fusion
27 Techniques to Predict Current and Future Year PM2.5 Concentrations in Unmonitored
28 Areas. (2010)," pp. 175-179 in Steyn DG, Rao St (eds). Air Pollution Modeling and Its
29 ApplicationXX. Netherlands: Springer.
8-78
-------
1 U.S. Environmental Protection Agency. 2013. Integrated Science Assessment for Ozone and
2 Related Photochemical Oxidants. Research Triangle Park, NC: EPA. (EPA document
3 number EPA 600/R-10/076F).
4 U.S. EPA. 2010. Quantitative Health Risk Assessment for Particulate Matter. Research Triangle
5 Park, NC: EPA. (EPA document number EPA-452/R-10-005).
6 U. S. EPA. 2009. Risk and Exposure Assessment to Support the Review of the SO 2 Primary
1 National Ambient Air Quality Standards: Final Report. Research Triangle Park, NC:
8 U. S. EPA. (EPA document number EPA-452/R-09-007).
9 U.S. EPA. 2007. Ozone Population Exposure Analysis for Selected Urban Areas. Research
10 Triangle Park, NC: EPA. (EPA document number EPA-452/R-07-010). Available at:
11 .
12 Woods and Poole Economics Inc. 2012. "Complete Demographic Database." Woods and Poole
13 Economics, Inc. Washington, DC. Available at:
14 .
15
16 Zanobetti, A. and J. Schwartz. 2008. "Mortality Displacement in the Association of Ozone with
17 Mortality: An Analysis of 48 Cities in the United States." American Journal of
18 Respiratory and Critical Care Medicine, 177:184-189.
19 Zanobetti, A. and J. Schwartz. 2011. "Ozone and Survival in Four Cohorts with Potentially
20 Predi sposing Di seases." American Journal of Respiratory and Critical Care Medicine.,
21 194:836-841.
22 Zhang, L.; D. J. Jacob; N.V. Smith-Downey; D. A. Wood; D. Blewitt; C. C. Carouge; A. van
23 Donkelaar; D. B. A. Jones; L.T. Murray and Y. Wang. 2011. "Improved Estimate of the
24 Policy-relevant Background Ozone in the United States Using the GEOS-Chem Global
25 Model with l/2°x2/3° Horizontal Resolution Over North America." Atmospheric
26 Environment,45:61'69-671'6.
27
8-79
-------
1 9 SYNTHESIS
2 9.1 INTRODUCTION
3 This assessment estimates exposures to Os and resulting mortality and morbidity health
4 risks based on the findings of the 63 ISA (U.S. EPA, 2013) that short-term 63 exposures are
5 causally related to respiratory effects, and likely causally related to cardiovascular effects, and
6 that long term 03 exposures are likely causally related to respiratory effects. The assessment
7 evaluates total exposures and risks associated with the full range of observed 63 concentrations,
8 as well as the incremental changes in exposures and risks between just meeting the existing
9 standard of 75 ppb and just meeting alternative standard levels of 70, 65, and 60 ppb using the
10 form and averaging time of the existing standard: the annual 4th highest daily maximum 8-hour
11 Os concentration, averaged over three consecutive years. We evaluated alternative standard
12 levels of 70, 65, and 60 consistent with recommendations from CASAC to consider alternative
13 standard levels between 60 and 70 ppb (Frey and Samet, 2012).
14 Following the conceptual framework described in Chapter 2, the assessment evaluates
15 exposures and lung function risk in 15 urban case study areas, and mortality and morbidity risks
16 based on concentration-response functions derived from epidemiology studies in 12 of these
17 urban case study areas1. The results from these assessments will help inform consideration of the
18 adequacy of the existing primary 63 standards, and potential risk reductions associated with
19 several alternative levels of the standard (for the current form and averaging time). In addition, to
20 place the urban case study area analyses in a broader context, Chapter 8 of this assessment
21 estimates the national burden of mortality associated with recent 63 levels, and evaluates the
22 representativeness of the 15 urban case study areas in characterizing 03 exposures and risks
23 across the U.S. This synthesis focuses on the urban case study area assessments of exposure and
24 risk for the scenarios of just meeting the existing and alternative standards. For this synthesis, we
25 discuss the results of the national-scale assessment as they relate to understanding the breadth of
26 63 risks across the U.S. and to the national representativeness of the urban case study area risk
27 results.
28 To facilitate interpretation of the results of the exposure and risk assessment, this chapter
29 provides a synthesis of the various results, focusing on comparing and contrasting those results
30 to identify common patterns, or important differences. These comparisons will focus on patterns
1 Three additional urban case 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 case 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
-------
1 across urban case study areas, across years of analysis, and across alternative standards. In
2 addition, factors related to each specific type of analysis that may influence comparisons
3 between the analyses are identified and discussed. The degree to which the integrated results are
4 representative of national patterns of exposure and risk is evaluated. Overall confidence in the
5 results, as well as relative confidence between the different analyses is also assessed. The chapter
6 concludes with an overall integrated characterization of exposure and risk in the context of key
7 policy-relevant questions raised in Chapter 2.
8 9.2 SUMMARY OF KEY RESULTS
9 9.2.1 Air Quality Considerations (Chapter 4)
10 Table 9-1 below gives information on the monitoring network, population, and observed
11 peak 63 concentrations for the 15 case study areas, for the years included in the exposure, lung
12 function risk, and epidemiology based risk assessments. The number of counties, number of Os
13 monitors, population, and design values (DV) are based on the area definitions used in the
14 exposure modeling and clinical-based lung function risk assessments, while the 2007 and 2009
15 annual 4* highest values are based on the Core Based Statistical Areas (CBSAs) used in the
16 epidemiology-based risk assessment. The "N/A" values in the 2007 and 2009 4th high columns
17 are for the three urban areas not included in the epidemiology-based risk assessment. The data
18 show a trend of lower peak Os concentrations (i.e., the 2008-2010 design values and 2009 4*
19 high values are generally lower than the 2006-2008 design values and 2007 4th high values,
20 respectively).
9-2
-------
1
2 Table 9-1 Area and Monitoring Information for the 15 Case Study Areas
Area Name
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
#of
Counties
33
7
10
16
8
11
13
9
10
5
27
15
7
17
26
#ofO3
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
high (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
high
(PPb)
77
83
75
N/A
72
N/A
79
73
91
108
81
74
96
74
N/A
4
5
6
7
8
9
10
11
12
13
14
15
16
In this analysis, we employed a photochemical model-based adjustment methodology
Simon et al. (2012) using the Higher-Order Decoupled Direct Method (HDDM) capabilities in
the Community Multi-scale Air Quality Model (CMAQ) (hereafter referred to as HDDM air
quality adjustment). The HDDM air quality adjustment methodology replaced the quadratic
rollback technique used in the first draft REA to estimate Os concentrations consistent with just
meeting existing and alternative 63 standards. The HDDM air quality adjustment procedure
estimates the change in observed hourly 63 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 63 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 case study areas for the 2006-2008 and 2008-2010 periods. In most locations, only NOx
reductions were used to adjust the distribution of 63 concentrations, because of the
ineffectiveness of VOC reductions in reducing peak Os concentrations needed to meet the
2 Attainment with the existing standard level of 75ppb is determined by the 4th highest maximum 8-hour O3
concentration, averaged over 3 years (hereafter referred to as the existing standard).
-------
1 existing and alternative standard levels. Sensitivity analyses were also conducted in some
2 locations to evaluate the impact of decreasing both NOx and VOC emissions.
3 The HDDM air quality adjustment methodology represents a substantial improvement
4 over the quadratic rollback method used to adjust Os concentrations in previous reviews. First,
5 quadratic rollback was a purely mathematical technique which attempted to reproduce the
6 distribution of observed Os concentrations just meeting various standards, while the new
7 methodology uses photochemical modeling to simulate the response in Os concentrations due to
8 changes in precursor emissions based on current understanding of atmospheric chemistry and
9 transport. Second, quadratic rollback used the same mathematical formula to adjust
10 concentrations at all monitors within each case study area for all hours, while HDDM allows the
11 adjustments to vary both spatially across each case study area and temporally across hours of the
12 day and across seasons. Finally, quadratic rollback was designed to only allow decreases in Os
13 concentrations, while the HDDM air quality adjustment allows both increases and decreases in
14 Os concentrations in response to reductions in NOx or VOC emissions. For example, in response
15 to reductions in NOx emissions, the HDDM methodology is able to capture increases in Os
16 concentrations that can occur in urban cores characterized by titration of Os by fresh NO
17 emissions and decreases in Os concentrations downwind.
18 Following HDDM adjustment of Os concentrations, several general trends are evident in
19 the changes in Os patterns across the case study areas and across the alternative standard levels.
20 In all 15 case study areas, peak Os concentrations tended to decrease while the Os concentrations
21 in the lower part of the distribution of Os tended to increase as the concentrations were adjusted
22 to meet the existing and alternative standards. In addition, Os concentrations in the high and mid-
23 range portions of the Os distribution generally decreased in the outer, more rural and suburban
24 portions of the urban case study areas, while the Os response to NOx reductions was more varied
25 within the urban cores. In particular, while the peak (annual 4* highest daily maximum 8-hour)
26 concentrations upon which the existing and alternative standards are defined generally decreased
27 in the urban core of the case study areas in response to modeled reductions in primarily NOx
28 emissions, the Os responses near the center of the Os distribution at these locations followed one
29 of three patterns when focusing on the mean of the daily maximum 8-hour Os concentrations
30 from May to September, as shown in Table 9-2.
31
9-4
-------
1
2
Table 9-2 General Patterns in Seasonal (May-Sept) Mean of Daily Maximum 8-hour
Concentrations after Adjusting to Meet Existing and Alternative Standards*
After Adjusting to Meet
Existing Standard
After Further Adjusting to Meet
Lower Alternative Standards
Case Study Areas
Showing Pattern
Decreased
Continued to decrease
Atlanta
Sacramento
Washington, D.C
Increased
Decreased
Baltimore
Cleveland,
Dallas
Detroit
Los Angeles
New York
Philadelphia
St. Louis
Increased
Continued to increase or remained
constant
Boston
Chicago
Denver
Houston
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
* These patterns refer to Os responses in the urban core of each urban case 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 63 concentrations at each census tract in the 15 case 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 REA and previous 63
NAAQS reviews (see Appendix 4-A for details). Consequently, the spatial variability of
observed and HDDM-adjusted 63 is better accounted for in these analyses compared to those in
the first draft REA.
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 case
study areas. Consequently, in cases of urban case study areas within which Os was predicted to
increase in some locations and decrease in others, the air quality inputs to this analysis represent
a "net" effect for each case study area. The spatial extent of the case study areas used in the
composite monitor averages were CBSAs. These CBSA areas are larger than the Zanobetti &
Schwartz, 2008 (Z & S) study areas (used in the first draft REA) 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
9-5
-------
1 (red) for the Z & S areas. The two adjustment methods were generally comparable in terms of
2 the changes in the upper quartile of the distribution. However, by design, quadratic rollback
3 always estimated decreases in the 75* percentile, median, and 25 percentile of the composite
4 monitor values, while HDDM estimated decreases in these values in some urban case study
5 areas, and increases in other areas consistent with atmospheric chemistry. HDDM-based
6 adjustments always produced increases in the lower tail of the distribution, while the lower tail
7 values generally remained unchanged with quadratic rollback. The differences between the two
8 adjustment procedures were the most pronounced in Los Angeles and New York, where the
9 largest reductions in NOx were required in order to meet the existing and alternative standards.
10 These large reductions in NOx caused a relatively large increase in lower Oj, concentrations
11 because of the reduction in NOx titration of O?,. As was noted in Chapter 4, the HDDM-based Os
12 estimates become more uncertain for larger changes in NOx and VOC emissions, and thus there
13 was less overall confidence in those results. Even in these cases, the HDDM approach is still
14 preferable because it captures better the overall shift in the distribution of Os concentrations.
15
9-6
-------
Atlanta: Z & S, April-October, 2006-2008
Baltimore: Z & S, April-October, Z006-ZOOB
Boston: Z & S. April-October. Z006-Z008
Cleveland: Z S S, April-October, Z006-ZOOB
Denver: Z & S, April-October, Z006-Z008
Detroit: Z & S, April-October, Z006-ZOOB
1
5
(
ba
3
o
n S
1
r T
se 7
^^ quadratic rollback
O
1
r-j-i
L2LJ
T
a
a —
a
OD
. .
fa -
Si -
a
a —
^^» quadratic rollback
^^~ ddm adjustment
S
~r •
i
base 75
Houston: Z & S, April-October, Z006-Z008
LosAngeles: Z S S, April-October, 2006-2008
NewVork: Z & S. April-October. Z006-Z008
Philadelphia: Z & S, April-October, Z006-Z008
Sacramento: Z S S, April-October, 2006-2008
SaintLouis: Z & S, April-October, 2006-2008
2 Figure 9-1 Distributions of composite monitor 8-hour daily maximum Os concentrations
3 from ambient measurements (black), quadratic rollback (blue), and the HDDM
4 adjustment methodology (red) for meeting the existing standard.
9-7
-------
1 9.2.2 Human Exposure Modeling (Chapter 5)
2 The population exposure assessment evaluates exposures to 63 using the Air Pollution
3 Exposure (APEX) model for the general population, all school-aged children (ages 5-18),
4 asthmatic school-aged children (ages 5-18), asthmatic adults (ages > 18), and older persons (ages
5 65 and older), with a focus on populations engaged in moderate or greater exertion (e.g. children
6 engaged in outdoor recreational activities). The strong emphasis on children, asthmatics, and
7 older adults reflected the findings of the last O3 NAAQS review (U.S. EPA, 2007) and the ISA
8 (U.S. EPA, 2013, Chapter 8) that these are important at-risk groups.
9 We assessed exposure in 15 urban case study areas - Atlanta, Baltimore, Boston,
10 Chicago, Cleveland, Dallas, Denver, Detroit, Houston, Los Angeles, New York, Philadelphia,
11 Sacramento, St. Louis, and Washington, D.C. - for recent 63 concentrations (2006-2010) and for
12 Os concentrations adjusted to just meet existing and alternative standards for two time periods
13 (2006-2008 and 2008-2010)3. The analysis provided estimates of the percent of several
14 populations of interest exposed to concentrations above three health-relevant 8-hour average Os
15 exposure benchmarks: 60, 70, and 80 ppb. The ISA includes studies showing statistically
16 significant effects at each of these benchmark levels (U.S. EPA, 2013). These benchmarks were
17 selected to provide some perspective on the public health impacts from exposures to various
18 concentrations that have been associated with (Vrelated health effects (e.g., lung inflammation
19 and increased airway responsiveness) in controlled human exposure and toxicological studies,
20 but cannot currently be evaluated in quantitative risk assessments. In addition, the exposure
21 assessment also identified the specific microenvironments and activities that contribute most to
22 exposure and evaluated at what times and how long individuals were in key microenvironments
23 and were engaged in key activities. This assessment focused on persons experiencing the highest
24 daily maximum 8-hour exposure within each study area. The assessment found that:
25 • Childhood is an important lifestage where higher exposures and risks can occur,
26 due to the higher time spent outdoors by children, the higher exposure
27 concentration experienced by children while outdoors (i.e. when they are
28 dismissed from school in the afternoon and during the summer, when they may be
29 at an outdoor camp all day) and engagement in moderate or high exertion level
30 activities.
3 Attainment with the O3 standard is based on the 4th highest maximum 8-hour O3 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 attainment in each of the two 3-year periods.
9-8
-------
1 • Persons spending a large portion of their time outdoors during afternoon hours
2 experienced the highest 8-hour 63 exposure concentrations given that 63
3 concentrations in other microenvironments were simulated to be lower than
4 ambient concentrations.
5 • Highly exposed children spend half of their outdoor time (on average) engaged in
6 moderate or greater exertion levels, such as in sporting activities. Highly exposed
7 adults also spent their outdoor time engaged in moderate or greater exertion levels
8 though on average, not as frequently as children.
9 Across the 15 urban case study areas, we find that children are of greatest concern for 03
10 exposures compared to other lifestages due to the greater amount of time they spend outdoors
11 engaged in moderate or higher exertion activities. The exposure analysis estimates that children
12 have the highest percent of exposures of concern of any of the at-risk populations or lifestages.
13 As a result, we focus on the results for children (ages 5-18) in the remainder of this discussion.
14 Figure 9-2 (reproduced from Figure 5-11) shows the results of the exposure assessment for all 15
15 urban case study areas, showing trends across the analytical years for the percent of children with
16 at least one 8-hour exposure greater than the 60, 70, and 80 ppb benchmarks.
17 The limited availability of longitudinal activity diary data and the general population
18 modeling approach used may underestimate the correlation in activity patterns for certain
19 susceptible populations (e.g., outdoor workers), and underestimate how often there are repeated
20 exposures to 03 concentrations above the exposure benchmarks. As a result, although we are
21 able to report the percent of the population with at least one exposure greater than the alternative
22 exposure benchmarks, we are less confident in the estimated percent of the population
23 experiencing more than one exposure. Individuals with repeated exposures may be at greater risk
24 of significant health effects (U.S. EPA, 2013, Section 6.2.1.1). In addition, the limited data on
25 responses to air quality alerts (e.g., averting behavior) indicates that a small percentage of the
26 population may engage in averting behavior in response to air pollution, which may overstate
27 actual exposures if individuals reduce their exposure during periods of high O?,.
28 The benchmark exposures of concern are not equivalent to ambient standard levels, as
29 exposures reflect the full pattern of 63 concentrations throughout a season, coupled with time
30 spent outdoors and indoors engaged in different activities. Thus, just meeting the existing
31 standard will result in shifts in the entire distribution of 63 over a three year period, and will
32 change the percent of populations experiencing each of the exposure benchmarks of concern.
33 Figure 9-2 shows that the percent of children above the 60 ppb benchmark declines consistently
34 across the 15 urban case study areas when just meeting potential alternative standards of 70, 65,
35 and 60. For most urban case study areas and years, the percent of children above the 60 ppb
36 benchmark is reduced by over half when 03 is adjusted to meet the 65 ppb alternative standard
9-9
-------
1 relative to the 75 ppb standard. In many urban case study areas and years, just meeting the 65
2 ppb alternative standard results in close to zero percent of children above the 60 ppb benchmark.
3 For the 70 and 80 ppb benchmarks, meeting an alternative standard of 70 ppb results in a small
4 percentage of children exceeding the benchmarks.
5 Year-to-year variability is relatively pronounced for exceedances of the 60 ppb
6 benchmark. In addition, we observe a geographic pattern to the years with the maximum percent
7 of exceedances of the exposure benchmarks reflecting the regional 63 patterns across years. In
8 general, northeastern urban case study areas saw the highest percentage of exceedances during
9 2007, while southern and western urban case study areas saw a higher percentage of exceedances
10 during 2006. However, these patterns are somewhat dependent on the 3-year averaging period
11 used to determine whether the standards are met. In general variability in the percent of children
12 exceeding the 60 ppb exposure benchmark across urban case study areas is similar to the
13 variability across years.
14 The percent of children with multiple exposures above the exposure benchmarks is
15 generally much lower compared to the percent of children with single exposures above the
16 benchmarks. However, as noted above, we have lower confidence in these estimates. Even for
17 the lowest benchmark level of 60 ppb, most locations and years have less than 10 percent of
18 children experiencing 2 or more exposures when just meeting the existing standard of 75 ppb,
19 less than 5 percent when just meeting an alternative standard of 70 ppb, and less than 1 percent
20 when just meeting an alternative standard of 65 ppb. For most urban case study areas and years,
21 less than 1 percent of children experience 2 or more exposures above the 70 ppb exposure
22 benchmark when just meeting the existing standard.
9-10
-------
0)
>»
§
« *
O w
Q- 3
X :=•
.
CO -
;; °
® X
= UJ
O ._
«
5 2
C fl^
03 "U
J= O
0
g
a)
Q.
Exposure
Benchmark
60
70
80
/ t / / /
2 Figure 9-2 Effects of just meeting existing (column 1) and alternative (columns 2 through
3 4) standards on percent of children (ages 5-18) with at least one Os exposure at or
4 above 60, 70, and 80 ppb-8hr while at moderate or greater exertion, years 2006-
5 2010.4
We were not able to adjust air quality to just meet the 60 ppb alternative standard in the New York City 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.
9-11
-------
1 Table 5-6 summarized the percent of the population of children (ages 5-18) with at least
2 one daily 8-hour exposure above the 60, 70, and 80 ppb benchmarks, providing both the mean
3 and maximum percentage across the five analytical years for each urban case study area. For Os
4 adjusted to just meet the existing standard of 75 ppb, the highest maximum percentage of
5 children exceeding the 60 ppb benchmark across years, 26 percent, occurs in Denver, which also
6 has the highest mean percentage across years. After just meeting the existing standard, Los
7 Angeles has the lowest maximum (10 percent) and mean (9.5 percent) percentage of children
8 exceeding the 60 ppb benchmark across years, likely reflecting the highly skewed nature of 63
9 concentrations in that urban case study area. For example, just meeting the existing standards in
10 Los Angeles moves the majority of 63 concentrations (sites and days) well below 60 ppb (See
11 Appendix 4-D). Patterns across urban case study areas are generally similar after just meeting
12 alternative standards of 70, 65, and 60 ppb, with the exception that the lowest maximum and
13 mean percentage of children for the alternative standard level of 65 ppb occurs in the New York
14 City urban case study area, which had very large (greater than 90 percent) reductions in NOx
15 emissions that were used to adjust air quality to just meet the 65 ppb standard level in that urban
16 case study area. This resulted in the distribution of 63 concentrations covering most days of the
17 year and most monitoring sites shifting dramatically downward, with most concentrations well
18 below 60 ppb across the New York City urban case study area. The level of confidence in the
19 results for the New York City and Los Angeles study areas for just meeting the alternative
20 standards is lower than that for some of the other urban case study areas due to the FtDDM-based
21 63 estimates becoming more uncertain for very large changes in precursor emissions.
22 Figure 9-3 (reproduced from Figure 5-19) shows the results of the exposure assessment
23 for all 15 urban case study areas, showing the effect on the percent of children with one or more
24 exposures above the 60 ppb benchmark of just meeting the existing and alternative standards.
25 For each alternative standard, Figure 9-2 shows the maximum percent of children exceeding the
26 benchmark across the modeled years 2006-2010. Patterns of results are similar for the 70 ppb
27 and 80 ppb benchmarks, however, the maximum percents of children exceeding those higher
28 benchmarks are much smaller for all alternative standards. The percent of children exceeding the
29 80 ppb benchmark is close to zero once the existing standard is met. The percent of children with
30 two or more exposures exceeding the 60 ppb benchmark level is substantially lower when just
31 meeting the existing standard, and is close to zero for the 70 ppb and 80 ppb benchmarks. This
32 percentage drops substantially when meeting the 70 ppb standard, and is close to zero in most
33 urban case study areas when meeting the 65 ppb and 60 ppb alternative standards. Patterns for
34 asthmatic children are very similar to patterns for all children.
9-12
-------
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Denver
Detroit
Houston
Los Angeles
New York
Philadelphia
Sacramento
St. Louis
Washington
0%
2% 4% 6% 8% 10°o 12% 14% 16% 18% 20% 22% 24% 26%
Percent of Children with at Least One 8-hr Daily Max Exposure 60 ppb
standard level (ppb) i i 60 i i 65 i i "0 i i "5
2 Figure 9-3 Effects of just meeting existing (75 ppb) and alternative standards on percent of
3 children (ages 5-18) exceeding 60 ppb exposure benchmark, highest value across
4 years for each urban case study area, 2006-2010.5
5
6 9.2.3 Health Risks Based on Controlled Human Exposure Studies (Chapter 6)
7 Using the estimates of exposure from APEX combined with results from controlled
8 human exposure studies, we estimated the number and percent of at-risk populations or lifestages
9 (all children aged 5-18, children with asthma aged 5-18, adults aged 18-35, adults aged 36-55,
10 and outdoor workers) experiencing selected decrements in lung function. The analysis focuses on
11 estimates of the percent of each at-risk population or lifestage experiencing a reduction in lung
12 function (mostly for durations of one to five hours) for three different levels of impact, 10, 15,
5 We were not able to adjust air quality to just meet the 60 ppb alternative standard in New York City urban case
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.
9-13
-------
1 and 20 percent decrements in FEV1. These levels of impact were selected based on the literature
2 discussing the adversity associated with increasing lung function decrements (US EPA, 2013,
3 Section 6.2.1.1). Consistent with the exposure assessment, we focus this summary on lung
4 function decrements in children as they are the lifestage likely to have the greatest percentage at-
5 risk due to higher levels of exposure and exertion. Within the overall population of children,
6 asthmatic children may have less reserve lung capacity to draw upon when faced with
7 decrements, and therefore a >10% decrement in lung function may be a more adverse event in an
8 asthmatic child than a healthy child.
9 Lung function risks (based on experiencing an estimated 10, 15, or 20 percent decrement
10 in lung function) were estimated for each of the 15 urban case study areas in which human
11 exposures were modeled. Two models were used to estimate lung function risks: one based on
12 application of a population level exposure-response (E-R) function consistent with the approach
13 used in the previous 63 NAAQS review, and one based on application of an individual level E-R
14 function (the McDonnell-Stewart-Smith (MSS) model), introduced in this review, which
15 incorporates individual differences in physiology, age, and activity patterns (McDonnell et al.,
16 2012). Because the individual level E-R function approach allows for a more complete estimate
17 of risk (incorporating risk responses at varying activity levels, not just moderate or greater
18 exertion), we focus on the results of that approach for this discussion.
19 The MSS model as implemented in APEX has a term that adjusts the lung function
20 response according to an individual's age. The MSS model was fit using data from subjects who
21 ranged in age from 18 to 35. Thus, the MSS model is not able to account for differences in lung
22 function at different age groups between the ages of 5 and 18. However, age does have a
23 pronounced effect on lung function response in the APEX model. APEX models differences in
24 physiological parameters due to age, and these result in age-dependent predictions of ventilation
25 rates, which are used in the MSS model. Ventilation rates also depend on the activities being
26 performed, which are also age-dependent. As a result of differences in physiology and activities,
27 the lung function responses vary by age (see Appendix 6-E).
28 Figure 9-4 (reproduced from Figure 6-6) shows the results of the lung function risk
29 assessment for all 15 urban case study areas, showing trends across the analytical years for the
30 percent of children with predicted lung function decrements greater than or equal to 10 percent6.
6 We have introduced a new method (relative to the O3 NAAQS review completed in 2008) for calculating the
percent of the at-risk populations (all children and asthmatic children) experiencing lung function decrements,
based on modeling of individual level responses to O3 exposures. This model yields significantly 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 MSS model can reflect greater sensitivity of children to O3 exposures because it
allows for age variability in the relationship between O3 and FEV1 decrements, and younger populations are
more responsive to O3 exposures than older populations.
9-14
-------
1 Specifically, Figure 9-4 shows that the percent of children (age 5-18) with greater than or equal
2 to 10 percent lung function decrement declines consistently across the 15 urban case study areas
3 when just meeting the existing 75 ppb standard, as well as the alternative standards of 70, 65, and
4 60. The percent of children at-risk at the 10 percent decrement level remains at or above 10
5 percent in many locations after just meeting the 60 ppb alternative standard. The percentage of
6 children with greater than or equal to a 15 or 20 percent lung function decrement is much lower
7 for all alternative standards, with close to zero percent of children at-risk when just meeting the
8 alternative standard of 60 ppb. In general variability in percent of children at-risk across urban
9 case study areas is similar to variability across years.
10
9-15
-------
c
0)
O
0)
G)
O
O
0)
O
0)
Q.
1
2
3
4
8 918 1=8 9 .*» * *
Figure 9-4 Effects of just meeting existing (column 1) and alternative (columns 2-4)
standards on percent of children (ages 5-18) with FEVi decrement > 10,15, and 20%, years
2006-2010.7
We were not able to adjust air quality to just meet the 60 ppb alternative standard in New York City urban case
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.
9-16
-------
1 Figure 9-5 (reproduced from Figure 6-11) shows the results of the lung function risk
2 assessment for all 15 urban case study areas, showing the effect on the risk of a 10 percent or
3 greater lung function decrement in children (ages 5-18) of just meeting the existing and
4 alternative Os standards. For each alternative standard, Figure 9-5 shows the maximum percent
5 risk over all of the modeled years 2006-2010.
6 There is no consistent pattern in the percent of children with 10 percent or greater lung
7 function decrement across urban case study areas just meeting the existing standard of 75 ppb.
8 The 5-year maximum estimated percent of children at-risk ranges from 17 to 22 percent across
9 urban case study areas. The percent reduction in 5-year maximum risk when just meeting the 70
10 ppb alternative standard is more consistent across urban case study areas, ranging from 8 to 23
11 percent (excluding New York City, which had a reduction of 29 percent). Reductions in risk
12 when just meeting the 65 ppb alternative standard are also generally consistent across urban case
13 study areas, with the exception of New York City. Incremental reductions in risk when just
14 meeting the alternative 65 ppb standard compared with just meeting the 70 ppb alternative
15 standard range from 17 to 31 percent excluding New York City, which has a reduction in risk of
16 more than twice as much as the next largest reduction. Incremental reductions in risk from just
17 meeting the alternative 60 ppb standard compared with just meeting the 65 ppb standard are
18 generally consistent, ranging from 16 to 46 percent, with somewhat larger reductions in risk
19 occurring in Cleveland and Denver. Overall, the 5-year maximum percent of children at-risk for
20 lung function decrements of 10 percent or more exceeds 13 percent, 10 percent, and 5 percent in
21 all urban case study areas except New York after just meeting alternative standards of 70, 65,
22 and 60, respectively. Patterns of risk reductions are also similar for the alternative lung function
23 decrement levels of 15 percent and 20 percent. However, the initial percent of the population
24 experiencing these decrements when just meeting the existing standard are substantially lower.
25 Patterns of risk responses using the population level exposure-response model are similar
26 to the MSS individual risk model. However, the starting values for the percent of the population
27 at risk are lower, reflecting the limits of the model in reflecting individual level responses, and
28 the limited coverage of the model for exposures at lower exertion levels. For children, the MSS
29 model gives results typically a factor of three higher than the population level E-R model used in
30 the previous Os NAAQS review.
31
9-17
-------
0% 2%
6% 8% 10% 12% 14% 16% 18%
percent of children with FEV1 decrement > 10%
standard level (ppb) ^m 60 ^m 65 i 1 70 ^m 75
20% 22%
T-T
24%
1 Figure 9-5 Impact of just meeting existing (75 ppb) and alternative standards on percent
2 of children (ages 5-18) with FEVi decrement > 10%, highest value for each urban
3 case study area, 2006-2010.8
4
5 9.2.4 Health Risks Based on Epidemiological Studies (Chapters 7 and 8)
6 The epidemiology-based risk assessment evaluated mortality and morbidity risks from
7 short-term O^ exposures and mortality risks from long-term exposures to O^ by applying
8 concentration-response (C-R) functions derived from selected epidemiology studies. The
9 analysis included both a set of urban case study area case studies and a national-scale
10 assessment. The urban case study analyses evaluated mortality and emergency department (ED)
11 visits, hospitalizations, and respiratory symptoms associated with recent 63 concentrations
12 (2006-2010) and with Os concentrations adjusted to just meet the existing and alternative Os
We were not able to adjust air quality to just meet the 60 ppb alternative standard in New York City 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.
9-18
-------
1 standards (see section 9.2.1 and Chapter 4). Mortality and hospital admissions (HA) were
2 evaluated in 12 urban case study areas, while ED visits and respiratory symptoms were evaluated
3 in a subset of areas with supporting epidemiology studies. The 12 urban case study areas were:
4 Atlanta, GA; Baltimore, MD; Boston, MA; Cleveland, OH; Denver, CO; Detroit, MI; Houston,
5 TX; Los Angeles, CA; New York, NY; Philadelphia, PA; Sacramento, CA; and St. Louis, MO.
6 The urban case study analyses focus on risk estimates for the middle year of each three-year
7 design value period (2006-2008 and 2008-2010) in order to provide estimates of risk for a year
8 with generally higher Oi concentrations (2007) and a year with generally lower Oi
9 concentrations (2009).
10 Most of the endpoints evaluated in epidemiology studies cover the entire study
11 population including children and adults. Because most mortality and hospitalizations occur in
12 older persons, these epidemiology-based risk estimates are better indicators of effects in adults
13 than in children. This is an important distinction from the human exposure and lung function risk
14 assessments, which focus on children. The only endpoints specific to children are asthma and all
15 respiratory hospital admissions using the New York specific epidemiology study, respiratory ER
16 visits in Atlanta, and respiratory symptoms in asthmatic children in Boston.
17 Both the urban case study area and national-scale assessments provide the absolute
18 incidence and percent of incidence attributable to Os. In addition, risks are presented in terms of
19 incidence per 100,000 population to control for the differences in the sizes of the populations
20 across urban case study areas, and to allow for comparison of risks using different definitions of
21 urban extent. In previous reviews, Oj risks have only been estimated for the portion of total Oj
22 attributable to North American anthropogenic sources (above what was referred to in previous
23 reviews as "policy-relevant background Os") In contrast, this assessment estimates risk for Oj,
24 concentrations down to zero, reflecting the lack of evidence for a detectable threshold in the C-R
25 functions (U.S. EPA, 2013, Chapter 2), and the understanding that U.S. populations may
26 experience health risks associated with Oj, resulting from emissions from all sources, both natural
27 and anthropogenic, within and outside the U.S. In order to better reflect how Os distributions are
28 likely to respond to just meeting existing and potential alternative standard levels, we adjusted
29 Os concentrations to just meet existing and potential alternative standard levels using reductions
30 in only U.S. anthropogenic emissions of Os precursors. Thus, the estimated changes in risk
31 between just meeting the existing standards and just meeting potential alternative standard levels
32 only reflect reductions in U.S. anthropogenic emissions.
33 However, consistent with the conclusions in the Os ISA (U.S. EPA, 2013), we have
34 relatively lower certainty about the shape of the C-R function towards the lower end of the
35 distribution of Os concentrations used in fitting the function due to the reduction in the number
36 of Os measurements in this portion of the distribution. We discuss this source of uncertainty
9-19
-------
1 below. In addition, we provide the distribution of mortality incidence across the range of Os
2 concentrations in Chapter 7 to inform discussions of uncertainty in the results.
3 9.2.4.1 Urban Case Study Results
4 Figures 9-6 and 9-7 (reproduced from Figures 7-4 and 7-5) show the results of the
5 mortality and adult (ages 65 and older) respiratory hospital admissions risk assessments for all 12
6 urban case study areas, showing the effect on the incidence per 100,000 population just meeting
7 the existing 75 ppb standard and alternative O?, standards of 70, 65, and 60 ppb in 2007 and
8 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
nver, CO
Detroit, M!
Houston, IX
LosAngeles, CA
New York, NY
Philadelphia, PA
Sacramento, CA
St. Louis, MO
75ppb
70ppb
65ppb
GOppb
2009 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
-NewYork. NY
-Philadelphia, PA
-Sacramento, CA
St. Louis, MO
75ppb
70ppb
bSppb
60ppb
4
5
Figure 9-6 Impacts of just meeting existing (75 ppb) and alternative standard levels on
mortality risk per 100,000 population for 2007 and 2009.9
' As noted earlier, we were not able to adjust air quality to just meet the 60 ppb alternative standard in New York
City 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.
9-21
-------
2007 Simulation year
i
2
3
4
5
i
Trend in ozone-related HA across standard levels
(HA per 100,000)
—*- Atlanta, GA
• Baltimore, MD
M Boston, MA
^*— Cleveland. OH
—#— Denver, CO
^^ Detroit, Ml
I Houston. TX
Los Angeles. CA
— New York, NY
—•-Philadelphia, PA
->- Sacramento, CA
-*- St. Louis, MO
75ppb
70ppb
65ppb
eoppb
2009 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
-LosAngeles, CA
-New York, NY
-Philadelphia, PA
-Sacramento, CA
-St. Louis, MO
75ppb
70ppb
65ppb
60ppb
Figure 9-7 Impacts of just meeting existing and alternative standard levels on adult (ages
65 and older) respiratory hospital admissions risk per 100,000 population for 2007
and2009.10
1 We were not able to adjust air quality to just meet the 60 ppb alternative standard in New York City urban case
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.
9-22
-------
1 In some urban case study areas which have large NOx emissions (e.g. from heavy
2 downtown traffic), Os levels are artificially low because the NOx emissions remove Os through a
3 chemical reaction (see section 9.2.1 and Chapter 4). In these places, when NOx emissions are
4 decreased to reduce peak O?, concentrations across the entire CBS A, which often includes
5 locations outside of the urban core areas, lower concentrations of Oj can go up. This can also
6 happen in other areas on the lowest Os days. This phenomenon occurs in some locations when
7 meeting lower alternative standards as well.
8 The overall trend across urban case study areas is small decreases in mortality and
9 morbidity risk as O^ concentrations are adjusted to just meet incrementally lower alternative
10 standard levels. In New York, there are somewhat greater decreases in these risks, reflecting the
11 relatively large emission reductions used to adjust air quality to just meet the 65 ppb alternative
12 standard, and the substantial change in the distribution of Os concentrations that resulted. We
13 were not able to adjust Oj concentrations to just meet the 60 ppb alternative standard in the New
14 York City urban case study area. Risks vary substantially across urban case study areas;
15 however, the general pattern of reductions across the alternative standards is similar between
16 urban case study areas. Because of the generally lower baseline Oj concentrations in 2009, risks
17 are generally slightly lower in 2009 relative to 2007; however, the patterns of reductions in risk
18 are very similar between the two years.
19 Mortality and morbidity risks generally do not show large responses to meeting existing
20 or alternative levels of the standard for several reasons. First, these risks are based on C-R
21 functions that are approximately linear along the full range of concentrations, and therefore
22 reflect the impact of changes in Os along the complete range of 8-hour average Os
23 concentrations. This includes days with low baseline11 Os concentrations that are predicted to
24 have increases in O3 concentrations, as well as days with higher starting Oj concentrations that
25 are predicted to have decreases in 03 concentrations as a result of just meeting existing and
26 alternative standards. Second, these risks reflect changes in the urban-area wide monitor average,
27 which will not be as responsive to air quality adjustments as the design value monitor, and which
28 includes monitors with both decreases and increases in 8-hour concentrations. Third, the days
29 and locations with predicted increases in O3 concentrations (generally those with low to
30 midrange starting Os concentrations) resulting from just meeting the existing or alternative
31 standard levels generally are frequent enough to offset days and locations with predicted
32 decreases in 03. The heat maps presented in Figures 7-2 and 7-3 demonstrate that just meeting
33 progressively lower alternative standard levels narrows the distribution of risk across the range
11 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.
9-23
-------
1 of Os concentrations. In addition, the distribution of risk tends to be more centered on area-wide
2 average concentrations in the range of 25 to 55 ppb after just meeting an alternative standard of
3 60 ppb. The focus of the epidemiological studies on urban case study area-wide average Os
4 concentrations, and the lack of thresholds coupled with the linear nature of the C-R functions
5 mean that in this analysis, the impact of a peak-based standard (which seeks to reduce peak
6 concentrations regardless of effects on low or mean concentrations) on estimates of mortality and
7 morbidity risks based on results of those studies is relatively small. For example, for mortality
8 and hospital admissions, we find a less than 10 percent reduction in risk for most urban case
9 study areas when just meeting the 70 ppb and 65 ppb alternative standards compared to just
10 meeting the existing standard, and a less than 25 percent reduction in risk for all urban case study
11 areas when just meeting the 60 ppb standard compared to just meeting the existing standard. The
12 general pattern for other morbidity risks is similar to hospital admissions. However, we are not
13 able to draw strong conclusions about the results across urban case study areas, because of the
14 limited number of urban case study areas represented for most of the endpoints.
15 We have applied city-specific mortality effect estimates to each urban case study area
16 based on the largest multi-city epidemiological study. However, for many of the urban case study
17 areas, the risk estimates have wide confidence intervals that can include zero, due to the lower
18 statistical power of some of the city-specific effect estimates relative to the national combined
19 effect estimate across cities. Furthermore, there is significant variability in these effect estimates
20 across the 12 urban case study areas, with some urban case study areas having effect estimates
21 from 5 to 7 times greater than other cites (see Chapter 7, section 7.4.1).12 The variability in effect
22 estimates, along with differences in Os concentrations, is a driver for the overall variability in the
23 risk results across cities. Smith et al (2009) reports an overall significant national mortality effect
24 estimate with confidence intervals that do not include zero, reflecting the much greater statistical
25 power available when pooling information across urban case study areas.
26 We also evaluated mortality risks in the 12 urban case study areas associated with long-
27 term Os exposures (based on the seasonal average (April to September) of the peak daily one-
28 hour maximum concentrations). Risks from long-term exposures after just meeting the existing
29 standard are substantially greater than risks from short-term exposures, ranging from 16 to 20
30 percent of respiratory mortality across urban case study areas. However, the percent reductions
31 in long-term mortality risks are similar to those for mortality from short-term exposures. For
32 example, we find a less than 10 percent reduction in risk relative to just meeting the existing
12 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.
9-24
-------
1 standard in most areas when just meeting the 70 ppb and 65 ppb alternative standards, and a less
2 than 20 percent reduction when just meeting the 60 ppb alternative standard level. Risk
3 reductions for the New York City urban case study area are much greater when just meeting the
4 65 ppb alternative standard compared to just meeting the existing standard, with a 24 percent
5 reduction in risk in 2007.
6 New York and Los Angeles have characteristics that make epidemiological risk estimates
7 particularly uncertain. In the case of New York, the expansion of the urban case study area
8 definition to the CBS A adds uncertainty due to the large and diverse nature of the CBS A. The
9 New York CBS A includes two urban case study areas which have separate effect estimates
10 available from the Smith et al. (2009) study. These separate effect estimates (for Newark, NJ and
11 Jersey City, NJ) are smaller than the effect estimate for New York, however, they are also based
12 on much smaller populations, and have relatively wider confidence bounds, reflecting low
13 statistical power. For consistency with other urban case study areas and to allow for comparison
14 between the CBSA-based risk estimates and the smaller study area based estimates (see the
15 sensitivity analyses in Chapter 7), we elected to apply the New York city effect estimate, which
16 is based on a very large population and has high precision, to all of the counties in the New York
17 CBSA. While this adds substantial uncertainty to the absolute incidence of mortality for the New
18 York CBSA, it does not affect the pattern of risk reductions when just meeting alternative
19 standards. In addition, as noted earlier, the Os adjustments to meet existing and alternative
20 standards in New York and Los Angeles also have additional uncertainties relative to the other
21 10 urban case study areas.
22 We conducted a number of sensitivity analyses based on a population normalized
23 mortality risk metric, e.g. mortality risk per 100,000 population. Maintaining the general linear,
24 no-threshold functional form, mortality risks per 100,000 population are generally robust to
25 alternative specifications of the C-R functions, although in several urban case study areas, using
26 effect estimates from Smith et al. (2009) which were derived using regional priors rather than
27 national priors results in higher risk estimates13. Using the effect estimates from Zanobetti and
28 Schwartz (2008) has no consistent effect on risk results across the urban case study areas. Using
29 effect estimates based on a copollutant model with PMio, mortality risks are higher in some
30 locations and lower in others. However, in all locations the confidence intervals are substantially
13 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.
9-25
-------
1 wider using the copollutant model with PMio (due to fewer days with both pollutants measured),
2 which makes it difficult to determine whether the increases and decreases in estimates relative to
3 the core estimates are real or the result of statistical error.
4 We selected the CBSA as the spatial definition for the urban case study areas. We made
5 this selection to address a downward bias that we identified resulting from a mismatch between
6 the smaller urban core areas used in the epidemiology studies and the larger areas where Os
7 concentrations are expected to change as a result of meeting the existing and alternative standard
8 levels (see Chapter 7). We included a sensitivity analysis evaluating the result of using a smaller
9 geographic area including only the counties used in the epidemiology study. As expected, using a
10 smaller geographic extent for the urban case study areas results in smaller, and in some cases
11 negative risk reductions when compared to using the CBSA definitions. This reflects the fact that
12 the controlling14 monitor in many of the 12 urban case study areas is located outside of the small
13 set of counties included in the Smith et al. (2009) urban case study area definitions, and some of
14 the monitors that are within that more limited spatial extent are more prone to Os titration due to
15 local NOx emission sources. As a result, those monitors are more likely to see increases in Oj,
16 which will, if other monitors with higher concentrations in the broader regions are not included,
17 lead to estimated increases in risk due to the application of a linear, no threshold C-R function.
18 This bias can be substantial, especially in St. Louis and several urban case study areas in the
19 Northeast, including Boston, New York, and Philadelphia, where the highest concentration
20 monitors are outside the Smith et al. (2009) urban case study area definitions.
21 Sensitivity analyses were conducted for scenarios of just meeting existing and alternative
22 standards using combinations of NOx and VOC emissions reductions (as compared to NOx
23 reductions alone). The addition of VOC emissions reductions had little impact with the exception
24 of New York and Los Angeles, where risk was decreased relative to the NOx-only reduction
25 scenario.
26 9.2.4.2 National-scale Assessment Results
27 The national-scale assessment evaluated only mortality associated with recent Oj
28 concentrations across the entire U.S for 2006-2008. The national-scale assessment is a
29 complement to the urban scale analysis, providing both a broader geographic assessment of Os-
30 related health risks across the U.S., as well as an evaluation of how well the 12 urban study areas
31 represented the full distribution of Os-related health risks in the U.S. The national-scale
32 assessment demonstrates that there are Oj risks across the U.S, not just in urban case study areas,
33 even though the Os concentrations in many areas were lower than the existing standard level.
14
The controlling monitor is the monitor with the highest design value within a defined non-attainment area.
9-26
-------
1 While we did not assess the changes in risk at a national level associated with just meeting
2 existing and alternative standards, just meeting existing and alternative standards would likely
3 reduce Os concentrations both in areas that are not meeting those standards and in locations
4 surrounding those areas, leading to risk reductions that are not included in the urban-scale
5 analysis.
6 9.3 COMPARISON OF RESULTS ACROSS EXPOSURE, LUNG FUNCTION RISK,
7 AND EPIDEMIOLOGY-BASED MORTALITY AND MORBIDITY RISK
8 ANALYSES
9 In considering the overall results across the human exposure, lung function risk, and
10 epidemiology-based risk assessments, we focus on the key policy-relevant metrics and levels for
11 each type of assessment. For the human exposure assessment, we selected exposures above the
12 60 ppb exposure benchmark for all children (ages 5-18). We select this exposure metric because
13 children represent a key at-risk lifestage, and the 60 ppb exposure benchmark is the lowest
14 exposure level associated with significant findings in controlled human exposure studies. For the
15 lung function risk assessment, we selected the results for lung function decrements greater than
16 or equal to 10 percent for all children (ages 5-18). We select this lung function risk metric
17 because children represent a key at-risk lifestage, and a 10 percent lung function decrement
18 represents a potentially more adverse event in asthmatic children. For the epidemiology-based
19 risk assessment we selected the core short-term exposure mortality results and the respiratory
20 hospital admission results, because these endpoints were estimated for all of the 12 urban case
21 study areas. Generally speaking, these metrics provide the most differentiation between the
22 alternative standards, helping to inform policy-relevant questions regarding adequacy of the
23 existing standard, and public health impacts of meeting alternative standards. The other metrics
24 analyzed in this REA (e.g. other exposure benchmarks and other lung function decrements) show
25 less response to just meeting the existing standard or potential alternative standard levels.
26 As discussed in Chapter 2, we designed the exposure and risk assessment to help inform
27 two fundamental questions related to the adequacy of the existing standard in protecting public
28 health and the degree of exposure and risk reductions associated with alternative standards
29 compared with the existing standard. The following discussion evaluates the three types of
30 analyses we conducted in terms of the consistency of the information provided to inform these
31 questions.
32 9.3.1 Evaluation of Exposures and Risks After Just Meeting the Existing Standard
33 To compare the results of the three assessments in urban case study areas, we plot the key
34 metrics from each analysis across urban case study areas for the two common years of analysis
35 (i.e., 2007 and 2009). For three urban case study areas (i.e., Chicago, Dallas, and Washington
9-27
-------
1 D.C.) we have only the exposure and lung function risk assessments, as these urban case study
2 areas did not have sufficient information to estimate epidemiology-based risks. The
3 epidemiology-based metrics are the percent of baseline short-term exposure mortality, based on
4 the core estimates using the C-R functions from Smith et al. (2009), and respiratory hospital
5 admissions based on the core estimates using the C-R functions from Medina-Ramon (2006),
6 attributable to 63. Figure 9-8 presents the exposures and risks after just meeting the existing
7 standard of 75 ppb. Each row represents one of the key analytical results; each column gives the
8 results for 2007 and 2009 for each urban case study area. The scale of each analytical metric for
9 each analysis differs, and thus the comparisons across analyses should focus on overall patterns
10 rather than on direct comparisons of numeric estimates.
11 All of the metrics show substantial variability among urban case study areas, although
12 there appears to be less variability in lung function risk and hospital admission risk compared
13 with the exposure metric and mortality risk. The differences between estimates for 2007 and
14 2009 are much higher for some urban case study areas (e.g. Baltimore and Philadelphia) for the
15 exposure metric than any of the risk metrics. This may reflect the explicit threshold nature of the
16 exposure metric, which focused on exposures above a benchmark level of 60 ppb. Differences
17 between years in exposures above the 60 ppb benchmark after just meeting the existing standard
18 are dependent on the number of days during each year with decreases in higher O^
19 concentrations, as well as the magnitude of the decreases in Oj, on those higher 63 concentration
20 days. These in turn are sensitive to the shape of the 63 distribution in the analytical year prior to
21 just meeting the existing standard (which determines the starting number of days above 60 ppb)
22 and the response to emissions reductions applied in meeting the existing standard for 2007 or
23 2009. There is some consistency between metrics in the urban case study areas with highest
24 values for the exposure and lung function risk metrics. However, there were still differences,
25 especially for Los Angeles, which had one of the higher values for lung function risk in 2009,
26 but had one of the lower percentages of children exposed above the 60 ppb benchmark. This
27 again points to the importance of the threshold nature of the exposure metric, combined with the
28 tendency for more substantial decreases in peak 63 concentrations relative to mid-range and low
29 concentrations when just meeting the existing standard.
30 There is little consistency within urban case study areas between the epidemiology risk
31 metrics and the exposure and lung function risk metrics, and there is also little consistency
9-28
-------
1 between the mortality and hospital admission risks. Houston has the lowest metric values in 2007
2 (except for mortality risk), but in 2009 has some of the higher risk metrics (except for hospital
3 admission risk). New York has the highest mortality risk in 2007 and 2009 but has among the
4 lowest hospital admission risks in both years.
9-29
-------
12007
2009
Figure 9-8 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.
9-30
-------
1 9.3.2 Reductions in Exposure and Risk Metrics after Just Meeting Alternative Standards
2 To compare the results of the three assessments for urban case study areas after just
3 meeting alternative standards relative to the existing standard, we express each result as a percent
4 of the metric value when just meeting the existing standard. Figure 9-9 presents the percent
5 reduction in exposures and risks after just meeting alternative standards relative to just meeting
6 the existing standard of 75 ppb. In this plot, each row represents one of the key analytical results
7 and each column gives the results for 2007 and 2009 for each urban case study area. The scales
8 are the same between analyses, and as such, it is informative to examine both the overall patterns
9 of change between alternative standards, and also the absolute value of the percent reductions in
10 risk metrics between analyses. In interpreting this chart, higher values mean greater reductions in
11 risk or exposure relative to just meeting the existing standards. Because these are percent
12 reductions, the maximum value is one hundred percent, which if reached would indicate that
13 risks or benchmark exposures are completely eliminated when the alternative standard is met in
14 the urban case study area as was seen for the 60 ppb exposure benchmark.
15 Many of the differences in results across the metrics are driven by how each metric is
16 affected by the Os data input to the analysis. In general, the impact of the HDDM adjustments to
17 Os vary based on three main considerations: 1) the degree to which the exposure or risk metric is
18 sensitive to changes across the various ranges of Os concentrations (e.g. high, mid-range, low);
19 2) whether the exposure or risk metric uses individual census tract concentrations or area-wide
20 average concentrations; and 3) changes in the distribution of 63 concentrations in the year of
21 analysis between recent Os concentrations and adjusted (meeting the existing or alternative
22 standards) 63 scenarios. With respect to 1), the exposure benchmark metric, which focuses only
23 on exposures above 60 ppb, will not be sensitive at all to changes in 63 concentrations in the
24 range below 60 ppb. The lung function risk metric, which depends on the dose rate and
25 individuals' characteristics, does not have a concentration threshold. However, because of the
26 logistic form of the response function, it is less sensitive to lower OT, concentrations and has very
27 few FEV1 responses greater than 10 percent when exposure concentrations are below 20 ppb and
28 very few FEV1 responses greater than 15 percent when exposure concentrations are below 40
29 ppb. On the other hand, the mortality and hospital admission risk metrics are based on non-
30 threshold, approximately linear C-R functions, and therefore will be sensitive to changes in 63
31 along the full range of 63. As discussed in Chapter 4, because 63 at lower concentrations may
32 increase following HDDM adjustment in some locations and on some days to just meet
33 alternative standards15, this can lead to increases in risk on some days, which can lead to a net
15 The frequency and magnitude of increases in spatially averaged mean concentrations in an urban case study area
occur during a season when adjusting air quality to just meet a standard vary considerably between the existing
9-31
-------
1 increase or decrease in risk over the entire year, depending on whether the days with increased
2 risk exceed days with decreased risk (generally due to a preponderance of days with lower 63
3 concentrations). With respect to 2), the exposure and lung-function risk metrics are based on
4 concentrations at individual census tracts since they depend on O^ exposure modeled by moving
5 each individual through their environment. Because of this, the exposure and lung-function risk
6 metrics are most affected by the spatial and temporal variability of Os concentrations across the
7 urban case study area. The mortality and hospital admission risk metrics are calculated applying
8 C-R functions to area-wide, daily maximum 8-hr average 63 concentrations. As a result, the
9 spatial variability in O?, concentrations between the monitors will only influence the
10 epidemiology-based risk estimates in how they influence the area-wide average. With respect to
11 3), all three metrics are influenced by how the distribution of Os concentrations changes between
12 recent O^ conditions and after adjustment to just meet existing and alternative O^ standard levels.
and alternative standards. The highest frequency of occurrence of days with increasing O3 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.
9-32
-------
Atlanfc Baltim Bostot Chicat Clevel Dallas Denve Detroi Housti
LosAi New> Philad Sacra St Lo
JD
m
Alternative Standard
65ppb
\60ppb
2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009
Analytical Year
2007 2009 2007 2009 2007 2009 2007 2009 2007 2009 2007 2009
Figure 9-9 Comparison of Percent Reduction in Key Risk Metrics for Alternative Standard Levels Relative to Just Meeting
the Existing 75 ppb Standard.
9-33
-------
1 The exposure and lung function risk metrics are most affected by the reductions in the
2 individual monitors' peak 63 concentrations, including the magnitude of these reductions and the
3 number of days that experience these reductions. In contrast, the mortality and hospital
4 admission risk metrics are affected by changes in the mean of the seasonal, area-wide average Os
5 concentrations, where the mean is determined by the frequency and magnitude of increases
6 versus decreases in area-wide, maximum daily 8-hr Os concentrations16. In addition to Os
7 concentrations, there are other factors that affect the variability across urban case study areas for
8 these three metrics, such as activity data and exposure factors for the exposure and lung function
9 risk metrics and the study-specific C-R functions for the mortality and hospital admission risk
10 metrics.
11 One clear observation is that the percent reductions in risk from meeting alternative
12 standard levels relative to meeting the existing standard for the two epidemiology-based
13 endpoints are much smaller than for the exposure benchmark and lung function risk endpoints.
14 The maximum percent reduction in the mortality and hospital admissions risk relative to just
15 meeting the existing standard across years, locations, and alternative standards is less than 25
16 percent, and for many years/locations, the reductions in these risks when just meeting the lowest
17 alternative standard, 60 ppb, are less than 10 percent. The exposure benchmark results show the
18 most reductions when comparing just meeting the existing standard to just meeting alternative
19 standards. Just meeting the 65 ppb standard results in reductions in the percent of children
20 exceeding the 60 ppb exposure benchmark by over 50 percent in all urban case study areas, and
21 by over 75 percent in 12 of the 15 urban case study areas evaluated. For most locations and
22 years, just meeting the 60 ppb alternative standard reduced the percent of children exceeding the
23 60 ppb exposure benchmark by over 90 percent compared to just meeting the existing standard.
24 Reductions in lung function risk were also much higher than reductions in mortality and hospital
25 admissions risk. Just meeting the 65 ppb standard results in reductions in lung function risk by
26 over 25 percent in most locations and years, and just meeting the 60 ppb standard results in
27 reductions by over 40 percent in most locations and years.
28 There is general consistency in the city-to-city patterns of reductions in the exposure and
29 lung function risk metrics, although the decreases in lung function risk are less than half as large
30 as the reductions in the percent of children exceeding the 60 ppb exposure benchmark (with the
31 clear exception of New York city, which we will discuss further below). The patterns of
32 reductions in mortality and hospital admission risk are generally consistent with the patterns for
16 As noted previously, changes in the spatial extent of the urban case 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 case
study area. For example, we found that we bias the risk estimates low when using urban case study area
definitions that include only urban core counties and not the counties with monitors experiencing the most
reductions in O3
9-34
-------
1 exposure and lung function risk for 2007, with the exception of Houston and Philadelphia.
2 However, for 2009, the patterns for mortality and hospital admission risk are quite different, both
3 from the 2007 results, and from the exposure and lung function risk results. This is due to the
4 generally lower O^ concentrations in 2009, which results in a greater number of days with
5 predicted increases in 63 concentrations at low concentrations, fewer days with very high
6 concentrations where predicted reductions in Os occur, and a smaller predicted decrease in Os
7 concentrations on those high days. This affects the mortality and hospital admissions risk more
8 than the exposure and lung function risk metrics because those metrics incorporate thresholds,
9 and therefore are not responsive to changes in O^ concentrations below those thresholds.
10 Additional considerations are important in interpreting the reduction in exposure and risk
11 between the existing standard and alternative standards. The REA analyses focus on reducing
12 peak Oj, concentrations, in particular the 4* high Oj, concentration averaged over 3-years so as to
13 simulate meeting the existing standard or various alternative standards. In addition, the air
14 quality adjustments are based on applying reductions in U.S. anthropogenic emissions. In this
15 way, the adjusted air quality reflects day-to-day 63 concentrations that could occur when
16 focusing on reducing high 63 concentrations rather than on reducing mean 63 concentrations. In
17 addition, because the analyses do not include reductions of 03 precursor emissions from sources
18 other than U.S. anthropogenic emissions (e.g. international emissions, biogenics, etc), the 63
19 concentrations in the adjusted air quality account for OT, created from natural and international
20 sources, even if 100 percent emissions reductions are applied to U.S. anthropogenic sources in
21 adjusting air quality scenarios.
22 Finally, with respect to the epidemiology based analyses, we note that 2007, which had
23 generally higher 63 concentrations than 2009, had more days where 63 concentrations decreased
24 as a result of adjusting peak 63 concentrations to just meet alternative standards. Thus just
25 meeting alternative standards resulted in net decreases in risk in all locations, with the exception
26 of Houston for just meeting the 70 ppb alternative standard. In contrast, 2009, which had
27 generally lower concentrations than 2007, had more days in the range where Os concentrations
28 were increased as a result of adjusting peak 03 concentrations to just meet alternative standards,
29 and thus the patterns reflect some locations where mortality and hospital admissions risk
30 increases. However, for 2009, in all locations, when just meeting the lowest alternative standard
31 of 60 ppb, mortality and hospital admission risks are decreased relative to just meeting the
32 existing standard.
33
9-35
-------
1 9.4 OVERALL ASSESSMENT OF REPRESENTATIVENESS OF EXPOSURE AND
2 RISK RESULTS
3 9.4.1 Representativeness of Selected Urban Case Study Areas in Reflecting Areas Across
4 the Nation with Elevated Risk
5 We selected urban case study areas for the exposure and risk analyses based on several
6 criteria (e.g. recent elevated 63 concentrations and presence of at-risk populations and lifestages)
7 we identified as likely indicators of areas and populations likely to experience high Os exposures
8 and risks (see Section 7.3.1). We then conducted several analyses to determine the extent to
9 which our selected urban case study areas actually represent the highest mortality and morbidity
10 risk areas. We compared the distributions of risk characteristics17 and mortality risk (based on
11 recent 63 concentrations) for the 12 urban case study areas used in the epidemiology-based risk
12 assessment with the corresponding national distributions. We also evaluated the degree to which
13 our selected urban case study areas represent the patterns of Os concentration changes
14 experienced by the overall U.S. population.
15 Based on the comparisons of distributions of risk characteristics, the selected urban case
16 study areas represent urban case study areas that are among the most populated in the U.S., have
17 relatively high peak 63 concentrations, and capture well the range of city-specific mortality risk
18 effect estimates. These three factors alone would suggest that the case study urban case study
19 areas should capture well the overall risk for other heavily populated urban case study areas in
20 the nation, with a potential for better characterization of the high end of the risk distribution. The
21 selected urban case study areas do not include those with the highest numbers of some at-risk
22 populations or lifestages, specifically older people with high baseline mortality rates. However,
23 most locations in the U. S. (except Florida) with high percentages of older people have low
24 overall populations, less than 50,000 people in a county, or low O?, concentrations. This suggests
25 that while the risk per exposed person per ppb of 63 may be higher in these locations, the overall
26 risk to the population is likely to be within the range of risks represented by the urban case study
27 locations.
28 Based on the comparisons of distributions of short-term 63 exposure mortality risk (using
29 the percent of mortality metric) for recent OT, concentrations, the 12 selected urban case study
30 areas are representative of the full distribution of U.S. Ch-related mortality risk in urban case
31 study areas. Two of the selected areas, New York and Philadelphia are representative of the
32 highest end of the distribution of short-term 03 mortality risk. Overall, Os mortality risk for
33 short-term 63 exposures in the 12 urban study areas are representative of the full distribution of
17 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).
9-36
-------
1 U.S. urban Os-related mortality, representing both high end and low end risk counties. For the
2 long-term Os exposure mortality risk metric (again using the percent of mortality), the 12 urban
3 study areas are representative of the central portion of the distribution of risks across all U.S.
4 counties; however, the selected 12 urban case study areas do not capture the very highest (greater
5 than 98th percentile) or lowest (less than 25th percentile) ends of the national distribution of long-
6 term exposure-related Os-related risk.
7 9.4.2 Representativeness of Selected Urban Case Study Areas in Reflecting
8 Responsiveness of Risk to Just Meeting Existing and Alternative Os Standards
9 While we selected urban case study areas to represent those populations likely to
10 experience elevated risks from Os exposure, we did not include among the selection criteria the
11 responsiveness of Os in the urban case study area to decreases in 63 precursor emissions that
12 would be needed to just meet existing or alternative standards.
13 In our preliminary evaluations of risk modeling results, we observed a consistent
14 presence of days with low to midrange starting 63 concentrations for which 63 concentrations
15 (using the 8-hour maximum metric) increased after adjustments to just meet the existing and
16 alternative standards across the selected urban case study areas. As noted above, this led to
17 estimates of increased risk on those days, and in some cases, estimates of increased risk over the
18 course of the 03 season, reflecting both the magnitude and frequency of the predicted increases
19 relative to the predicted decreases in 63 concentrations. As explained above, this pattern was
20 more pronounced when using a more spatially limited definition of the urban case study areas,
21 but even when using the CBS A definitions, there were still days when the area-wide average Os
22 increased, primarily due to predicted increases in 63 in the core counties of the urban case study
23 areas.
24 In order to better understand how prevalent this type of air quality response was across
25 the U.S., we conducted several additional analyses of Os concentrations. These included
26 evaluations of trends at Os monitors during a period of time with significant Os precursor
27 emission reductions, and evaluations of temporal and spatial patterns of 63 changes across the
28 U.S., based on air quality modeling results, to simulate how Os would change across the U.S. in
29 response to NOx (and VOC) emissions reductions (relative to recent 2007 levels) similar to those
30 used in the HDDM adjustments (see section 9.2.1 above). The latter analysis includes an
31 assessment of the association of different types of Os responses with population counts to help
32 characterize the degree to which populations in the U.S. experience Os conditions like those in
33 the selected 15 urban case study areas (see Chapter 8).
34 Overall, both types of analyses showed that decreases in Os precursor emissions lead to
35 decreases in Os concentrations in areas with higher starting Os concentrations, which tend to be
36 rural or suburban case study areas, and on days with higher Os concentrations. The analyses also
9-37
-------
1 indicate that in urban core areas (those with high levels of fresh NOx emissions), decreases in
2 NOx emissions can lead to increases in 63, primarily for days when initial 63 concentrations are
3 suppressed due to NOx titration. The observed widespread decreases of median Os in suburban
4 and rural locations when NOx emissions are decreased suggest the efficacy of large NOx
5 emissions reductions on reducing Oj over large regions of the country.
6 These results suggest that many of the urban case study areas may show Os responses
7 that are typical of other large urban case study areas in the U.S., but may not represent the
8 response of Oj in other populated areas of the U.S., including suburban case study areas, smaller
9 urban case study areas, and rural areas. These smaller urban case study areas would be more
10 likely than our urban case study areas to experience area-wide average decreases in mean Oj,
11 concentrations as Os standards are met. Even though large urban case study areas throughout the
12 U.S. have high population density, 73 percent of the U.S. population lives outside of these high
13 population density areas18, and thus, a large proportion of the population is likely to experience
14 greater mortality and morbidity risk reductions in response to reductions in 8-hour Os
15 concentrations than are predicted by our modeling in the selected urban case study areas. The
16 analyses presented in Section 8.2.3.2 show that populations in the case study areas we selected
17 are approximately twice as likely to experience increasing mean Os concentrations as
18 populations in the U.S. as a whole. Because our selection strategy for risk modeling was focused
19 on identifying areas with high risk, we tended to select large urban population centers. As
20 discussed in the previous section, this strategy was largely successful in including those urban
21 case study areas in the upper end of the Oj risk distribution. However, this also has led to an
22 overrepresentation of the populations living in locations where we estimate increasing mean
23 seasonal Os in response to adjusting air quality to just meet the existing and alternative standards
24 using NOx emissions reductions. The implication of this is that our estimates of mortality and
25 morbidity risk reductions for the selected urban case study areas are likely to understate the
26 average risk reduction that would be experienced across the population and should not be seen as
27 representative of potential risk reductions for most of the U.S. population.
28 9.5 OVERALL ASSESSMENT OF CONFIDENCE IN EXPOSURE AND RISK
29 RESULTS
30 As with any complex analysis using estimated parameters and inputs from numerous data
31 sources and models, there are many sources of uncertainty that may affect our exposure and risk
32 estimates. These sources of uncertainty are discussed in each of the chapters related to air
33 quality, exposure, lung function risk, and epidemiology based mortality and morbidity risk. The
18 High population density areas are defined here as locations with population densities greater than 1000
people/km2
9-38
-------
1 overall effect of the combined set of uncertainties on confidence in the interpretation of the
2 results of the analyses is difficult to quantify. However, we provide our judgment of our overall
3 confidence here, with an understanding that alternative judgments may also be supported.
4 The degree to which each analysis was able to incorporate quantitative assessments of
5 uncertainty differed, due to differences in available information on uncertain parameters and
6 complexities in propagating uncertainties through the models. In general, we followed the World
7 Health Organization tiered approach to uncertainty characterization (WHO, 2008), which
8 includes both quantitative and qualitative assessments. Each chapter includes a table identifying
9 and characterizing the potential impact of key uncertainties on risk estimates, including the
10 degree to which we were able to quantitatively address those uncertainties.
11 In considering our overall confidence in the results, there are several key considerations
12 discussed below related to sources of uncertainty which we were not able to fully quantify, but
13 which may have a large impact on both overall confidence and confidence in individual analyses.
14 9.5.1 Uncertainties in Modeling Os Responses to Meeting Standards
15 There is inherent uncertainty in all deterministic air quality models, such as CMAQ, the
16 photochemical grid model used to develop the model-based Os adjustment methodology.
17 Evaluations of air quality models against observed pollutant concentrations build confidence that
18 the model performs with reasonable accuracy despite both structural and parametric
19 uncertainties. A comprehensive model performance evaluation provided in Appendix 4-B shows
20 generally acceptable model performance that is equivalent to or better than typical state-of-the
21 science regional modeling simulations described in Simon et al. (2012). Two additional sources
22 of uncertainties in the HDDM adjustment methodology are the applicability of HDDM
23 sensitivities over large emissions perturbations and the variability in data used to create
24 regressions which allowed the application of these sensitivities to ambient data. Both sources of
25 uncertainty are shown to be reasonably small in chapter 4 with the first having a mean error of
26 less than Ippb for 50% NOx cuts and less than 4 ppb for 90% NOx cuts. The uncertainty
27 introduced from the application of regressions to determine sensitivities were quantified by
28 propagating uncertainties in the sensitivities through to uncertainties in the final predicted Oj
29 concentrations which had standard errors less than 1.4 ppb for all adjustment scenarios. New
30 York and Los Angeles had the largest uncertainties in these two areas due to the fact that they
31 required the largest reductions in NOx emissions. Uncertainties stemming from the application
32 of 8-months of model data to 5-years of ambient data and the across-the-board emissions cut
33 assumptions are further discussed in chapter 4 but are not expected to substantially degrade
34 confidence in the air quality results.
9-39
-------
1 9.5.2 Uncertainties in Modeling Exposure and Lung-function Risk
2 With regard to the exposure and lung-function risk estimates, the modeling explicitly
3 incorporates population variability in many of the modeling inputs. We did not attempt to
4 probabilistically incorporate the many sources of uncertainty in model parameters or input data
5 due to limitations in the ability to specify distributions characterizing our confidence in those
6 variables. To explore the impacts of some of the more important sources of uncertainty, we
7 conducted a limited set of sensitivity analyses. For the exposure assessment, the estimate of
8 repeated exposures above exposure benchmarks is based on the limited set of diaries of activity
9 data available in the Consolidated Human Activity Database (CHAD) database (see Chapter 5).
10 The method for constructing activity patterns over the course of an Oj, season may not fully
11 capture the behavior of children who have systematically high outdoor activity levels. As a
12 result, while we are able to report the percent of children with two or more exposures, modeling
13 of the distribution of multiple exposures is limited, and the ability to identify the percent of the
14 population with unusually high numbers of multiple exposures is not possible.
15 For the lung function risk assessment, sensitivity analyses indicate that the MSS model
16 parameter related to the impact of the ventilation rate was most influential in determining the
17 estimated number of children with FEVi decrements greater than 10 percent. Estimates of lung
18 function decrements are also influenced by how much variability in individual response is
19 assumed in the MSS model. Sensitivity analyses indicate that when a greater amount of
20 variability is allowed in the MSS model, the percent of children ages 5-18 with FEVi decrements
21 greater than 10 percent can increase substantially. In addition, we performed analyses to
22 understand the age-related factors in APEX that could influence the estimated FEVi decrements.
23 It was found that the four most influential factors influencing the relationship between the
24 predicted FEVi decrement and age are the decreasing level of exertion, the decreasing equivalent
25 ventilation rate (with increasing age), the higher time spent outdoors by children, and the higher
26 exposure concentration experienced by children while outdoors. These all lead to children having
27 higher FEVi decrements than adults, and are more influential than the MSS model age term.
28 9.5.3 Uncertainties in Modeling Epidemiological-based Risk
29 A major issue in using the results of the epidemiology studies in estimating risk is the
30 narrow geographic definition used for urban case study areas in the epidemiology studies. In
31 many of the urban case study areas, we observe two distinct patterns of 63 response to the
32 reductions in precursor emissions we evaluated to just meet the existing and alternative standard
33 levels. The first pattern generally occurs in areas outside the urban core (e.g. suburban and rural
34 areas), and on days when 63 concentrations are on the higher end of the distribution of 63
35 concentrations, and is characterized by predicted decreases in 8-hour Os concentrations. These
9-40
-------
1 tend to be the locations where the highest 8-hour design values occur. The second pattern
2 generally occurs in the urban core, and on days when 63 concentrations are on the lower end of
3 the distribution of Os concentrations, and is characterized by predicted increases in 8-hour Os
4 concentrations. The narrow definitions of urban case study areas used in the epidemiological
5 studies generally included the urban core areas, but did not include all of the suburban or rural
6 areas. The narrow geographic definitions led to a clear downward bias in the estimates of risk
7 changes that would be associated with just meeting the standards in the urban case study areas,
8 because the risk changes would reflect the locations with a tendency towards increases in 8-hour
9 Os, but would not include locations outside the urban core with decreases in 03. In many cases,
10 the narrowly defined geographic definitions used in the epidemiology studies did not even
11 include the location with the monitor that was violating the standard. We addressed this bias by
12 expanding the urban case study area to the CBSA. However, this adds additional uncertainty to
13 the risk estimates, and reduces our confidence that we have a good match between the basis of
14 the C-R function (just urban core locations) and the risk analysis context (including both urban
15 core counties and other counties in the CBSA). A clear implication of this decision is that the
16 absolute incidence estimates will be larger than if the analysis was limited to a smaller number of
17 counties. For this reason, we have placed more emphasis on risk metrics that have been
18 normalized for population size (e.g. risks per 100,000 population and percent risk), so as to
19 facilitate comparisons between cities of different population sizes and to reduce the influence of
20 population size on the risk metrics.
21 The epidemiology studies used as the source for C-R functions for short-term exposure
22 mortality and morbidity endpoints all use time-series approaches to estimate the effect of daily
23 variations in 63 concentrations on daily mortality or morbidity incidence. The effect estimates
24 developed in these epidemiology studies were based on air quality and health information
25 observed over periods of time in the past (1987-2000). These effect estimates were based on day-
26 to-day variations in area-wide 63 concentrations estimated from observed concentrations at
27 monitors that reflect a specific set of emissions and atmospheric conditions. In our REA
28 analyses, we apply these effect estimates to adjusted air quality scenarios that are reflective of
29 substantial changes in 63 concentrations across an area due to, in some cases, large decreases in
30 NOx and VOC emissions reductions. The resulting spatial and temporal patterns of Os may not
31 be the same as the spatial and temporal patterns of 63 that existed at the time of the
32 epidemiology study. The potential for different spatial and temporal patterns in 63
33 concentrations between the adjusted air quality scenarios and the air quality observed during the
34 epidemiology study period potentially adds uncertainty to the estimates of risk, as it is not clear
35 the degree to which the exposure surrogate used in the epidemiology study correlates with the
36 exposure surrogate used in the risk analysis. The degree of this potential uncertainty increases
9-41
-------
1 with the amount of emissions reductions applied in the adjusted air quality scenario. This is
2 because as the amount of emissions reductions applied increases, the spatial and temporal
3 patterns of Os concentrations become increasingly different from those patterns observed for
4 recent O?, concentrations (2006-2010) that are more similar, although not identical (due to
5 reductions in NOx between 2000 and 2006), to the patterns for the time period covered by the
6 epidemiology studies (1987-2000). We are not able to quantify the effect or magnitude of this
7 uncertainty, because we do not know the relationship between Oj, variability and the C-R
8 functions. However, to the extent that the uncertainty is shown to be important, it seems
9 reasonable to conclude that the larger the adjustment to the O?, distributions, the more likely there
10 could be a mismatch in the exposure surrogates.
11 Overall, these sources of uncertainty cause us to have reduced confidence in estimates of
12 short-term risk based on modeling the larger (CBSA-based) study areas using the multi-city time
13 series-based effect estimates. This reduces the utility of the risk assessment in directly informing
14 the decision regarding the level of the standard since we have lower confidence in estimates of
15 absolute risk associated with a given standard level. However, the risk assessment can still be
16 useful in providing estimates of the general magnitude and direction of changes in risk associated
17 with an alternative standard level.
18 9.6 OVERALL INTEGRATED CHARACTERIZATION OF RISK IN THE
19 CONTEXT OF KEY POLICY RELEVANT QUESTIONS
20 Our analyses set out to inform two questions: 1) what are the magnitudes of exposures of
21 concern and risks for Os-related health effects that are estimated to occur with Os concentrations
22 that just meet the existing O^ standard?; and 2) to what extent do alternative standards reduce
23 estimated exposures and risks of concern attributable to Os, focusing on at-risk populations and
24 lifestages? In evaluating risk, we did not limit the assessment to just the absolute risk that is
25 attributable to U.S. or North American emissions, as this is not relevant to answer the two
26 questions. Instead, we estimated total risk from all Oj concentrations and the distribution of risk
27 over the range of Os concentrations. Our estimates of changes in risk from meeting alternative Os
28 standard levels relative to meeting the existing standard reflect only the impact of reductions in
29 U.S. precursor emissions on Oj distributions, recognizing that these emissions are most likely to
30 be affected by implementation of the standards.
31 To inform these questions, we conducted air quality, exposure, and risk analyses for
32 selected urban case study areas. We evaluated changes in the distribution of Os concentrations
33 along the full range of Os concentrations down to zero. We have utilized a new method
34 (compared to the Oj NAAQS review completed in 2008) for estimating Oj concentrations
35 consistent with attaining existing and alternative standards, based on modeling the response of
9-42
-------
1 Os concentrations to reductions in U.S. anthropogenic NOx and VOC emissions, using the
2 HDDM capabilities in CMAQ. This modeling incorporates all known emissions, including
3 emissions from non-anthropogenic sources and anthropogenic emissions from sources in and
4 outside of the U.S. As a result, background O^ concentrations are directly modeled and, therefore,
5 do not need to be separately specified. Application of this approach also addresses the
6 recommendation by the National Research Council of the National Academies (NRC, 2008) to
7 explore how emissions reductions might affect temporal and spatial variations in 63
8 concentrations, and to include information on how NOx versus VOC control strategies might
9 affect exposure to O^ and potential risks.
10 We estimated exposures and risks using several different metrics. Consistent with the
11 available evidence, we estimated the percentages of different study populations and lifestages
12 with exposures exceeding several health-based exposure benchmarks. We estimated lung
13 function risks based on a model of individual risk of lung function decrements that incorporates a
14 dose-equivalent threshold and individual exposures, activity levels, and physiology. We
15 estimated mortality and morbidity risks based on non-threshold C-R functions derived from
16 epidemiology studies. These three different analyses result in differing sensitivities of results to
17 changes in the O?, concentration distribution. Because the three metrics are affected differently in
18 the analyses by changes in Oj, at low concentration levels, it is important to understand these
19 changes in Os at low concentrations in interpreting differences in the results across metrics.
20 We also evaluated the degree to which exposures of concern and lung function risk were
21 reduced in the portions of urban case study areas (urban core areas) that were more likely to
22 experience an increase in low concentrations of Os, and in some cases an overall net increase in
23 epidemiology based mortality and morbidity risk (results for this assessment are presented in
24 Appendix 9A). We compared these estimates of changes in exposures and lung function risk to
25 estimates of changes in exposures and lung function risk in the areas outside of the urban core
26 areas to judge whether for exposures of concern and lung function risk we see the same pattern
27 of risk reduction between those areas.
28 Both exposures of concern and lung-function risk estimates in the core urban case study
29 areas showed similar patterns compared with the areas outside the urban cores when just meeting
30 the existing and potential alternative standards. Thus, we observe that in urban core areas which
31 in some cases showed overall increases in epidemiology based mortality and morbidity risk
32 when looking across these same air quality scenarios (see section 9.5.3), we generally see
33 reductions in exposures of concern and lung function risk. These findings illustrate that
34 populations within core urban case study areas are likely to experience risk reductions for health
35 endpoints reflected in the exposure and lung-function analyses.
9-43
-------
1 The mortality and morbidity risk assessment is the analysis that is most sensitive to the
2 increases in 63 in the lower part of the distribution of initial 63 concentrations at some monitors
3 and on some days after meeting the existing and alternative standards in some urban case study
4 areas. As demonstrated in the heat maps (Figures 7-2 and 7-3), the increases in 03 (and resulting
5 estimated increases in risk) occur largely on days with initial 63 concentrations in the range of 10
6 to 40 ppb. In addition, mean Os concentrations for the urban case study areas change little
7 between air quality scenarios for meeting the existing and alternative standards, because mean
8 concentrations reflect both the increases in 63 at lower concentrations and the decreases in 63
9 occurring on days with high 03 concentrations. This leads to small net changes in mortality and
10 morbidity risk estimates for many of the urban case study areas. For New York, we find there is
11 a larger decrease relative to other urban case study areas (nearly five times as large as the next
12 largest result for Los Angeles), in mortality and respiratory hospital admissions when just
13 meeting the 65 ppb alternative standard compared to just meeting the existing standard,
14 reflecting the large degree of air quality adjustment needed to meet the standard at all monitors in
15 New York. Both the net change in risk and the distribution of risk across the range of 63
16 concentrations in the urban case study areas may be relevant in considering the degree of
17 additional protection provided by just meeting existing and alternative standards.
18 The dampened response of short-term mortality risk can be contrasted with lung function
19 risk estimates based on application of results from controlled human exposure studies. The lung
20 function risk estimates primarily reflect changes in the upper end of the O^ distribution and
21 reflect counts of exceedances of lung function decrement benchmarks, rather than summing risks
22 across all days in the season. In addition, lung function risks are based on detailed micro-
23 environmental exposure modeling which uses individual monitor values instead of composite
24 monitor values, thereby resulting in less dampening of spatial variability in 63 within a given
25 urban study area.
26 The exposure benchmark analysis is the least sensitive to changes in 63 in the lower part
27 of the distribution of initial Os concentrations, because the lowest of the exposure benchmarks is
28 at 60 ppb, well above the portion of the distribution of initial O^ concentrations that increased.
29 Since the modeled exposures will always be less than or equal to the monitor concentrations, a
30 benchmark of exposure at 60 ppb is above the range of OT, concentrations where the HDDM
31 approach estimates increases in concentrations. Thus, this metric is most reflective of the
32 decreases in 63 at high concentrations that are expected to result from just meeting the existing
33 and alternative standards.
34 The lung function risk analysis is less sensitive than the mortality and morbidity risk
35 assessments to increases at very low concentrations of Os, because the risk function is logistic
36 and shows little response at lower 03 dose rates that tend to occur when ambient concentrations
9-44
-------
1 are lower (generally less than 20 ppb for the 10 percent FEVi decrement and generally less than
2 40 ppb for the 15 percent FEVi decrement). However, because there are still some increases in
3 Os concentrations that occur in the 50 to 60 ppb range where the estimated risk is more
4 responsive, there may be some reduction in the magnitude of the risk decrease (this is evident
5 when comparing the lung function risk metric with the exposure benchmark metric in figure 9-
6 8).
7 The exposure-based lung function risk assessment is based on controlled human exposure
8 studies which studied responses in healthy adults. Although the lung function model based on
9 this population shows less responsiveness at lower ambient concentrations, the applicability of
10 this model to the responses of more sensitive populations and lifestages, including children and
11 asthmatics, is uncertain. In addition, although the most complete information for generating an
12 exposure-response function is available for FEVI as a measure of lung function, there are other,
13 potentially more public health relevant effects, such as lung inflammation, which have also been
14 shown to respond to Os. As such, the lung function risk analysis should be seen as providing
15 useful but not complete information on risks of health responses to 63.
16 Exposures above health benchmarks and risks remain after adjusting 63 to just meet the
17 existing standard. The percentage of children with at least one 8-hour 03 exposure exceeding 60
18 ppb is greater than 10 percent in at least one of the five analytical years for all of the 15 urban
19 case study areas. The percent of children with a predicted decrement in lung function greater
20 than or equal to 10 percent is greater than 16 percent in at least one of the five analytical years
21 for all of the 15 urban case study areas, and for a 15 percent decrement is less than 7 percent for
22 all years and areas. Cb-attributable mortality is slightly less than one percent up to four percent of
23 total mortality across the 12 urban case study areas, with little variation between 2007 and 2009.
24 (Vattributable respiratory hospital admissions are between 2 and 3 percent across the 12 urban
25 case study areas, with little variation between 2007 and 2009. The percent attributable risk for
26 other morbidity endpoints is somewhat higher than for respiratory hospital admissions, but we
27 only estimated these endpoints for a more limited set of urban case study areas due to data
28 limitations.
29 The degree of reduction in exposures and risks when adjusting 63 from just meeting the
30 existing standard to just meeting lower alternative standard levels varies considerably between
31 metrics. The greatest degree of reduction occurs in exposures above the 60 ppb exposure
32 benchmark, followed by reductions in lung function decrements greater than or equal to 10
33 percent, with the smallest changes in mortality and respiratory hospital admissions. Although the
34 magnitude of reduction differs between the different exposure and risk metrics, there are
35 generally the same patterns of reductions for the exposure benchmark and lung function risk
36 metrics, showing consistent reductions across all 15 urban case study areas. Risk reductions also
9-45
-------
1 occur in most of the urban case study areas for mortality and respiratory hospital admissions.
2 However, these reductions are small, and reflect net changes in risk that include days with risk
3 increases as well as risk decreases. For most urban case study areas, the greatest incremental
4 reductions in exposures above the 60 ppb benchmark occurred when just meeting 70 ppb
5 compared to just meeting the existing standard. Just meeting lower standards of 65 ppb and 60
6 ppb had incrementally smaller reductions in the percent of children exposed above 60 ppb.
7 Incremental lung function risk reductions are more even between alternative standards, with
8 similar or greater incremental reductions for the 65 ppb and 60 ppb alternatives compared with
9 the incremental reductions for just meeting 70 ppb. Incremental reductions in mortality and
10 respiratory hospital admissions risk are small between alternative standards, but more urban case
11 study areas have somewhat larger risk reductions when comparing just meeting the 60 ppb
12 alternative to just meeting the 65 ppb standard, than when comparing 65 ppb to 70 ppb or 70 ppb
13 to 75 ppb. Long-term exposure mortality risk results show larger absolute estimates of mortality
14 risk and more consistent reductions across urban case study areas. However, percent changes in
15 long-term exposure mortality are similar to those for short-term exposure mortality.
16 In conclusion, we have estimated that exposures and risks remain after just meeting the
17 existing standards and that in many cases, just meeting alternative standard levels results in
18 reductions in those exposures and risks. Meeting alternative standards has larger impacts on
19 metrics that are not sensitive to changes in lower Os concentrations. When meeting the 70, 65,
20 and 60 ppb alternative standards, the percent of children experiencing exposures above the 60
21 ppb health benchmark falls to less than 20 percent, less than 10 percent, and less than 3 percent
22 in the worst Os year for all 15 case study urban case study areas, respectively. Lung function risk
23 also drops considerably as lower standards are met. When meeting the 70, 65, and 60 ppb
24 alternative standards, the percent of children with lung function decrements greater than or equal
25 to 10 percent falls to less than 21 percent, less than 18 percent, and less than 14 percent in the
26 worst 63 year for all 15 case study urban case study areas, respectively. Mortality from short-
27 and long-term Os exposures and respiratory hospitalization risk is not greatly affected by
28 meeting lower standards, reflecting the impact of increasing O?, on low concentration days, and
29 the non-threshold nature of the C-R function.
9-46
-------
1 9.7 REFERENCES
2 Frey, C. and J. Samet. 2012. CASAC Review of the EPA's Policy Assessment for the Review of
3 the O3 National Ambient Air Quality Standards (First External Review Draft - August
4 2012). U.S. Environmental Protection Agency Science Advisory Board, (Document
5 number EPA-CASAC-13-003).
6 Medina-Ramon, M.; A. Zanobetti and J. Schwartz. 2006. "The Effect of O3 and PMio on Hospital
7 Admissions for Pneumonia and Chronic Obstructive Pulmonary Disease: A National
8 Multicity Study. "American Journal of Epidemiology, 163(6), 579-588.
9 McDonnell, W.F.; P. W. Stewart; M. V. Smith; C. S. Kim C.S. and E. S. Schelegle. 2012.
10 "Prediction of lung function response for populations exposed to a wide range of O3
11 conditions." Inhalation Toxicology, 24:619-633.
12 Simon, H.; K. R. Baker; F. Akhtar; S. L. Napelenok; N. Possiel; B. Wells and B. Timin. 2013.
13 "A Direct Sensitivity Approach to Predict Hourly O3 Resulting from Compliance with the
14 National Ambient Air Quality Standard." Environmental Science and Technology, 47,
15 2304-2313.
16 Smith, RL; B. Xu and P. Switzer. 2009. "Reassessing the Relationship Between O3 and Short-
17 term Mortality in U.S. Urban Communities." Inhalation Toxicology, 21: 37-61.
18 U.S. Environmental Protection Agency. 2007. O3 Health Risk Assessment for Selected Urban
19 case study areas. Research Triangle Park, NC: Office of Air Quality Planning and
20 Standards. (EPA document number EPA 452/R-07-009). Available at:
21 .
22 U.S. EPA. 2013. Integrated Science Assessment for O3 and Related Photochemical Oxidants:
23 Final. Research Triangle Park, NC: U.S. Environmental Protection Agency. (EPA
24 document number EPA/600/R-10/076F).
25 World Health Organization. 2008. Part 1: Guidance Document on Characterizing and
26 Communicating Uncertainty in Exposure Assessment, Harmonization Project Document
27 No. 6. Published under joint sponsorship of the World Health Organization, the
28 International Labour Organization and the United Nations Environment Programme.
29 WHO Press, World Health Organization, 20 Avenue Appia, 1211 Geneva 27,
30 Switzerland.
31 Zanobetti, A. and J. Schwartz. 2008. "Mortality Displacement in the Association of O3 with
32 Mortality: An analysis of 48 Cities in the United States." American Journal of
33 Respiratory and Critical Care Medicine, 177:184-189.
34
9-47
-------
United States Office of Air Quality Planning and Standards Publication No. EPA-452/P-14-004a
Environmental Protection Air Quality Strategies and Standards Division February 2014
Agency Research Triangle Park, NC
------- |