EPA
United States
Environmental
Protection Agency
Particulate Matter National Ambient
Air Quality Standards:
Scope and Methods Plan
for Health Risk and Exposure
Assessment
February 2009
Office of Air Quality Planning and Standards
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
-------
DISCLAIMER
This planning document has been prepared by staff from the Office of Air Quality
Planning and Standards, U.S. Environmental Protection Agency. Any opinions, findings,
conclusions, or recommendations are those of the authors and do not necessarily reflect the
views of the EPA. This document is being circulated to facilitate consultation with the Clean Air
Scientific Advisory Committee (CASAC) and to obtain public review. Comments on this
document should be addressed to Beth Hassett-Sipple, U.S. Environmental Protection Agency,
Office of Air Quality Planning and Standards, C504-06, Research Triangle Park, North Carolina
27711 (email: hassett-sipple.beth@epa.gov).
-------
EPA-452/P-09-002
February 2009
Particulate Matter National Ambient Air Quality Standards:
Scope and Methods Plan
for Health Risk and Exposure Assessment
US Environmental Protection Agency
Office of Air and Radiation
Office of Air Quality Planning and Standards
Health and Environmental Impacts Division
Ambient Standards Group
Research Triangle Park, North Carolina 27711
-------
This page intentionally left blank.
-------
Table of Contents
List of Acronyms/Abbreviations iv
1 INTRODUCTION 1-1
1.1 Background on Last PM NAAQS Review 1-3
1.1.1 Overview of Health Risk Assessment for Fine Particles from Last
Review 1-4
1.1.2 Overview of Health Risk Assessment for Thoracic Coarse Particles
from Last Review 1-7
1.2 Goals for Framing the Assessments in the Current Review 1-7
1.3 Overview of Current Assessment Plan 1-8
1.3.1 Air Quality Assessment 1-8
1.3.2 Risk Assessment 1-9
1.3.3 Exposure Assessment 1-11
2 AIR QUALITY CONSIDERATIONS 2-1
2.1 Introduction 2-1
2.2 Air Quality Inputs to Risk Assessments 2-2
2.2.1 Recent Air Quality: PM2.5 2-2
2.2.2 Recent Air Quality: PMi0-2.5 2-3
2.2.3 Air Quality Data Related to Exceptional Events 2-3
2.3 Development of Estimates of PM Air Quality Assuming "Just Meeting" Current
NAAQS and Potential Alternative NAAQS 2-4
2.3.1 Background and Conceptual Overview 2-4
2.3.2 Historical Approach 2-5
2.3.3 Alternative Approach Under Consideration 2-6
2.4 Policy Relevant Background 2-7
2.5 Broader Air Quality Characterization 2-9
3 SCOPE AND APPROACH FOR THE HEALTH RISK ASSESSMENT 3-1
3.1 Introduction 3-1
3.2 Framework for the PM25 Health Risk Assessment 3-4
3.2.1 Air Quality Considerations 3-5
3.2.2 Selection of Health Effects Endpoint Categories 3-7
3.2.3 Specification of Concentration-Response Functions 3-8
3.2.4 Selection of Urban Study Areas 3-18
3.2.5 Baseline Health Effects Incidence Data and Demographic Data 3-19
3.3 Characterization of Uncertainty and Variability in the Context of the PM2 5 Risk
Assessment 3-21
3.3.1 Differentiating Between Uncertainty and Variability 3-21
3.3.2 Addressing Variability 3-22
3.3.3 Uncertainty Characterization - Qualitative Assessment 3-24
3.3.4 Uncertainty Characterization - Quantitative Analysis 3-25
3.4 Presentation of risk estimates to inform consideration of standards 3-28
3.5 National Scale Health Impact Assessment for Long-term Exposure Mortality
Related to PM2.5 Exposures 3-30
3.6 Overview of Health Risk Assessment Approach Developed for PMio-2.s 3-32
-------
4 SCOPE AND APPROACH FOR POPULATION EXPOSURE ANALYSIS 4-1
4.1 Introduction 4-1
4.2 The APEX Population Exposure Model 4-2
4.3 Measures of Exposure 4-9
4.4 Selection of Urban Areas and Time Periods 4-9
4.5 Development of Model Inputs 4-10
4.5.1 Population Demographics 4-10
4.5.2 Commuting 4-10
4.5.3 Ambient PM2.s Concentrations 4-11
4.5.4 Meteorological Data 4-11
4.5.5 Specification of Microenvironments 4-11
4.5.6 Indoor Sources 4-14
4.5.7 Activity Patterns 4-14
4.6 Pilot Exposure Assessment 4-18
4.7 Exposure Modeling Issues 4-20
4.8 Uncertainty and Variability 4-21
5 SCHEDULE AND MILESTONES 5-1
6 REFERENCES 6-1
Appendix Preparation of Model Inputs for the Pilot Exposure Assessment A-l
11
-------
List of Tables
Table 3-1 Provisional Summary of Short-Term PM2.5 Epidemiological Studies Being Considered
for Selecting C-RFunctions for PM2.5Risk Assessment 3-11
Table 3-2 Provisional Summary of Long-term Epidemiological Studies Being Considered for
Selecting C-R Functions for PM2.5 Risk Assessment 3-15
Table 3-3 Provisional List of Studies to be Considered for PMi0-2.5 Risk Assessment:
Cardiovascular and Respiratory Morbidity Associated with Short-term PMio-2.5
Exposure 3-34
Table 4-1 Microenvironments to be Modeled 4-8
Table 4-2 Studies in CHAD used in this analysis 4-16
Table 5-1 Key Milestones for the Risk and Exposure Assessment 5-2
List of Figures
Figure 1-1 Overview of Risk Assessment Model 1-10
Figure 1-2 Overview of Exposure Assessment Model 1-12
Figure 3-1 Detailed Overview of Risk Assessment Model 3-6
Figure 3-2 Overview of Uncertainty Analysis Approach 3-26
Figure 4-1 Overview of the APEX Model 4-5
Figure 4-2. The Mass Balance Model 4-13
in
-------
LIST OF ACRONYMS/ABBREVIATIONS
A/C
Act
AERMOD
AHRQ
AHS
APEX
AQS
ARIES
BASE
BenMAP
CARS
CAMx
CASAC
CEM
CHAD
CMAQ
CONUS
COV
C-R
CSA
CTM
EPA
eVNA
FEM
FIP
FRM
GAMs
GLMs
HCUP
HEI
IEC
IMPROVE
ISA
ISCST
km
ME
MENTOR
NAAQS
NAPS
NCEA
Air conditioning
Clean Air Act
American Meteorological Society/EPA Regulatory Model Improvement
Committee (AERMIC) air quality dispersion modeling system
Agency for Healthcare Research and Quality
American Housing Survey
EPA's Air Pollutants Exposure model, version 4
EPA's Air Quality System
Aerosol Research and Inhalation Epidemiological Study
Building Assessment Survey and Evaluation
Benefits Mapping Analysis Program
California Air Resources Board
Comprehensive Air quality Model
Clean Air Scientific Advisory Committee
Continuous Emission Monitoring
EPA's Consolidated Human Activity Database
Community Multiscale Air Quality
Continental United States
Coefficient of Variation
Concentration-response relationship
Consolidated Statistical Area
Chemical transport models
United States Environmental Protection Agency
enhanced Voronoi Neighbor Averaging
Federal Equivalent Method
Federal Implementation Plan
Federal Reference Method
Generalized additive models
Generalized linear models
Healthcare Cost and Utilization Project
Health Effects Institute
Industrial Economics, Incorporated
Interagency Monitoring of Protected Visual Environment
Integrated Science Assessment
EPA's Industrial Source Complex Short-Term model
Kilometer
Microenvironment
Modeling ENvironment for TOtal Risk
National Ambient Air Quality Standards
National Air Pollution Surveillance
National Center for Environmental Assessment
IV
-------
NEI
NEM
NERL
NCDC
NMMAPS
NCore
NOx
03
OAQPS
OAR
OMB
ORD
PM
PMX
PM2.5
PMiQ-2.5
PRB
QA
RR
SAB
REA
SD
SEDD
SHEDS-PM
SID
SO2
SOX
TEACH
TSP
VOC
National Emissions Inventory
NAAQS Exposure Model
National Exposure Research Laboratory
National Climatic Data Center
National Morbidity, Mortality, and Air Pollution Study
National Core Monitoring Network
Nitrogen oxides
Ozone
Office of Air Quality Planning and Standards
Office of Air and Radiation
Office of Management and Budget
Office of Research and Development
Particulate Matter
The legal definition for PMx, as defined in the Code of Federal
Regulations, includes both a 50% cut-point and a penetration curve. A
50% cut-point of X um diameter means that 50% of particles with
aerodynamic diameter of X are removed by the inlet and 50% pass
through the inlet and are collected on the filter. Depending on the specific
penetration curve specified, particles larger than X um aerodynamic
diameter are collected with an efficiently than decreases rapidly for
particles larger than X while the collection efficiency for particles smaller
than X increases rapidly with decreasing size until 100 % efficiency is
reached.
Particles with a 50% upper cut-point of 10± 0.5 um aerodynamic diameter
and a penetration curve as specified in the Code of Federal Regulations.
Particles with a 50% upper cut-point of 2.5 um aerodynamic diameter and
a penetration curve as specified in the Code of Federal Regulations.
Particles with a 50% upper cut-point of 10 um aerodynamic diameter and
a lower 50% cut-point of 2.5 um aerodynamic diameter.
Policy-Relevant Background
Quality assurance
Relative risk
Science Advisory Board
Risk and Exposure Assessment
Standard deviation
State Emergency Department Databases
Stochastic Human Exposure and Dose Simulation model for PM
State Inpatient Database
Sulfur Dioxide
Sulfur Oxides
Toxicity and Exposure Assessment for Children's Health
Total suspended particulate
Volatile organic compounds
-------
i 1 INTRODUCTION
2 The U.S. Environmental Protection Agency (EPA) is presently conducting a review of
3 the particulate matter (PM) national ambient air quality standards (NAAQS). EPA's overall plan
4 and schedule for this PM NAAQS review are presented in the Integrated Review Plan for the
5 National Ambient Air Quality Standards for Paniculate Matter (US EPA, 2008a). That plan
6 outlines the Clean Air Act (CAA) requirements related to the establishment and reviews of the
7 NAAQS, the process and schedule for conducting the current PM NAAQS review, and two key
8 components in the NAAQS review process: an Integrated Science Assessment (ISA) and a Risk
9 and Exposure Assessment (REA). It also lays out the key policy-relevant issues to be addressed
10 in this review as a series of policy-relevant questions that will frame our approach to determining
11 whether the current primary and secondary NAAQS for PM should be retained or revised.
12 The ISA prepared by EPA's Office of Research and Development (ORD), National
13 Center for Environmental Assessment (NCEA) provides a critical assessment of the latest
14 available policy-relevant scientific information upon which the NAAQS are to be based. The
15 ISA will critically evaluate and integrate scientific information on the health and welfare effects
16 associated with exposure to PM in the ambient air. The REA, prepared by EPA's Office of Air
17 and Radiation (OAR), Office of Air Quality Planning and Standards (OAQPS), will draw from
18 the information assessed in the ISA. The REA will include, as appropriate, quantitative
19 estimates of human and ecological exposures and/or risks associated with recent ambient levels
20 of PM, with levels simulated to just meet the current standards, and with levels simulated to just
21 meet possible alternative standards.
22 The REA will be developed in two parts addressing: (1) human health risk and exposure
23 assessment and (2) visibility impairment and other welfare-related effects assessment. This
24 document describes the scope and methods planned to conduct the human health risk and
25 exposure assessments to support the review of the primary (health-based) PM NAAQS. A
26 separate document describes the scope and methods planned to conduct quantitative assessments
27 to support the review of the secondary (welfare-based) PM NAAQS (U.S. EPA, 2009).
28 Preparation of these two planning documents coincides with the development of the first draft
1-1
-------
1 PM ISA (U.S. EPA, 2008b) to facilitate the integration of policy-relevant science into all three
2 documents.
3 This planning document is intended to provide enough specificity to facilitate
4 consultation with CASAC, as well as for public review, in order to obtain advice on the overall
5 scope, approaches, and key issues in advance of the conduct of the risk and exposure analyses
6 and presentation of results in the first draft REA. NCEA has compiled and assessed the latest
7 available policy-relevant science available to produce a first draft of the ISA and related Annexes
8 (US EPA, 2008b). The first draft ISA has been reviewed by staff and used in the development of
9 the approaches described below. This includes information on atmospheric chemistry, source
10 emissions, air quality, human exposure, and related health effects. CASAC consultation on this
11 planning document coincides with its review of the first draft ISA. CASAC and public
12 comments on this document will be taken into consideration in the development of the first draft
13 REA, the preparation of which will coincide and draw from the second draft ISA. The second
14 draft REA will draw on the final ISA and will reflect consideration of CASAC and public
15 comments on the first draft REA. The final REA will reflect consideration of CASAC and
16 public comments on the second draft REA. The final ISA and final REA will inform the policy
17 assessment and rulemaking steps that will lead to a final decision on the PM NAAQS.
18 This introductory chapter includes background on the current PM standards and the
19 quantitative risk assessment conducted for the last review; the key issues related to designing the
20 quantitative assessments in this review, building upon the lessons learned in the last review; and
21 an overview introducing the planned assessments that are described in more detail in later
22 chapters. The planned assessments are designed to estimate health risks and/or human exposures
23 that are associated with recent ambient levels, with ambient levels simulated to just meet the
24 current standards, and with ambient levels simulated to just meet alternative standards that may
25 be considered. The major components of the assessments (e.g., air quality analyses, quantitative
26 health risk assessment, risk characterization, and quantitative exposure assessment) briefly
27 outlined in the Integrated Review Plan (U.S. EPA, 2008a, section 5), are conceptually presented
28 in Figures 1-1 and 1-2, and are described in more detail below in Chapters 2, 3, and 4,
29 respectively. The schedule for completing these assessments is presented in Chapter 5.
1-2
-------
1 1.1 BACKGROUND ON LAST PM NAAQS REVIEW
2 As a first step in developing this planning document, we considered the work completed
3 in previous reviews of the primary NAAQS for PM (U.S., EPA 2008a, see section 1.3) and in
4 particular the quantitative assessments supporting those reviews. The most recent review of the
5 PM standards, completed in 2006 (71 FR 61144, October 17, 2006) \ evaluated the existing
6 NAAQS and concluded that the standards needed to be revised to provide increased public health
7 protection. The rationale for the final decision on the appropriate revisions to the primary PM
8 NAAQS included consideration of: (1) evidence of health effects related to short- and long-term
9 exposures to particles; (2) insights gained from the quantitative risk assessment; and (3) specific
10 conclusions regarding the need for revisions to the current standards and the elements of the PM
11 standards (i.e., indicator, averaging time, form, and level), that taken together, are requisite to
12 protect public health with an adequate margin of safety.
13 In summary, with regard to the primary standards for fine particles, EPA revised the
3
14 level of the 24-hour PM2.5 standard to 35 |ig/m , retained the level of the annual PM2.5 annual
3
15 standard at 15 |ig/m , and revised the form of the annual PM2.5 standard by narrowing the
16 constraints on the optional use of spatial averaging. With regard to the primary standards for
17 thoracic coarse particles, EPA retained PMio as the indicator for purposes of regulating the
18 coarse fraction of PMio (referred to as thoracic coarse particles or coarse-fraction particles;
19 generally including particles with a nominal mean aerodynamic diameter greater than 2.5 jim
20 and less than or equal to 10 jim, or PMio-2.s). Specifically, EPA retained the 24-hour PMio
21 standard at 150 |ig/m3 and revoked the annual PMio standard because available evidence
22 generally did not suggest a link between long-term exposure to current ambient levels of thoracic
23 coarse particles and health or welfare effects.
24 In the last PM NAAQS review, EPA focused on particle mass and primarily
25 distinguished between two categories of particle pollution based on size (i.e., fine- and thoracic
26 coarse-fraction particles), and conducted parallel evaluations of the available scientific evidence
27 relating to each category. The importance of specific PM components and sources were
See also http://www.epa.gOv/ttn/naaqs/standards/pm/s jm_index.html.
1-3
-------
1 evaluated within the context of this basic size differentiation. In that review, it was determined
2 that size-fractionated particle mass, rather than particle composition, remained the most
3 appropriate approach for addressing ambient PM. The EPA conducted a quantitative health risk
4 assessment2 for selected health endpoints to provide additional information and insights that
5 could help inform decisions on the standards. The limitations of such an assessment were clearly
6 articulated.3 These assessments are briefly described below. EPA did not conduct an exposure
7 assessment for the PM NAAQS review completed in 2006.
8 1.1.1 Overview of Health Risk Assessment for Fine Particles from Last
9 Review
10 The approach used in the last PM NAAQS review to develop quantitative risk estimates
11 associated with exposures to fine particles, using PM2.5 as the indicator, was built upon the more
12 limited risk assessment conducted during the review completed in 1997 (Abt Associates, 1996,
13 2002). The expanded and updated assessment conducted for the review completed in 2006
14 included estimates of risks of mortality (total non-accidental, cardiovascular, and respiratory),
15 morbidity (hospital admissions for cardiovascular and respiratory causes), and respiratory
16 symptoms (not requiring hospitalization) associated with recent short-term (daily) ambient PM2.5
17 levels and risks of total, cardiopulmonary, and lung cancer mortality associated with long-term
18 exposure to PM2.s in nine urban areas. Nine areas were included in this assessment to provide
19 some sense of the variability in the PM2.s-related risk estimates across the U.S. including:
20 Boston, MA; Detroit, MI; Los Angeles, CA; Philadelphia, PA; Phoenix, AZ; Pittsburgh, PA; San
21 Jose, CA; and St. Louis, MO.
22 EPA recognized that there were many sources of uncertainty and variability inherent in
23 the inputs to this assessment and that there was a high degree of uncertainty in the resulting
The risk assessment was discussed in the Staff Paper (U.S. EPA, 2005, Section 4) and presented more fully in a
technical support document, Particulate Matter Health Risk Assessment for Selected Urban Areas (Abt Associates,
2005). The assessment scope and methodology were developed with considerable input from the CASAC Panel and
the public, with CASAC concluding that the general assessment methodology and framework were appropriate
(Hopke, 2002).
3 The EPA continues to support the development and application of risk assessment methods with the goal of
improving the characterization of risks and the communication of uncertainties in such risk estimates.
1-4
-------
1 PM2 5 risk estimates. Such uncertainties generally related to a lack of clear understanding of a
2 number of important factors, including, for example:
3 . the shape of concentration-response functions, particularly when effect thresholds
4 could neither be discerned nor determined not to exist;
5 . issues related to selection of appropriate statistical models for the analysis of the
6 epidemiologic data;
7 . the role of potentially confounding and modifying factors in the concentration-
8 response relationships;
9 . the method for simulating how daily PM2 5 ambient concentrations would likely
10 change in any given area upon meeting a particular standard, since strategies to
11 reduce emissions had not yet been defined; and
12 . the issue of whether there would be differential reductions in the many components
13 within PM2 5 and, if so, whether this would result in differential reductions in risk.
14 While some of these uncertainties were addressed quantitatively in the form of estimated
15 confidence ranges around central risk estimates, other uncertainties and the variability in key
16 inputs were not reflected in these confidence ranges, but rather were addressed through separate
17 sensitivity analyses or characterized qualitatively (U.S. EPA, 2005, Chapter 4; Abt Associates,
18 2005). The concentration-response relationships used in the quantitative risk assessment were
19 based on findings from human epidemiologic studies that relied on fixed-site, population-
20 oriented, ambient monitors as a surrogate for actual ambient PM2.5 exposures. The assessment
21 included a series of base case estimates that, for example, included various cutpoints intended as
22 surrogates for alternative assumed population thresholds. Other uncertainties were addressed in
23 various sensitivity analyses (e.g., the use of single- versus multi-pollutant models, use of single-
24 versus multi-city models, use of a distributed lag model) and had a more moderate and often
25 variable impact on the risk estimates in some or all of the cities.
1-5
-------
1 Key observations and insights from the PM2 5 risk assessment, together with important
2 caveats and limitations, were discussed in Section II.B of the 2006 proposal notice (71 FR 2637
3 to 2641, January 17, 2006). In general, estimated risk reductions associated with going from just
4 meeting the current suite of PM2.5 standards to just meeting alternative suites of annual and 24-
5 hour standards for all the various assumed cutpoints showed patterns of increasing estimated risk
6 reductions as either the annual or 24-hour PM2.5 standard, or both, were reduced over the range
7 considered in the assessment, and the estimated percentage reductions in risk were strongly
8 influenced by the assumed cutpoint level (EPA, 2005, see Figures 5-1, 5-2, 5A-1, and 5A-2). In
9 comparing the risk estimates for the only two locations (Philadelphia, PA and Los Angeles, CA)
10 that were included in both this assessment and the prior assessment, the magnitude of the risk
11 estimates associated with just meeting the current annual PM2.5 standard, in terms of percentage
12 of total incidence, were very similar for premature mortality associated with long-term
13 exposures.
14 In making final decisions for the PM2.5 NAAQS in 2006, the Administrator relied
15 primarily on evidence-based considerations to inform his conclusions on the levels for the 24-
16 hour and annual standards. The Administrator believed, at that time, that the estimates of risks
17 likely to remain upon attainment of the 1997 suite of PM25 standards were indicative of risks that
18 could be reasonably judged important from a public health perspective, and, thus, supported
19 revision of the standards. However, the Administrator judged that the quantitative risk
20 assessment had important limitations and did not provide an appropriate basis for selecting either
21 the level of the 24-hour or annual PM2.5 standard. The Administrator more heavily weighed the
22 implications of the uncertainties associated with the quantitative risk assessment than CAS AC in
23 their comments on the proposed rulemaking and disagreed with CASAC and many public
24 commenters that the risk assessment results could appropriately serve as a primary basis for a
25 decision for the level of either the 24-hour or the annual PM2.5 standards.4
' See discussion in Section II.F of the preamble to the final rule, 71 FR 61167-61177, October 17, 2006.
1-6
-------
1 1.1.2 Overview of Health Risk Assessment for Thoracic Coarse Particles from
2 Last Review
3 The general overview and discussion of key components of the quantitative risk
4 assessment used to develop risk estimates for PM2.5 presented above is also applicable to the risk
5 assessment conducted for PMi0-2.5 as part of the last review. However, the scope of the risk
6 assessment for PMi0-2.5 was much more limited than that for PM2.5, reflecting the much more
7 limited body of epidemiologic evidence and air quality information available for PMio-2.5- As
8 discussed in section 4.5 of the Staff Paper (U.S. EPA, 2005), the PMio-2.5 risk assessment
9 included risk estimates for just three urban areas for two categories of health endpoints related to
10 short-term exposure to PMio-2.s: hospital admissions for cardiovascular and respiratory causes
11 and respiratory symptoms.
12 Estimates of hospital admissions attributable to short-term exposure to PMio-2.5 were
13 developed for Detroit, MI (cardiovascular and respiratory admissions) and Seattle, WA
14 (respiratory admissions), and estimates of respiratory symptoms were developed for St. Louis,
15 MO. While one of the goals of the PMio-2.5 risk assessment was to provide estimates of the risk
16 reductions associated with just meeting alternative PMio-2.5 standards, EPA concluded that the
17 nature and magnitude of the uncertainties and concerns associated with this portion of the risk
18 assessment weighed against use of these risk estimates as a basis for recommending specific
19 standard levels (U.S. EPA, 2005, see p. 5-69). These uncertainties and concerns were
20 summarized in the proposal notice (see FR 71 2662, January 17, 2006) and discussed more fully
21 in the Staff Paper (U.S. EPA, 2005, Chapter 4) and associated technical support document (Abt
22 Associates, 2005).
23 1.2 GOALS FOR FRAMING THE ASSESSMENTS IN THE CURRENT
24 REVIEW
25 A critical step in designing the quantitative risk and exposure assessments is to clearly
26 identify the policy-relevant questions to be addressed by these assessments. As identified above,
27 the Integrated Review Plan presents a series of key policy questions (U.S. EPA, 2008a, section
28 3.1). To answer these questions, EPA will integrate information from the ISA and from air
29 quality, risk, and exposure assessments as we evaluate both evidence-based and risk-based
30 considerations.
1-7
-------
1 More specifically, to focus the REA, we have identified the following goals for the risk
2 assessment: (1) to provide estimates of the potential magnitude of premature mortality and/or
3 selected morbidity health effects in the population, including sensitive subpopulations, where
4 data are available to assess these subgroups, associated with recent ambient levels of fine and
5 thoracic coarse particles and with just meeting the current suite of PM standards and any
6 alternative standards that might be considered in selected urban study areas; (2) to develop a
7 better understanding of the influence of various inputs and assumptions on the risk estimates to
8 more clearly differentiate alternative standards that might be considered including potential
9 impacts on various sensitive subpopulations; and (3) to gain insights into the distribution of risks
10 and patterns of risk reduction and uncertainties in those risk estimates. In addition, we are
11 considering conducting an assessment to provide nationwide estimates of the potential
12 magnitude of premature mortality associated with long-term ambient fine particle exposures to
13 more broadly characterize this risk on a national scale and to support the interpretation of the
14 more detailed risk results generated for the selected urban study areas.
15 We have also identified the primary goal for the exposure assessment as providing insight
16 on population exposures to inform the interpretation of available epidemiologic studies. More
17 specifically, the design of the exposure assessment is planned to inform our understanding of
18 how exposure-related factors contribute to the heterogeneity in responses to ambient PM
19 concentrations observed in epidemiologic studies and/or to provide insights on other issues
20 related to uncertainties in the existing epidemiology evidence.
21 1.3 OVERVIEW OF CURRENT ASSESSMENT PLAN
22 This plan outlines the scope and approaches and highlights key issues in our plans to
23 assess human health risks and/or population exposures posed by ambient fine and thoracic coarse
24 particles. The characterizations of the variability and uncertainties associated with the qualitative
25 and quantitative analyses is an important component of our assessments.
26 1.3.1 Air Quality Assessment
27 Chapter 2 describes assessments planned for the current review of the primary NAAQS
28 for PM including air quality analyses to be conducted to support quantitative risk and exposure
-------
1 assessments in selected urban study areas as well as to support evidence-based considerations
2 and to place the results of the quantitative assessments into a broader public health perspective.
3 Air quality inputs will include: (1) recent air quality data for PM2.5 and PMio-2.5 from suitable
4 monitors for each selected urban study area; (2) estimates of policy-relevant background (PRB)5
5 concentrations for each selected urban study area, and (3) simulated air quality that reflects
6 changes in the distribution of PM air quality estimated to occur when an area just meets the
7 current or alternative PM standards under consideration. While incremental risk reductions do
8 not require estimates of PRB, estimates of the risks in excess of PRB remaining upon meeting
9 the current or potential alternative standards, do require us to estimate PRB. Both kinds of risk
10 estimates are considered relevant to inform the Administrator's decision on the adequacy of a
11 given standard. The approach to estimating PRB for PM2.5 and PMio-2.5 for use in conducting the
12 health risk assessment will be informed by the discussion and evaluation contained in the first
13 draft ISA and will build on the approach used in the last review (Langstaff, 2004, 2005). For the
14 exposure assessment, EPA plans to focus on historical air quality data for PM2.5 only considering
15 time periods evaluated in selected epidemiologic studies. The exposure assessment will not
16 evaluate air quality simulated to just meet current or alternative PM2.5 standards.
17 1.3.2 Risk Assessment
18 Chapter 3 discusses the planned health risk assessment, outlined in Figure 1-1. This
19 assessment will build upon the methodology, analyses, and lessons learned from the assessments
20 conducted for the last review as briefly summarized in section 1.1 above. In the Integrated
21 Review Plan, we recognized a potentially broad scope for the quantitative risk assessment and
22 proposed to focus our efforts on fine particles (PM^.s), and to consider, to the extent relevant
23 information is available, risks associated with thoracic coarse particles (PMio-2.s), as well as risks
24 associated with specific PM components, sources, and/or environments (U.S. EPA, 2008a,
25 section 5.5).
5 For the purposes of the risk and exposure assessments, background PM is defined as the distribution of PM
concentrations that would be observed in the U.S. in the absence of anthropogenic (man-made) emissions of primary
PM and precursor emissions (e.g., VOC, NOX, SOX, and NH3) in the U.S., Canada, and Mexico. We refer to
background levels so defined as policy-relevant background (PRB). See section 2.4 for additional information.
1-9
-------
Overview of Risk Assessment Model
Air Quality
•Recent air quality
•Air quality simulated to just
meet current and alternative
NAAQS
•Policy relevant background
Concentration-Response
•C-R functions derived from
epidemiological studies for
various health endpoints
Baseline Incidence and Demographics
•Baseline health effects incidence rates
•Population data
Health
Risk
Model
Risk Estimates
•Recent air quality
•Current or alternative
NAAQS scenarios
3 Figure 1-1 Overview of Risk Assessment Model
4 We plan to focus the risk assessment on selected health effect endpoints (e.g., emergency
5 department visits and hospitalizations for ischemic heart disease) for which the weight of the
6 evidence supports the judgment that the overall health effect category (e.g., cardiovascular
7 morbidity) is at least likely caused by exposure to fine particles (PM^.s) either alone and/or in
8 combination with other pollutants. The planned quantitative risk assessment is designed to
9 estimate risks associated with short- (24-hour average) and long-term (annual average) ambient
10 PM2.5 concentrations in selected urban study areas. We are considering expanding the focus of
11 the risk assessment for fine particles to include additional health effect categories (e.g., birth
12 outcomes) that are within broader health effect categories (e.g., reproductive, developmental,
1-10
-------
1 prenatal and neonatal outcomes) that have been initially classified in the first draft ISA as having
2 suggestive evidence of a causal association with ambient PM2.5 measurements.
3 With respect to evaluating the public health impacts of thoracic coarse particles, we plan
4 to build on the limited risk assessment conducted in the last review for PMi0-2.5 as outlined in
5 section 1.2 above. Based upon the information in the first draft ISA, we plan to focus the risk
6 assessment for thoracic coarse particles on health effect categories that we judge to be
7 sufficiently suggestive of a causal association with short-term (24-hour) PMio-2.5 ambient
8 measurements.
9 We have considered the extent to which evidence supports a quantitative risk assessment
10 for specific PM components, sources, and/or environments. Based upon information assessed in
11 the first draft PM ISA, we have provisionally concluded that the available data do not support a
12 quantitative risk assessment for any specific components of PM2.5 or PM10_2.5, nor for specific
13 sources, environments, or other size fractions (e.g., ultrafine particles).
14 1.3.3 Exposure Assessment
15 As part of the last review, EPA did not conduct an exposure assessment. Chapter 4 discusses our
16 plan to conduct a quantitative exposure assessment in this review primarily to provide insights on
17 population exposures with respect to informing the interpretation of the available epidemiologic
18 evidence. The design of the exposure assessment is summarized in
19 Figure 1-2. The Integrated Review Plan (U.S., EPA, 2008a, section 5.4) outlined a second
20 purpose for conducting a population exposure assessment that is, assessing population exposures
21 above benchmark levels of concern, and providing input to quantitative risk assessments based
22 on evidence from clinical studies. At this time, based upon information presented in the first
23 draft ISA, we are unaware of any results from human clinical studies that would provide the
24 basis for exposure-response functions that could inform the selection of benchmark levels of
25 concern or a quantitative risk assessment; therefore, we do not plan to address the second
26 purpose in this review.
27 The exposure assessment will build upon the methodology, analyses, and lessons learned
28 from assessments conducted for other recent NAAQS reviews (U.S. EPA, 2007, U.S. EPA,
29 2008c). We plan to focus the exposure assessment on PM2 5. We believe, at
1-11
-------
1 this time, that the available monitoring data for PMi0-2.5 do not provide enough spatial coverage
2 for exposure modeling to be useful.
3
4
Air quality and meteorological
measurements; radii of
influence
Overview of Exposure Assessment Model
Demographic data
Diary events/activities
and personal information
Physiological distributions
Microenvironmental data
Population exposure
indicators for
• Total population
• Sensitive populations
Figure 1-2 Overview of Exposure Assessment Model
EPA plans to model population exposures to ambient PM2.5 in a small number of
6 generally representative urban areas across the U.S. Criteria to select the specific urban study
7 areas will include identifying locations where epidemiologic studies have been conducted that
8 are planned to be used to support the quantitative risk assessment. We plan to select urban study
9 areas that will be representative of a variety of populations, geographic areas, climates, and
10 different PM2.5 composition and co-pollutant levels. The exposure modeling locations and
11 periods chosen are planned to be selected from the epidemiologic studies evaluated. We do not
12 plan to develop estimates for population exposures associated with simulation of air quality to
13 just meet the current or any alternative standards under consideration.
1-12
-------
i 2 AIR QUALITY CONSIDERATIONS
2 2.1 INTRODUCTION
3 A number of air quality analyses are planned to provide inputs for the risk and exposure
4 assessments that will be conducted for selected urban study areas as well as to provide a broader
5 understanding of PM air quality, in order to inform: (1) evidence-based considerations; (2) our
6 understanding of the risk and exposure assessment results to better characterize potential
7 nationwide public health impacts associated with exposures to fine and thoracic coarse particles;
8 and (3) policy considerations related to evaluating possible alternative NAAQS. EPA plans to
9 focus air quality assessments on both fine particles (PM^.s) and thoracic coarse particles (PMio-
10 2.5), although the availability of ambient air monitoring data is much more robust for PM2.5 than
11 for PMio-2.5. Specific goals for the planned air quality assessments include:
12 * Characterizing air quality in various locations across the U.S. in terms of PM2.5 and
13 PMio-2.5 considering differences in PM ambient concentrations, composition, and
14 spatial and temporal patterns to help inform the selection of specific cities that we
15 plan to include in the risk and exposure assessments. Analyses for this purpose have
16 been ongoing.
17 • Characterizing policy-relevant background (PRB) based on chemical transport
18 modeling conducted for and described in the first draft ISA (U.S. EPA, 2008b,
19 section 3.6).
20 • Providing air quality distributions for PM2.5 and PMio-2.5 for a number of alternative
21 scenarios in the selected urban study areas including:
22 o Recent air quality;
23 o Simulation of air quality to just meet the current suite of primary PM
24 standards; and
25 o Simulation of air quality to just meet potential alternative primary standards
26 for fine and thoracic coarse particles under consideration.
27 • Providing a broader characterization of current PM2.5, PMio-2.5, and PMio,
28 concentrations nationally (beyond the locations evaluated in the risk and exposure
29 assessments.
2-1
-------
1 2.2 AIR QUALITY INPUTS TO RISK ASSESSMENTS
2 Major inputs to the PM risk assessment are ambient PM air quality data. For the
3 assessments described in this chapter, EPA plans to use 2005-2007 air quality data obtained from
4 EPA's Air Quality System (AQS). These ambient measurement data have been collected by
5 State and local air monitoring agencies that are guided by EPA rules, guidance documents, and
6 grant terms. The period of 2005-2007 has been selected because these data are the most recent
7 data that have been: (1) certified by the State/local monitoring organizations as being complete
8 and accurate to the best of their knowledge and (2) EPA has reached final decisions on which of
9 these data may be excluded under the Exceptional Events Rule (71 FR 13560, March 22, 2007)
10 for 24-hour PM2.5 NAAQS designations.
11 2.2.1 Recent Air Quality: PM2.5
12 For PM2.5, in general, only data collected by Federal reference or equivalent methods
13 (FRMs or FEMs) will be used in the risk and exposure assessments, consistent with the use of
14 such data in most of the health effects studies. However, if an epidemiologic study used non-
15 FRM/FEM data from a continuous PM2.5 monitor(s) as an independent variable for a
16 concentration-response function, consideration will be given to using the same type of data in the
17 quantitative risk assessment for the same location. In order to be consistent with the approach
18 generally used in the epidemiological studies that estimated PM2.5 concentration-response (C-R)
19 functions for short-term effects, we plan to average ambient PM2.5 concentration on each day for
20 which measured data are available for estimating health effects associated with 24-hour ambient
21 concentrations. Consistent with the approach used in the prior two PM2 5 risk assessments, a
22 composite monitor data set will be created for each assessment location based on a composite of
23 all monitors with at least 11 observations per quarter. As in the last review, some monitoring
24 sites may be omitted, if needed, to best match the set of monitors that were used in the
25 epidemiological studies. Most assessment locations will not have a composite estimate for all
26 days because of the variability in local monitoring schedules, in which case adjustments will be
27 made to standardize the risk estimates to 365 days per year.
2-2
-------
1 2.2.2 Recent Air Quality: PMi0.2.5
2 For PMio-2.5, there are air quality analysis challenges not present in the case of PM2.5.
3 PMio-2.5 air quality can be estimated from PM2.5 and PMio concentrations at co-located monitors
4 by subtracting the former from the latter. When inconsistent samplers and filter weighing
5 procedures are used, some of the PMio-2.5 concentrations that are calculated may be negative, and
6 all the calculated values have more uncertainty than when matched, low-volume samplers (which
7 are standard for PM2.5 monitoring) are used. Relatively few sites had collocated matched low-
8 volume samplers operating in 2005-2007, as this became the reference method for PMio-2.5 only
9 in a rulemaking completed late in 2006 (71 FR, 61144, October 17, 2006). Moreover, the
10 historical monitoring network design strategies have been different for PM2.5 and PMio, resulting
11 in many of each type of monitor not having a collocated monitor of the other type. We plan to
12 use the same approach to estimating PMio-2.5 in an assessment location as is used in the
13 epidemiological studies that provide C-R functions. We plan to focus the air quality assessment
14 for thoracic coarse particles on 24-hour ambient PMio and PM2.5 concentrations since that is the
15 focus of any planned PMio-2.5 risk assessment.
16 2.2.3 Air Quality Data Related to Exceptional Events
17
18 State and local agencies and EPA have systematically reviewed PM2.5 data for purposes
19 of making requests and decisions regarding the exclusion of data under the Exceptional Events
20 Rule. We would include these decisions regarding specific data that should be excluded from
21 consideration when testing to see if a monitoring site has air quality meeting the current or
22 alternative PM2.s standards. For thoracic coarse particles, we plan to assume that on any day
23 when the PM2 5 concentration value has been approved for exclusion from use in determining
24 compliance with the 24-hour PM2.5 standard, it is also appropriate to treat the PMio and PMio-2.5
25 concentrations as excludable. In addition, any approved exceptions for PMio data would result in
26 exclusion of PMio-2.5 estimates from those days and from consideration in the rollback process
27 (see section 2.3). PMio data have not been as systematically reviewed by State and local
28 agencies and EPA for purposes of making requests and decisions regarding the exclusion of data,
29 so there is more uncertainty about whether specific data should be excluded from consideration
30 when testing to see if a monitoring site has air quality meeting the current PMio standard or
2-3
-------
1 alternative PMi0 or PMi0-2.5 standards. There may be days for which States have requested
2 exclusion of PMio data and EPA has not made a decision on the request. In these situations, a
3 sensitivity analysis may be conducted, in which such requests are presumed to be approved.
4 2.3 DEVELOPMENT OF ESTIMATES OF PM AIR QUALITY
5 ASSUMING "JUST MEETING" CURRENT NAAQS AND
6 POTENTIAL ALTERNATIVE NAAQS
7 2.3.1 Background and Conceptual Overview
8 In order to simulate air quality concentrations that "just meet" the current or potential
9 alternative PM2.5 standards in a study area, we consider what mathematical approach (commonly
10 referred to as rollback) should be used to transform recent air quality into profiles of adjusted air
11 quality that simulate just meeting the current or alternative standards under consideration. The
12 form of the current PM2.5 standards requires that the 3-year average (rounded to the nearest 0.1
13 Hg/m3) of the annual means from single monitors or the average of multiple monitors must be at
14 or below the level of the annual standard and the 3-year average (rounded to the nearest 1 |ig/m3)
15 of the ninety-eighth percentile values at each monitor cannot exceed the level of the 24-hour
16 standard. In determining attainment of the annual average standard, a State may choose to use
17 either the spatially averaged6 concentrations across all population-oriented monitors, subject to
18 meeting certain criteria detailed in Appendix N of Part 50 of the CFR, or it may use the highest
19 3-year average concentration based on individual monitors. The form of the current 24-hour
20 PMio standard requires that the expected number of exceedances of the level of the standard per
21 year, averaged over 3 years, is not greater than 1.0. The "expected" refers to an adjustment made
22 to the count of actually monitored exceedances to adjust for monitoring schedules which are not
23 every day.
24 The challenge in developing estimates of PM air quality for a scenario in which an
25 assessment location is "just meeting" the current standards or alternative standards under
26 consideration is to estimate as realistically as possible how concentrations on all days at all
27 monitors will be affected, not just how the key air quality statistic from the controlling monitor
6 In the last review of the PM2 5 standards, the criteria for spatial averaging were made more restrictive, and presently
no area follows the spatial averaging approach nor do we think it likely that any areas will do so in the future.
2-4
-------
1 (or set of monitors being averaged) will be affected. The definition of "just meeting" alternative
2 PM standards uses the same approach as "just meeting" the current standards, although some sets
3 of potential alternative standards may include PMi0-2.5 standards instead of the current PMio
4 standard.
5 There are many possible ways to create characterizations of air quality to represent
6 scenarios "just meeting" specified PM2.5 (or PMio-2.s) standards. The previous two reviews have
7 used a method called proportional rollback, which is described below in section 2.3.2. This
8 choice was based on analyses of historical PM2.5 data which found, from comparing the
9 reductions over time in daily ambient PM2 5 levels in two locations with sufficient ambient air
10 quality data, that reductions tended to be roughly proportional (Abt Associates, 2005, Appendix
11 B). We recognize that the pattern of changes that have occurred in the past may not necessarily
12 reflect the temporal and spatial patterns of changes that would likely result from future efforts to
13 attain the PM2.5 standards; therefore, we are considering examining an alternative prospective
14 approach for rollback, as described in section 2.3.3.
15 2.3.2 Historical Approach
16 Prior PM2.5 risk assessments that simulated PM2.5 reductions that would result from just
17 meeting a set of standards used a proportional adjustment ("proportional rollback") which
18 decreased non-background PM levels on all days by the same percentage for all concentrations
19 exceeding the PRB (U.S. EPA, 2005). The portion of the distribution below the
20 estimated background concentration was not rolled back, since air quality strategies adopted to
21 meet the standards would not be expected to reduce the background contribution to PM
22 concentrations. The percentage amount of rollback was just enough so that neither the 24-hour
23 nor the annual levels of the suite of standards under consideration were exceeded. Generally,
24 the amount of rollback required to just meet the 24-hour and the annual levels were not the same,
25 so, in practice, this brought the design value7 for one of the 24-hour or the annual standards at
7 In the risk assessment conducted for the last review, the annual average PM2 5 concentration at each monitor was
calculated for each of the years 2001, 2002, and 2003, and these three annual average concentrations were then
averaged. The maximum of these monitor-specific 3-year averages of annual averages is the annual design value.
At each monitor, the 98th percentile PM2 5 concentration was calculated for each of the years 2001, 2002, and 2003,
and these three 98th (99th) percentile concentrations were then averaged to calculate the 24-hour PM2 5 design value.
2-5
-------
1 the controlling monitor to be equal to the level of the corresponding standard, while the design
2 value of the other standard was reduced to a level below the standard.
3 In the risk assessment for this review, we will again evaluate the proportional rollback
4 approach by comparing it with historical changes in distributions of PM2.5 concentrations in
5 selected locations (and PMi0-2.5 air quality, to the extent possible). Specifically, EPA plans to
6 evaluate historical PM2.5 air quality changes between 1999 and 2007 to assess the implications of
7 using a proportional (PRB-adjusted) rollback approach. This type of analysis is similar to
8 analyses conducted for Los Angeles and Philadelphia in the last risk assessment (Abt Associates,
9 2005, Appendix B). One difference is that in the last review, the composite (multi-monitor) 24-
10 hour PM2.5 concentrations were the subject of the analysis, while for this review, we plan to
11 analyze monitors individually for the purposes of evaluating air quality changes over time.8 We
12 also plan to consider the premises and outcomes of the proportional rollback approach against
13 our insights regarding known and likely future emission reductions, e.g., whether it is reasonable
14 to expect that future patterns of changes in PM air quality would generally be similar to historical
15 patterns of changes in air quality.
16 2.3.3 Alternative Approach Under Consideration
17 In this review, we are also considering an alternative approach to simulating just meeting
18 current or alternative standards under consideration. This alternative approach would take into
19 consideration information about possible future patterns of emissions reductions reflecting both
20 federal regulations that are in place that will affect PM2.5 and precursor emissions as well as
21 possible actions by states to meet current NAAQS. With respect to the historical rollback
22 approach described above, we recognized in the last review that the historical changes in
23 ambient PM2.5 concentrations found to be generally proportional may not have been the result of
24 control strategies designed to meet a PM2.5 NAAQS, but likely resulted from stationary source
25 control programs for PMi0 and other pollutants (especially sulfur and nitrogen oxides) and from
26 multi-pollutant reductions achieved by the national motor vehicle emission control program.
8 This type of analysis has been conducted for the ongoing review of the primary NO2 NAAQS (see Rizzo, 2008 for
a more complete explanation).
2-6
-------
1 Therefore, we recognize that the pattern of changes that had occurred in the past may not
2 necessarily reflect changes that may result from future efforts to attain the PM2.5 standards.
3 More specifically, EPA is investigating the possibility of developing and using a new,
4 alternative approach for simulating PM2.5 air quality "just meeting" the current or alternative
5 PM2 5 NAAQS. This "model-based rollback" approach relies upon results from EPA's
6 Community Multiscale Air Quality (CMAQ) model (Byun et al., 2006 and Byun et al., 1999)
7 that reflect both federal regulations that are in place (or "on the books") as well as possible
8 actions by States to meet the current NAAQS. By using CMAQ modeled response, we can
9 provide information about the non-linear nature of the air quality response to reductions in
10 precursor emissions and direct PM2 5 emissions on PM2 5 concentrations at monitor locations.
11 Since the available modeled response does not reflect meeting the current or potential alternative
12 standards in all study areas, we also need to consider how to make further adjustments to
13 modeled responses that may be needed in some study areas to simulate just meeting some of the
14 standards under consideration. This modeling approach may serve as the basis for a
15 methodology for monitor rollback that we believe has the potential to better reflect expected
16 reductions in PM2 5 concentrations at monitor locations. Therefore, we plan on evaluating the
17 utility of this alternative approach compared to the historical approach based on "proportional
18 rollback" to determine the most appropriate method for adjusting monitors to "just meet" the
19 current and potential alternative PM2.5 NAAQS.
20 2.4 POLICY RELEVANT BACKGROUND
21 For the purposes of the risk assessment, background PM is defined as the distribution of
22 PM concentrations that would be observed in the U.S. in the absence of anthropogenic (man-
23 made) emissions of primary PM and emissions of precursors to secondary PM (e.g.,
24 VOC, NOx, SOx, and NH3) in the U.S., Canada, and Mexico. We refer to background levels so
25 defined as policy-relevant background (PRB), since this definition of background is intended to
26 facilitate separating pollution levels that can be addressed by U.S. regulations (or through
27 international agreements with neighboring countries) from levels that are generally
28 uncontrollable by U.S. regulations.
2-7
-------
1 For this assessment, we are planning to estimate levels of PRB using a CTM-based
2 approach which involves coupling the global-scale circulation model GEOS-Chem (Fiore et al.,
3 2003) with the regional scale air quality model CMAQ (Byun et al., 2006 and Byun et al, 1999).
4 The GEOS-Chem model is run on a global scale and is used to provide estimates of transported
5 pollutants from emissions of natural and anthropogenic sources outside the U.S., Canada, and
6 Mexico. These transported pollutant concentrations are used to provide the boundary condition
7 concentrations for two CMAQ simulations covering the continental U.S. and adjacent portions of
8 Canada and Mexico (CONUS), one simulation of current conditions to evaluate model
9 performance and one to estimate PRB. In the CMAQ simulation to estimate PRB, only natural
10 emissions in the U.S., Canada, and Mexico are considered. The details of this modeling
11 approach (EPA 2008b, section 3.6) are briefly summarized below.
12 The two models have been applied to simulate one year of air quality data for 2004.
13 The base case CMAQ run for 2004 includes meteorology and all the anthropogenic and natural
14 sources both within and outside of the U.S., Canada and Mexico. This run was performed to
15 provide a comparison of model predictions with measurements. The first draft ISA characterizes
16 the CMAQ performance for the annual average concentrations and for most of the seasonal
17 averages of PM2 5 at remote sites as very good in the East and Midwest. In the West, predictions
18 at remote sites are generally too low in all seasons. The first draft ISA further notes that
19 degraded performance in the West is not unexpected because the grid resolution in the CMAQ
20 model simulation (36 km for this application) will smooth out significant variations in terrain
21 that influence measured concentrations, particularly concentrations attributable to anthropogenic
22 emissions which in the West are often concentrated in basin settings where local meteorological
23 conditions coupled with local emissions of primary particles may dominate PM2.5 concentrations.
24 However, looking across the U.S., the model does correctly reproduce broad geospatial
25 differences in that predicted PM2.5 concentrations are lower at western locations than they are in
26 the East, consistent with measured ambient data. Also, natural emissions in the West are less
27 concentrated in basin settings and terrain and, therefore, western terrain may have less effect on
28 model performance when estimating PRB.
2-8
-------
1 In addition to the "base case" run which includes all anthropogenic and biogenic
2 emissions, CMAQ was also run for a second scenario to estimate PRB, with the same boundary
3 conditions but with only natural emissions from within the U.S., Canada, and Mexico. The
4 hourly outputs from this second CMAQ run were used to calculate seasonal and annual average
5 estimates of PRB within seven regions of the U.S. These data are provided in Table 3-26 of the
6 first draft ISA (U.S. EPA, 2008b, page 3-128). We plan to use 24-hour average concentrations
7 of PRB from this CMAQ run (not reported in the first draft ISA) as input to portions of the risk
8 assessment.
9 2.5 BROADER AIR QUALITY CHARACTERIZATION
10 Information presented in the REA will draw upon air quality data analyzed in the ISA as
11 well as national and regional trends in air quality as evaluated in EPA's Air Quality Status and
12 Trends document (U.S., 2008d), and EPA's Report on the Environment (U.S. EPA, 2008e). We
13 plan to use this information, and additional analyses, as needed, to develop a broad
14 characterization of current air quality across the nation. For example, tables of areas and
15 population in the U.S. exceeding current PM2.5 and PMio standards and potential alternative
16 standards would be prepared. Additional information would be generated on the expected
17 number of days on which the 24-hour PM2.5 and PMio standards are exceeded, adjusting for the
18 number of days monitored. Further, daily PM2.5 levels in locations and time periods relevant to
19 areas assessed in key short-term epidemiological studies would be characterized. Information on
20 the spatial, temporal, and compositional characterization of PM2.s across the national monitoring
21 network would be compiled. To the extent possible, we plan to compare these data to the same
22 parameters in the selected urban study areas considered in the quantitative risk assessment to
23 help place the results of that assessment into a broader context.
2-9
-------
i 3 SCOPE AND APPROACH FOR THE HEALTH RISK
2 ASSESSMENT
3 3.1 INTRODUCTION
4 This chapter presents an overview of the design of the human health risk assessment to be
5 conducted in the current review of the PM NAAQS. This design reflects goals laid out in the
6 Integrated Review Plan (U.S. EPA, 2008a, section 5.5) including: (1) to provide estimates of the
7 potential magnitude of premature mortality and/or selected morbidity health effects in the population
8 associated with recent ambient PM levels and with just meeting the current suite of PM standards and
9 any alternative standards that might be considered in selected urban study areas; (2) to develop a better
10 understanding of the influence of various inputs and assumptions on the risk estimates; and (3) to gain
11 insights into the distribution of risks and patterns of risk reduction and uncertainties in those risk
12 estimates. In addition, we are considering conducting an assessment to provide nationwide estimates of
13 the potential magnitude of premature mortality associated with long-term exposure to ambient PM2.5 to
14 more broadly characterize this risk on a national scale and to support the interpretation of the more
15 detailed risk estimates generated for selected urban study areas. As part of the risk assessment, where
16 feasible, quantitative characterizations of the uncertainties associated with the risk estimates will be
17 developed.
18 As outlined in section 1.3.2, the planned health risk assessment will build upon the
19 methodology, analyses, and lessons learned from the assessments conducted for the last review. In the
20 Integrated Review Plan, we recognized a potentially broad scope for the quantitative risk assessment
21 and proposed to focus our efforts on fine particles (PM2.s), and to consider, to the extent relevant
22 information is available, risks associated with thoracic coarse particles (PMio-2.s), as well as risks
23 associated with specific PM components, sources, and/or environments (U.S. EPA, 2008a, section 5.5).
24 With respect to fine particles, based upon the information assessed in the first draft ISA, we
25 plan to focus the risk assessment on health effect endpoints for which the weight of the evidence as
26 assessed in the ISA supports the judgment that the overall health effect category is at least likely
27 caused by exposure to fine particles (PM2.5) either alone and/or in combination with other pollutants.
28 The planned quantitative risk assessment, is designed to estimate risks associated with short- (24-hour
5-1
-------
1 average) and long-term (annual average) ambient PM2 5 concentrations in selected urban study areas.
2 We are considering expanding the focus of this risk assessment to include additional health effect
3 categories beyond those classified as casual or likely causal, when available evidence presented in the
4 ISA is sufficiently suggestive of a causal association to support conducting quantitative risk
5 assessment and when inclusion of that endpoint category will allow us to address potentially important
6 policy issues related to reviewing the PM2.5 NAAQS. For example, we are considering including
7 information on birth outcome effects associated with ambient PM2.5 which would allow us to evaluate
8 additional potentially sensitivity subpopulations (i.e., pregnant women and infants) not previously
9 evaluated in the quantitative risk assessment conducted in the last review. EPA recognizes that a
10 decision to include these additional endpoint categories needs to consider the increased uncertainty that
11 their inclusion could introduce into the risk assessment; specifically, the potential for these endpoints
12 not to be associated with PM2.5 exposure, despite the generation of risk estimates.
13 Building upon the assessment completed in the last review, we plan to focus the PM2.5
14 assessment on modeling risk for a set of selected urban study areas, expanding the number of study
15 areas modeled from nine areas to a somewhat larger set of urban areas (e.g., 15-20 study areas) in order
16 to provide greater population coverage and to better portray the observed heterogeneity in PM2.5-
17 related risk across selected urban study areas. EPA is considering ways to put the quantitative risk
18 assessment results conducted for a limited number of locations and selected health endpoints into a
19 broader context to better characterize the nature, magnitude, extent, variability, and uncertainty of the
20 public health impacts associated with PM2.5 exposures. This includes plans for (1) an exposure
21 assessment (Chapter 4.0); (2) evaluation of the urban study areas with respect to key PM2.5 risk-related
22 parameters to help inform judgments about the representativeness of the urban areas included in the
23 assessment (section 3.4); and (3) consideration of a national-scale health impact assessment (section
24 3.5).
25 With respect to evaluating the public health impacts of thoracic coarse particles (PMio-2.s), we
26 recognize that the first draft ISA presents more limited data for this size fraction. As outlined in
27 section 1.3.2, the planned health risk assessment will build upon the methodology, analyses, and
28 lessons learned from the assessments conducted for the last review. Based on the information assessed
29 in the first draft ISA, no health effect categories have been classified as having a likely causal or causal
5-2
-------
1 association with ambient PMio-2.s. Taking this into account, we plan to focus this assessment on
2 selected health effect endpoints within broader health effect categories for which the evidence, as
3 assessed in the first draft ISA is sufficiently suggestive of a casual association to support conducting a
4 quantitative risk assessment related to short-term (24-hour) ambient PMio-2.5 exposures. The planned
5 approach to model risk for PMi0-2.5 will, in many respects, follow the approach planned for evaluating
6 PM2.s-related risks though with a much more limited set of health endpoints and a smaller number of
7 selected urban study areas. We recognize that the current standard uses a PMio indicator to provide
8 protection against exposures to thoracic coarse particles and that a large number of epidemiological
9 studies use PMio as an air quality metric. However, based upon the information presented in the first
10 draft ISA, it is difficult to evaluate the health effects associated with fine versus thoracic coarse
11 particles in studies that use PMio as a metric.
12 Regarding PM composition, based on information analyzed in the first draft ISA, we do not
13 plan to develop separate risk estimates for PM2.5 or PMi0-2.5 components and we plan to continue to
14 model risk using the basic mass size-fraction approach used in the last review.9 In addition, based on
15 information presented in the first draft ISA, we believe the data are too limited to support conducting a
16 quantitative risk assessment for ultrafine particles.
17 The following discussion begins by presenting the framework for the risk assessment
18 developed to evaluate PM2.5 with more detailed discussions of key components of the risk assessment
19 model including air quality considerations, selection of health effects endpoints to include in the
20 assessment, and specification of concentration-response (C-R) functions (section 3.2). As part of the
21 assessment, where feasible, quantitative characterizations addressing uncertainty and variability
22 associated with the PM2.5 risk estimates will be developed (section 3.3). This section not only presents
23 how we plan to assess and characterize uncertainty potentially impacting the risk assessment but also
24 discusses the degree to which variability related to PM2.5 risk is captured in the analysis design.
25 Section 3.4 discusses the types of risk metrics that may be generated for PM2.5, including how these
26 results could be used to inform consideration of existing and alternative standards. In section 3.5, we
9 See section 2.3.3 of the first draft ISA (U.S. EPA, 2008b) for a discussion of evidence related to PM25
components/sources and associations with specific health effects. We do not plan to incorporate component-specific risk
modeling into our analysis, however, as discussed in Section 3.3.2, the use of location-specific effects estimates potentially
reflects underlying differences in PM2 5 composition as well as additional factors related to PM2 5-related risk including
underlying health status of the study population and exposure-related factors such as prevalence of air conditioning use.
3-3
-------
1 discuss the approach being considered for conducting a national-scale health impact assessment.
2 Lastly, in section 3.6 we present plans to estimate risks associated with thoracic coarse particles (PMio-
3 2.5). Given the similarity of the risk assessment approaches for PM2.5 and PMio-2.5, much of section 3.6
4 references earlier sections covering elements of the PM2.5 risk assessment approach.
5 3.2 FRAMEWORK FOR THE PM2 5 HEALTH RISK ASSESSMENT
6 Overview of Modeling Approach
7 Building on the risk assessment conducted for the last review, the modeling approach we plan
8 to use is based on a risk model whose components are illustrated in Figure 3-1. The calculation of
9 risks is based on Equation 3-1 which combines information about changes in ambient PM2.5 air quality
10 concentrations (Ax) with C-R relationships (reflected by B, the PM2.5 coefficient derived from
11 epidemiological studies) and baseline health incidence data for specific health endpoints (y) to derive
12 estimates of the change in incidence (Ay) of specific health effects attributable to ambient PM
13 concentrations during the period examined.10
14 Equation 3-1 Ay =
15 This type of risk model is based on epidemiological studies characterizing the relationship
16 between ambient PM2 5 levels measured at fixed-site population-oriented monitors and the incidence of
17 specific health endpoints in the population, therefore, it does not require more detailed individual-level
18 exposure modeling and relies instead, on the use of ambient monitoring data. Specifically, a change in
19 the level of ambient PM2.5 is translated through the effect estimate (coefficient B) to a change in the
20 baseline rate of a particular health effect(s) in the study population. This adjustment to the baseline
21 incidence rate can then be combined with population estimates to generate an overall change in the
22 incidence of a specific health endpoint(s) which is attributable to the change in ambient PM2 5
23 concentrations.
24 As illustrated in Figure 3-1, this risk assessment approach requires specifying a number of
25 modeling components related to (a) characterizing air quality, (b) establishing the C-R functions, and
5-4
-------
1 (c) specifying the baseline incidence rates and population demographics. The remainder of this section
2 discusses each of these modeling components in detail.
3 3.2.1 Air Quality Considerations
4 There are several air quality inputs to the risk assessment as illustrated in Figure 3-1. These
5 have been described in Chapter 2 and include: (a) ambient PM2.5 levels for each selected urban study
6 area (including identification of population-oriented monitors and a method for aggregating those
7 monitors and linking them to the study population, in a manner consistent with the approach used in
8 the epidemiological studies underlying the concentration-response functions) for each year of the
9 assessment, (b) quarterly average PRB concentrations for each selected urban study area, and (c)
10 ambient air quality data sets adjusted to simulate air quality conditions that just meet current and
11 alternative PM NAAQS under consideration. Additional detail on these inputs is presented below:
12 • Characterizing recent ambient PMi.s levels for selected urban study areas: EPA plans to
13 use 3 years (2005-2007) of ambient PM2.5 measurement data to characterize recent air quality
14 conditions (see section 2.2). In aggregating monitoring data (to form composite monitor(s) for
15 each study area) and linking those monitors to study populations within a particular study area,
16 we plan to match, to the extent possible, the approach for analyzing air quality data used in the
17 epidemiological studies from which the C-R functions are obtained. For example, in order to
18 be consistent with the approach generally used in the epidemiological studies from which C-R
19 functions have been estimated for effects associated with long-term PM2.5 exposures, we plan
20 to develop and use ambient data for a single composite monitor based on monitored data from
21 all eligible monitors in that study area. Note, that some epidemiological studies have used
22 more sophisticated (and spatially-refined) methods for associating ambient PM2 5 data with a
23 study population (e.g., Jerrett et al., 2005). In cases where we include C-R functions from
24 studies using alternative methods to link ambient PM2 5 concentrations with health effects
25 information in our risk assessment, we may consider a more refined approach for linking PM2.5
26 monitoring data with study populations, to match the approach used in the study.
10 The health risk model given in Equation 3-1 is based on a concentration-response function in which the natural logarithm
of the incidence of the health effect is a linear function of PM2 5 concentration. We plan to consider other mathematical
forms where epidemiological studies have reported effects using other model forms.
5-5
-------
Figure 3-1 Overview of Risk Assessment Model for PM2. (including key model components)
Air Quality
Ambient Monitoring for
Selected Urban Study
Areas
Policy Relevant
Background (PRB)
Air Quality Adjustment
Procedures (Rollback) to
Simulate Just Meeting
Current and Alternative
NAAQS
Recent Ambient
PM2 5 Levels
1
Changes in
Distribution of
PM25 Air Quality
Concentration-Response
Selection of 0 ... ,. ,
._ .. .... Specification of
Epidemiological ^
Selection of health I J Studies to Provide > concentration -
effects endpoints Concentration- Response
... ._ ,. Functions
Response Functions
Baseline Health Effects Incidence Rates and Demographics
Estimates of City-specific
Baseline Health Effects
Incidence Rates
I City-specific Demographic Data r
Health
Risk
Model
Risk Estimates:
Recent Air Quality
- Risk above PRB
Simulating Just Meeting
Current NAAQS
- Risk above PRB
- Incremental difference
in risk compared to
recent air quality
Simulating Just Meeting
Alternative NAAQS
Under Consideration
-Risk above PRB
-Incremental difference
in risk compared to just
meeting current NAAQS
3-6
-------
1 • Characterizing PRB: As noted in section 2.4, we will rely on regionally-differentiated
2 characterization of PRB provided by the assessment summarized in the first draft ISA.
3 • Method for adjusting ambient air quality levels to simulate air quality just meeting
4 current and potential alternative PMi.s NAAQS: As discussed in section 2.3, EPA is
5 considering two methods for simulating PM2.5 levels to just meet current or alternative
6 NAAQS - a proportional rollback approach and a model-based approach.
7 3.2.2 Selection of Health Effects Endpoint Categories
8 As noted in section 3.1, based on review of the first draft ISA, we plan to focus the risk
9 assessment primarily on fine particles, estimating potential health impacts associated with both
10 short-term and long-term exposures to PM2.5. In selecting health effects endpoints to include as
11 an initial matter in the risk assessment, we have considered the following factors based upon
12 review of the first draft ISA (U.S. EPA, 2008b; Chapters 2, 6, and 7): (a) the extent to which the
13 health effect endpoints are considered significant from a public health standpoint, (b) the overall
14 weight of the evidence from the collective body of epidemiological, clinical, and toxicological
15 studies and the inferences made in the first draft ISA as to whether there is a causal or likely
16 causal relationship between PM2.5 and the health effect category, (c) whether there is sufficient
17 evidence to support a causal or likely causal relationship for the specific health endpoint within
18 the health effect category to warrant inclusion in the risk assessment, and (d) whether there are
19 well-conducted studies reporting estimated C-R functions for specific health endpoints within the
20 broader health effect endpoint category associated with ambient PM2.5 levels.
21 Based upon review of the first draft ISA, we plan to consider the following health effect
22 endpoint categories in this assessment:
23 Health Effect Categories Associated with Short-term PMi.s Exposure
24 • cardiovascular morbidity (causal association)
25 • respiratory morbidity (likely causal association)
26 • mortality (likely causal association)
27 Health Effect Categories Associated with Long-term PM2.s Exposure
28 • cardiovascular morbidity (likely casual association)
5-7
-------
1 • respiratory morbidity (likely casual association)
2 • mortality (likely casual association)
3 We are considering expanding the focus of the PM2.s risk assessment to include
4 additional health effect endpoints from health effect categories that have been initially judged in
5 the first draft PM ISA to have a suggestive causal association with ambient PM2.5 measurements.
6 We plan to consider including these additional endpoints when they allow us to address
7 potentially important policy issues related to reviewing the current PM2 5 standards. For
8 example, we are considering including information on birth outcome effects associated with
9 ambient PM2.5 which would allow us to evaluate additional potentially sensitive subpopulations
10 (i.e., pregnant women and infants) not previously evaluated in the quantitative risk assessment
11 conducted in the last review. u EPA recognizes that a decision to include this additional
12 endpoint category would need to appropriately characterize the increased uncertainty associated
13 with these additional outcomes.
14 3.2.3 Specification of Concentration-Response Functions
15 As noted above, the risk assessment conducted in this review will build on the approach
16 developed and applied in the last review. EPA will rely on a weight-of evidence approach, based
17 on the ISA's evaluation of new and previously reviewed epidemiologic studies including
18 identification of relevant C-R functions that characterize the relationships between short- and
19 long-term PM2.5 exposures and health outcomes, particularly those conducted at or near current
20 ambient concentrations. Quantitative relationships provided in the specific studies (or to be
21 derived by EPA from the data presented in the epidemiologic studies) describe the change in
22 concentration (generally based on ambient fixed-site monitors) associated with a change in
23 health response. These C-R relationships will be combined with air quality data, baseline
24 incidence data, and population data to develop population health risk estimates.
1: As noted in the first draft ISA, there are limitations in the evidence assessing the relation between PM2 5 exposure
and reproductive/developmental effects (U.S. EPA, 2008b, page 7-84). Specifically, there are fewer studies with
often inconsistent results in comparison to the evidence available for other endpoint categories (e.g., cardiovascular/
respiratory morbidity and mortality). In addition, characterizing PM2 5 exposure at the etiologically relevant time
period for developmental effects, and understanding the biological mechanisms underpinning these relations remain
important challenges. Despite these limitations, studies have reported associations between ambient PM2 5 exposure
and low birth weight, preterm birth, and respiratory-specific infant mortality, respectively. These results support the
consideration of risk associated with developmental effects in relation to ambient PM2 5 exposure.
-------
1 We plan to use specific criteria to select the epidemiological studies that will be used to
2 provide C-R functions for the quantitative risk assessment including:
3 • The study addresses one of the health effects endpoint categories identified for
4 inclusion in the risk assessment.
5 • The study was peer-reviewed study, evaluated in the first draft ISA, and judged
6 adequate by EPA staff for purposes of inclusion in the risk assessment. Criteria
7 considered by staff include: whether the study provides C-R relationships for
8 locations in the U.S., whether the study has sufficient sample size to provide effect
9 estimates with a sufficient degree of precision and power, whether the study is a
10 multi-city study, and whether adequate information is provided to characterize
11 statistical uncertainty.
12 • The study directly measured PM2.5 (i.e., it did not use a surrogate measure such as
13 airport visibility).
14 • The study is not superseded by another study (e.g., if a later study is an extension or
15 replication of a former study, the later study would effectively replace the former
16 study), unless the earlier study has characteristics that are clearly preferable.
17 In addition to the above criteria, other factors, which may be specific to a particular
18 health effect endpoint, or even to a set of studies, may be considered. For example, several of
19 the studies have improved upon the method of estimating the exposure metric used in most
20 studies which have generally relied upon population-oriented monitoring data. Instead of
21 assigning the same ambient PM2.5 concentration to all individuals in a city (based on a central
22 monitor or the average of several monitors in a city), these studies have assigned "exposures"
23 according to monitors that better approximate conditions near subjects' residences (for example,
24 see Jerrett et al., 2005). In addition, at least one long-term exposure mortality study (Villeneuve
25 et al., 2002) takes into account that exposure changes over time. These and similar studies may
26 provide additional insights into whether reductions in mortality are attributable to recent, or more
27 historical changes in patterns of long-term PM2.5 exposure.
28 We also plan to consider the overall study design, including the method used to adjust for
29 covariates (including confounders and effects modifiers) in identifying candidate studies. For
30 example, if a given study uses ecological-defined variables (e.g., smoking rates) as the basis for
31 controlling for confounding, concerns may be raised as to the effectiveness of that control.
5-9
-------
1 These factors related to confounding control and consideration of effects modification also will
2 be considered in identifying studies for use as the basis of C-R functions.
3 Based on application of these criteria to the set of studies reporting PM2.5 effect estimates
4 evaluated in the first draft ISA, we have identified a provisional set of studies as candidate
5 studies for use in specifying C-R functions relating short-term PM^ambient concentrations with
6 health effects. Table 3-1 includes information on this initial set of studies, as well as the health
7 effect category (and specific endpoints) evaluated in the studies. We plan to continue to refine
8 this list in order to identify a final set of studies for use in deriving C-R functions for inclusion in
9 the risk assessment.
10 With regard to long-term PM^exposure studies, application of the criteria listed above
11 has resulted in a preliminary set of studies listed in Table 3-2 . We plan to continue to refine this
12 list prior to selecting specific C-R functions for use in the quantitative risk assessment evaluating
13 PM2.5-related health effects associated with long-term ambient exposures (annual average
14 concentrations).
15 Once the final set of epidemiological studies is chosen, the next step will be the selection
16 of C-R functions from those studies. A number of factors need to be considered in specifying C-
17 R functions related to short- and long-term exposure studies. The factors being considered in
18 selecting C-R functions include:
19 • Single- and multi-pollutant models (pertains to both short-term and long-term
20 exposure studies): Epidemiological studies often consider health effects associated with
21 ambient PM2.5 independently as well as together with co-pollutants (e.g., ozone, nitrogen
22 dioxide, sulfur dioxide, carbon monoxide). To the extent that any of the co-pollutants
23 present in the ambient air may have contributed to health effects attributed to PM2 5 in
24 single pollutant models, risks attributed to PM2 5 may be overestimated if C-R functions
25 are based on single pollutant models. This would argue for inclusion of models reflecting
26 consideration of co-pollutants. Conversely, in those instances where co-pollutants are
27 highly correlated with PM2 5, inclusion of those pollutants in the health impact model can
28 produce unstable and statistically insignificant effect estimates for both PM2 5 and the co-
29 pollutants. This situation would argue for inclusion of a model based exclusively on
30 PM2.5. Given that single and multi-pollutant models each have potential advantages and
31 disadvantages, we plan to include both types of C-R functions in the risk assessment.
3-10
-------
Table 3-1. Provisional Summary of Short-Term Epidemiological Studies Being Considered as Basis for Selecting C-R Functions for the PM2.5 Risk Assessment
Location
All cause mortality
Cardiovascular
mortality
Respiratory
mortality
Respiratory hospital
admissions
Cardiovascular
hospital admissions
ER visits
Respiratory
Symptoms (Not
requiring
hospitalization)
Region: West/Southwest (Minus CA)
Anchorage, AK
El Paso, TX
Las Vegas, NV
Phoenix, AZ
Seattle, WA
Spokane, WA
Franklin et al.(2008)
Franklin et al.(2007)**
Franklin et al.(2007)
Franklin et al.(2008)**
Franklin et al.(2007)
Slaughter et al.(2005)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et a 1. (2007)
Wilson et al. (2007)
Mar (2003) [reanalysis of
Mar (2000)]
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Chimonas et al. (2007)
Dominici et al. (2006)***
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)****
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Sheppard (2003) [reanalysis
ofSheppardetal. (1999)]-
asthma
Dominici et al. (2006)
Bell et al. (2008)
Slaughter et al. (2005)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Slaughter et al. (2005)
Slaughter et al.
(2005) (respiratory)
Slaughter et al. 2003
Mar etal. 2004
Region: California
Bakersfield/ Kern Co.,
CA
Contra Costa, CA
Fresno, CA
Los Angeles, CA
Orange, CA
Riverside, CA
Franklin et al.(2008)
Ostro et al. (2007)
Franklin et al.(2008)
Franklin et al.(2007)
Ostro et al. (2007)
Franklin et al.(2008)
Franklin et al.(2007)
Moolgavkar (2003)
[reanalysis of
Moolgavkar (2000a)]
Jerrett et al. (2005)
Franklin et al.(2008)
Ostro et al. (2007)
Franklin et al.(2007)
Franklin et al.(2008)
Ostro et al. (2006)
Ostro et al. (2007)
Franklin et al.(2008)
Franklin et al.(2007)
Ostro et al. (2006)
Ostro et al. (2007)
Franklin et al.(2008)
Franklin et al.(2007)
Moolgavkar (2003)
[reanalysis of
Moolgavkar (2000a)]
Jerrett et al. (2005)
Ostro et al. (2007)
Ostro et al. (2007)
Franklin et al.(2008)
Ostro et al. (2006)
Ostro et al. (2007)
Franklin et al.(2007)
Franklin et al.(2008)
Ostro et al. (2006)
Franklin et al.(2008)
Franklin et al.(2007)
Ostro et al. (2006)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Ostro et al. (2006)
Franklin et al.(2007)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Moolgavkar 2003
(Reanalysis of Moolgavkar,
2000)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
3-11
-------
Table 3-1. Provisional Summary of Short-Term Epidemiological Studies Being Considered as Basis for Selecting C-R Functions for the PM2.5 Risk Assessment
Location
Sacramento, CA
San Diego, CA
San Jose, CA
Santa Clara Co., CA
All cause mortality
Franklin et al.(2008)
Ostro et al. (2007)
Franklin et al.(2007)
Franklin et al.(2008)
Ostro et al. (2007)
Franklin et al.(2007)
Fairley (2003)
[reanalysis of Fairley
(1999)]
Ostro et al. (2006)
Ostro et al. (2007)
Cardiovascular
mortality
Franklin et al.(2008)
Ostro et al. (2006)
Ostro et al. (2007)
Franklin et al.(2007)
Franklin et al.(2008)
Ostro et al. (2006)
Ostro et al. (2007)
Franklin et al.(2007)
Fairley (2003)
[reanalysis of Fairley
(1 999)]
Ostro et al. (2007)
Respiratory
mortality
Franklin et al.(2008)
Ostro et al. (2006)
Franklin et al.(2007)
Franklin et al.(2008)
Ostro et al. (2006)
Franklin et al.(2007)
Fairley (2003)
[reanalysis of Fairley
(1999)]
Ostro et al. (2006)
Respiratory hospital
admissions
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Cardiovascular
hospital admissions
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
ER visits
Respiratory
Symptoms (Not
requiring
hospitalization)
Region: South
Atlanta, GA
Beaumont, TX
Birmingham, AL
Dallas, TX
Durham, Guilford, and
Wake Counties, NC
Houston, TX
Kingston/Harriman, TN
Memphis, TN
Palm Beach, FL
Tampa, FL
Klemm et al. (2004)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Schwartz (2003b)
[reanalysis of Schwartz
etal. (1996)]
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Holloman et al. (2004)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Klemm and Mason
(2003) [reanalysis of
Klemm et al. (2000)]
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2007)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Metzgeretal. (2004)
(cardiovascular);
Peel et al. (2005)
(respiratory)
Tolbert et al. (2007)
(cardiovascular)
Schwartz and Neas,
2000
3-12
-------
Table 3-1. Provisional Summary of Short-Term Epidemiological Studies Being Considered as Basis for Selecting C-R Functions for the PM2.5 Risk Assessment
Location
All cause mortality
Cardiovascular
mortality
Respiratory
mortality
Respiratory hospital
admissions
Cardiovascular
hospital admissions
ER visits
Respiratory
Symptoms (Not
requiring
hospitalization)
Region: Midwest/Central
Akron, OH
Chicago, IL
Cincinnati, OH
Cleveland, OH
Columbus, OH
Dayton, OH
Detroit, Ml
Erie, PA
Harrisburg, PA
Indianapolis, IN
Kansas City, MO
Milwaukee, Wl
Minneapolis, MN
Pittsburgh, PA
Portage, Wl
St. Louis, MO
Steubenville, OH
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Ito (2003) [reanalysis of
Lippmann et al. (2000)]
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Chock et al. (2000)
Schwartz (2003b)
[reanalysis of Schwartz
etal. (1996)]
Schwartz (2003b)
[reanalysis of Schwartz
etal. (1996)]
Franklin et al.(2008)
Schwartz (2003b)
[reanalysis of Schwartz
etal. (1996)]
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Ito (2003) [reanalysis of
Lippmann et al. (2000)]
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Ito (2003) [reanalysis of
Lippmann et al. (2000)]
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Klemm and Mason
(2003) [reanalysis of
Klemm et al. (2000)]
Klemm and Mason
(2003) [reanalysis of
Klemm et al. (2000)]
Franklin et al.(2008)
Klemm and Mason
(2003) [reanalysis of
Klemm et al. (2000)]
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Ito (2003) [Reanalysis of
Lippmann et al. 2000]
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Schwartz and Neas,
2000
Schwartz and Neas,
2000
Schwartz and Neas,
2000
3-13
-------
Table 3-1. Provisional Summary of Short-Term Epidemiological Studies Being Considered as Basis for Selecting C-R Functions for the PM2.5 Risk Assessment
Location
Toledo, OH
Topeka, KA
All cause mortality
Franklin et al.(2008)
Schwartz (2003b)
[reanalysis of Schwartz
etal. (1996)]
Cardiovascular
mortality
Franklin et al.(2008)
Respiratory
mortality
Franklin et al.(2008)
Klemm and Mason
(2003) [reanalysis of
Klemm et al. (2000)]
Respiratory hospital
admissions
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Cardiovascular
hospital admissions
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
ER visits
Respiratory
Symptoms (Not
requiring
hospitalization)
Schwartz and Neas,
2000
Region: Northeast
Baltimore, MD
Boston, MA
New York, NY
Philadelphia, PA
Washington, DC
Schwartz (2003b)
[reanalysis of Schwartz
etal. (1996)]
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2008)
Lipfert et al. (2000)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2008)
Lipfert et al. (2000)
Franklin et al.(2007)
Franklin et al.(2007)
Klemm and Mason
(2003) [reanalysis of
Klemm et al. (2000)]
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2007)
Franklin et al.(2008)
Franklin et al.(2007)
Franklin et al.(2007)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Bell et al. 2008
Dominici et al. (2006)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Babin et al (2007)
Dominici et al. (2006)
Bell et al. (2008)
Symons et al. (2006)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Dominici et al. (2006)
Bell et al. (2008)
Ito et al. (2007)
(respiratory)
Babin et al (2007)
(respiratory)
Schwartz and Neas,
2000
"Studies were included in this table regardless of whether their results were statistically significant, as long as they reported results for one of the health endpoint categories in the table.
"Location-specific results are shown in these papers only in a figure (but may be available from the study authors).
***Dominici et al. (2006) included the 204 U.S. counties with populations > 200,000, but reported results by region of the country. The individual counties included in this study are listed at:
http://www.biostat.jhsph.edu/MCAPS/county-info.html.
****Bell et al. (2008) used the same set of counties as Dominici et al. (2006) (minus 4 counties); results are reported by region of the country; if there are county-specific results for the county of the location,
3-14
-------
Table 3-2. Provisional Summary of Long-Term Epidemiological Studies Being Considered
for Selecting C-R Functions for the PMi.s Risk Assessment
Study
Dockery et al., 1993
(Six Cities Study)
Krewski et al.
(2000) - Reanalysis
of Six Cities Study
Villeneuve et al.
(2002)
Laden et al. (2006)
Eftim et al. (2008)
Popeetal. (1995)-
American Cancer
Society (ACS)
Study
Krewski et al.
(2000) - Reanalysis
of ACS Study
Pope et al. (2002) -
ACS extended
Pope et al. (2004) --
ACS study
Jerrett et al. (2005)
Enstrom (2005)
McDonnell et al.
(2000) (AHSMOG
study)
Health effect endpoint
categories
(mortality categories, unless
otherwise noted)
• all-cause,
• lung cancer,
• cardiopulmonary
• all-cause,
• lung cancer,
• cardiopulmonary
• all-cause
• all-cause,
• lung cancer,
• cardiovascular
• all-cause
• all-cause,
• lung cancer,
• cardiopulmonary
• all-cause,
• lung cancer,
• cardiopulmonary
• all-cause,
• lung cancer,
• cardiopulmonary
• cardiovascular
• all-cause,
• lung cancer,
• cardiopulmonary
• all-cause,
• cardiopulmonary
• all-cause,
• lung cancer
• nonmalignant respiratory
diseases*
Study locations
6 U.S. cities - Watertown, MA;
Kingston/Harriman, TN; St. Louis,
MO; Steubenville, OH; Portage, WI;
Topeka, KS
6 Harvard Six Cities locations and
110 counties (within the 50 ACS
metropolitan areas)
ACS locations (50 metropolitan
areas)
ACS locations (50 metropolitan
areas)
ACS metropolitan areas:
1979-1983: 61; 1999-2000: 116;
average: 51.
ACS metropolitan areas:
1979-1983: 61; 1999-2000: 116;
average: 51.
Los Angeles
1 1 CA counties
1 1 airsheds in CA
3-15
-------
Study
Health effect endpoint
categories
(mortality categories, unless
otherwise noted)
Study locations
Chen et al. (2005)
(AHSMOG study)
• ischemic heart disease/coronary
heart disease
San Francisco, South Coast, and San
Diego air basins in CA, plus a
random sample from the rest of CA
Goss et al. (2004)
Mortality:
• all-cause
Morbidity:
Cystic Fibrosis pulmonary
exacerbations
Locations around the U.S.**
Lipfert et al. (2006a
and b)
• all-cause
Cohort in the following states: CT,
DC, IL, IN, MD, MI, NJ, NY, OH,
PA, AL, FL, TN, VA, AR, IA, KS,
LA, MN, MS, WI, AZ, CA, OK, TX,
UT, WA
Chen et al. (2005)
(AHSMOG study)
• ischemic heart disease/coronary
heart disease
San Francisco, South Coast, and San
Diego air basins in CA, plus a
random sample from the rest of C A
Miller at el. (2007)
Mortality:
• cardiovascular
Morbidity:
First event of any of the
following:
• Any cardiovascular event,
• Coronary heart disease
• Cerebrovascular disease
•MI
• Coronary revascularization
Stroke
36 US metropolitan areas
* As either the underlying or a contributing cause of death.
** Subjects were taken from patients enrolled in the Cystic Fibrosis Foundation National Patient
Registry in 1999 and 2000.
3-16
-------
1 • Single- versus multi-city studies (typically a factor in short-term exposure studies). All
2 else being equal, we judge C-R functions estimated in the assessment location as
3 preferable to a function estimated in some other location, to avoid uncertainties that may
4 exist due to differences associated with geographic location. There are several
5 advantages, however, to using estimates from multi-city studies versus studies carried out
6 in single cities. Multi-city studies are applicable to a variety of settings, since they
7 estimate a central tendency across multiple locations. Multi-city studies also tend to have
8 more statistical power and provide effect estimates with relatively greater precision than
9 single-city studies due to larger sample sizes, reducing the uncertainty around the
10 estimated health coefficient. However, multi-city studies may also mask potential
11 differences in PM2.5 composition and the impact that compositional differences may have
12 on the magnitude of effects estimates. By contrast, single-city studies, while often having
13 lower statistical power and varying study designs which can make comparison across
14 cities challenging, do reflect location-specific factors such as PM2 5 compositional
15 differences, differences in underlying health status, and differences in exposure-related
16 factors such as air conditioner use and urban density with larger populations exposed near
17 high-traffic roads. Because single- and multi-city studies have different advantages, we
18 plan to include both types of functions in this analysis, where they are available. We plan
19 to place greater weight on the use of C-R relationships reflecting adjusted single-city
20 estimates from multi-city studies. This would include Empirical Bayes adjusted city-
21 specific estimates. These types of effect estimates benefit both from increased statistical
22 power, as well as the potential for specification of city-specific effect estimates.
23 Conversely, if a multi-city study only provides aggregated effect estimates, but does
24 differentiate those estimates regionally, we plan to use those regional-specific estimates
25 rather than a single national-level estimate by matching selected urban study areas to
26 these regions.
27 • Multiple lag models (pertinent to short-term exposure time-series studies). If
28 information is available for a distributed lag model, we plan to use that model. Where
29 there are multiple lags presented, but a distributed lag model is not included, we plan to
30 consider information presented in the first draft ISA to determine if there is biological
31 support for selecting a specific lag period for a given health effect endpoint.
32 • Seasonally-differentiated effects estimates (pertinent to short-term studies). In those
33 instances where studies presented effect estimates associated with short-term ambient
34 PM2 5 concentrations differentiated by season, we plan to use these seasonal estimates.
35 We plan to link seasonal effect estimates with seasonal PM2 5 air quality data in
36 conducting the risk assessment for selected urban study areas.
37 • Shape of the functional form of the risk model: In the risk assessment conducted in
38 the last review, EPA included C-R relationships that reflected linear or log-linear C-R
39 functions that extended down to estimated PRB levels for effects related to short-term
40 exposure and down to lowest measured ambient levels for effects related to long-term
41 exposure, as well as adjusting these models to reflect various alternative "cutpoint"
42 models. The alternative cutpoint models imposed an assumed threshold on the original
43 C-R function, below which there is little or no population response. The first draft ISA
3-17
-------
1 concludes that there is little support in the literature for a population threshold for
2 mortality effects associated with long-term or short-term PM2.5 ambient concentrations,
3 although it also states that alternate model forms (including presence of a threshold -
4 inference added) can not be ruled out on a city-specific basis (see U.S. EPA, 2008b, pp.
5 2-21 to 2-22). For this reason, while we plan to emphasize non-threshold C-R functions
6 in the risk assessment model, we may also consider various cutpoints, or hypothetical
7 population thresholds, as part of the uncertainty analysis.
8 In addition to the factors listed above, there are additional factors related to the design of
9 individual epidemiological studies which we plan to consider in selecting the C-R functions to be
10 included in the assessment. For example, studies often include adjustment for covariates with
11 varying degrees of freedom, reflecting the tradeoff between bias and over-adjustment (loss of
12 efficiency). In these cases, we plan to consider any information provided for specific studies
13 within the first draft ISA and also plan to consider which model form has the strongest statistical
14 fit, while still considering overall biological plausibility.
15 3.2.4 Selection of Urban Study Areas
16 We plan to build on the risk assessment conducted for the last review and continue to
17 focus the risk assessment on a set of selected urban study areas. In the last review, nine urban
18 areas were evaluated. In this review, we plan to focus on 15 to 20 urban areas. The decision to
19 continue to focus on modeling a set of selected urban study areas reflects the goal of providing
20 risk estimates that have higher overall confidence due to the use of location-specific data when
21 available for these urban locations. In addition, given the greater availability of location-specific
22 data a more rigorous evaluation of the impact of uncertainty and variability can be conducted for
23 a set of selected urban study areas, than would be possible for a broader regional or national-
24 scale analysis. We plan to consider the following factors in the selection of urban study areas:
25 • Air quality data: The urban area has sufficient recent (2005-2007) air quality data to
26 conduct the risk assessment (See section 2.2.1).
27 • Location-specific C-R functions: There are C-R functions available from
28 epidemiological studies identified in section 3.2.3, for one or more of the selected health
29 endpoints. This primarily applies to short-term epidemiological studies, which more
30 often include city-specific effect estimates (see Table 3-1). C-R functions available from
31 long-term epidemiological studies generally combine data from multiple cities. Specific
32 cities evaluated in the key long-term studies would be considered for inclusion in the risk
33 assessment (see Table 3-2). We plan to include urban study areas that have been
3-18
-------
1 assessed in epidemiological studies that have evaluated health effects associated with
2 both short- and long-term PM2.5 exposures and, to the extent possible, locations where
3 both morbidity and mortality health endpoints have been evaluated.
4 • Baseline incidence rates and demographic data: The required urban area-specific
5 baseline incidence rates and population data are available for a recent year for at least one
6 of the health endpoints.
7 • Geographic heterogeneity: Because PM2.5 composition and populations vary
8 geographically across the U.S., we plan to select a set of urban study areas in which each
9 region of the country is represented. We plan to define these regions in such a way as to
10 reflect differences in factors related to PM2.5 composition, sources, co-pollutants,
11 exposure, and/or effect estimates.
12 • Representing areas with relatively larger vulnerable populations: Baseline incidence
13 rates (e.g., mortality rates) and PM exposures are higher in some parts of the country than
14 others. We plan to select a set of urban study areas that will include representation of
15 sensitive subpopulations (e.g., those with higher baseline incidence rates of the health
16 effect endpoints being evaluated, lower air conditioning usage which has been related to
17 higher ambient PM exposures).
18 • Consideration of epidemiology studies with more refined exposure metrics: We plan
19 to include urban study areas for which there is a C-R function estimated using a more
20 refined metric of exposure (e.g., smaller geographic units linked to nearest PM monitors,
21 rather than constructing a single composite monitor for an entire metropolitan area),
22 where available.
23 3.2.5 Baseline Health Effects Incidence Data and Demographic Data
24 As noted earlier (section 3.2.1), the most common epidemiological-based health risk
25 model expresses the reduction in health risk (Ay) associated with a given reduction in PM2.5
26 concentrations (Ax) as a percentage of the baseline incidence (y). To accurately assess the
27 impact of PM2.s air quality on health risk in the selected urban areas, information on the baseline
28 incidence of health effects (i.e., the incidence under recent air quality conditions) in each
29 location is needed. Where at all possible, we plan to use county-specific incidences or incidence
30 rates (in combination with county-specific populations). A summary of available baseline
31 incidence data for specific categories of effects is presented below:
32 • Availability of baseline incidence data on mortality: County-specific (and, if desired,
33 age- and race-specific) baseline incidence data are available for all-cause and cause-
3-19
-------
1 specific mortality from CDC Wonder.12 The most recent year for which data are
2 available online is 2005.13
3 • Availability of baseline incidence data for hospital admissions and emergency room
4 (ER) visits:
5 o Cause-specific hospital admissions baseline incidence data are available for each
6 of 40 states from the State Inpatient Databases (SID).
7 o Cause-specific ER visit baseline incidence data are available for 26 states from
8 the State Emergency Department Databases (SEDD).
9 o SID and SEDD are both developed through the Healthcare Cost and Utilization
10 Project (HCUP), sponsored by the Agency for Healthcare Research and Quality
11 (AHRQ).
12 o The data generated from HCUPnet (HCUP's online interactive tool) are state-
13 level summary statistics, whereas the data from the HCUP distributor are at the
14 individual discharge level.
15 o In addition to being able to estimate State-level rates, SID and SEDD can also be
16 used to obtain county-level hospital admission and ER visit counts by aggregating
17 the discharge records by county.
18 EPA is in the process of obtaining the county-specific hospital admission and ER visit
19 baseline incidence data for the most recent single year available for most of the States included
20 in the HCUP data. While we recognize that there is year-to-year variability in baseline incidence
21 data, a single year of data is being obtained due to resource constraints. We plan to examine the
22 potential variability in baseline incidence data and the impact this might have on the risk
23 estimates in sensitivity analyses based on endpoints and locations where we can obtain multi-
24 year baseline incidence data at little or no cost and by examining the variability in baseline
25 incidence rates at the State level.
12 http://wonder.cdc.gov/mortsql.html
13 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.
3-20
-------
1 3.3 CHARACTERIZATION OF UNCERTAINTY AND VARIABILITY IN
2 THE CONTEXT OF THE PM2 5 RISK ASSESSMENT
3 3.3.1 Differentiating Between Uncertainty and Variability
4 An important issue associated with any population health risk assessment is the
5 characterization of uncertainty and variability. Variability refers to the heterogeneity in a
6 population or variable of interest that is inherent and cannot be reduced through further research.
7 For example, there may be variability among C-R functions describing the relation between
8 PM2.5 and mortality across selected urban study areas. This variability may be due to differences
9 in population (e.g., age distribution), population activities that affect exposure to PM2.5 (e.g., air
10 conditioning use), levels and composition of PM2.5 and/or co-pollutants, and/or other factors that
11 vary either within or across urban areas.
12 Uncertainty refers to the lack of knowledge regarding both the actual values of model
13 input variables (parameter uncertainty) and the physical systems or relationships (model
14 uncertainty - e.g., the shapes of concentration-response functions). In any risk assessment,
15 uncertainty is, ideally, reduced to the maximum extent possible, through improved measurement
16 of key parameters and ongoing model refinement. However, significant uncertainty often
17 remains and emphasis is then placed on characterizing the nature of that uncertainty and its
18 impact on risk estimates. The characterization of uncertainty can include both qualitative and
19 quantitative analyses, the latter requiring more detailed information and often, the application of
20 sophisticated analytical techniques such as 2-stage Monte Carlo simulation.
21 While the goal in designing a quantitative risk assessment is to reduce uncertainty to the
22 extent possible; with variability, the goal is to incorporate the sources of variability into the
23 analysis approach to insure that the risk estimates are representative of the actual response of a
24 study population (including the distribution of that adverse response across the study
25 population). An additional aspect of variability which is pertinent to this risk assessment is the
26 degree to which the set of selected urban study areas provide coverage for the range of PM2.5-
27 related risk experienced by the U. S. population.
3-21
-------
1 We plan to more fully differentiate variability and uncertainty in the design of the risk
2 assessment to more clearly address (a) the extent to which the risk estimates represent the
3 distribution of health impacts across a population, including impacts on more susceptible and/or
4 vulnerable subpopulations14 and (b) the extent to which risk estimates are impacted by key
5 sources of uncertainty which could prevent a clear differentiation between regulatory alternatives
6 based on risk estimates.
7 The remainder of this section discusses how we are planning to address variability and
8 uncertainty within the PM NAAQS risk assessment. The treatment of variability is discussed first
9 (section 3.3.2) by identifying sources of variability associated with the modeling of PM25-related
10 risk and noting which of those sources are reflected in the risk modeling approach presented
11 here. Next, the treatment of uncertainty is addressed, which will include both a qualitative and
12 quantitative component. The qualitative component is described first (section 3.3.3), including
13 plans for identifying and describing key sources of uncertainty, and noting whether those sources
14 of uncertainty are addressed quantitatively in the risk assessment model. A preliminary list of
15 key sources of uncertainty for the risk assessment is provided as part of this discussion. The
16 quantitative component of the uncertainty characterization approach, which is structured around
17 single-factor and multi-factor sensitivity analysis methods, is then described (section 3.3.4).
18 3.3.2 Addressing Variability
19 Key sources of variability associated with the modeling of population-level risk
20 associated with PM2.5 exposure are presented below, including whether, and to what extent, we
21 plan to address each source of variability:
22 , PM2.s composition: We plan to address differences in PM2.5 composition by using city-
23 specific effects estimates, which can reflect, among other location-specific factors related
24 to PM2.5 exposure and risk, differences in PM2.5 composition. We do not plan to
25 explicitly consider PM2.5 composition within the risk assessment because C-R functions
26 for specific PM2.5 components/sources have not been identified.
14 Susceptibility refers to innate (e.g., genetic or developmental) or acquired (e.g., age, disease, or smoking) factors
that make individuals more likely to experience effects with exposure to PM. Vulnerability refers to PM-related
effects due to factors including socioeconomic status (e.g., reduced access to health care) or particularly elevated
exposure levels.
3-22
-------
1 • Spatial gradients in PMi.s (and related population exposure): This source of
2 variability is likely to be less-well captured in the risk assessment primarily because the
3 majority of epidemiological studies providing effect estimates are themselves limited in
4 reflecting more detailed patterns of PM2 5 exposure among populations. More
5 specifically, the epidemiological studies typically use an average ambient concentration
6 developed across population-oriented monitors as a surrogate for exposure. Note,
7 however that the exposure assessment described in Chapter 4 may allow this issue to be
8 investigated to some degree, particularly as it impacts on exposure error misclassification
9 in the epidemiological studies underpinning the C-R functions used in this risk
10 assessment. In addition, a few epidemiological studies being considered for inclusion in
11 this analysis include more refined characterization of population-level exposure (e.g.,
12 based on more spatially differentiated linkages between population-level monitors and
13 segments of the study population). We plan to consider the use of those studies with more
14 refined population exposure characterization to examine the issue of spatial gradients in
15 PM2.5 and demographics and the degree to which this source of variability impacts risk
16 estimates.
17 • Demographics (i.e., greater concentrations of susceptible subpopulations in certain
18 locations): We plan to include multiple urban study areas reflecting differences in
19 demographics in different regions of the country to address this issue. In addition, as
20 noted in the previous bullet, we plan to consider studies with more refined
21 characterization of population-level exposure, to provide insights into the degree to which
22 this source of variability impacts risk estimates.
23 • Behavior related to PMi.s exposure (e.g., outdoor time, air conditioning use): We
24 plan to include multiple urban study areas reflecting differences in a variety of factors
25 related to PM2.5 exposure (e.g., time spent outdoors, air conditioner use, housing stock
26 which can impact PM2.5 infiltration, and commuting patterns).
27 • Susceptibility to specific populations to PMi.s exposure (note - this could include a
28 number of factors e.g., magnitude of the effect estimate, underlying health status): We
29 plan to consider this source of variability by using effect estimates and lag structures
30 specific to each urban study location.
31 • Differences in baseline incidence of disease: This source of variability would
32 potentially be captured through the use of localized baseline incidence data (e.g., county -
33 level).
34 • Longer-term temporal variability in ambient PM2.s levels (reflecting meteorological
35 trends, as well as future changes in the mix of PM2.5 sources and regulations impacting
36 PM2 5): This is more difficult to incorporate into the analysis and reflects a combination
37 of variability as well as uncertainty.
3-23
-------
1 3.3.3 Uncertainty Characterization - Qualitative Assessment
2 We plan to include a qualitative discussion of uncertainty in the risk assessment which
3 will include: (1) identification and description of key sources of uncertainty, noting whether they
4 are addressed quantitatively in the risk assessment model and (2) a qualitative assessment of
5 those sources of uncertainty in terms of their potential impact on risk using a "high," "medium,"
6 and "low" designation. A preliminary list of potentially important sources of uncertainty has
7 been developed for this plan and is presented below (note, some of these sources may be
8 addressed in the quantitative uncertainty analysis, when feasible):
9 • Statistical uncertainty associated with the fit of the C-R function
10 • Shape of the C-R function: (e.g., shape of the function, including the potential for a
1 1 population threshold). Of particular concern is uncertainty related to the shape of the C-R
12 function at lower exposure levels where there is less exposure and response data.
13 • Potential role of co-pollutants and different lag structures: these are related to the C-
14 R function (and nature of the associated effects estimate).
15 • Transferability of C-R functions from study locations to urban study area locations:
16 this reflects variation in (a) PM2.5 composition, (b) the possible role of copollutants in
17 influencing risk, (c) relationship between ambient PM2.5 and actual exposure, and (d)
18 differences in population characteristics. However, it is anticipated that the transferability
19 issue will play less of a role in the upcoming analysis, since studies used to derive C-R
20 functions will often be matched to our urban study area locations. However, there may
21 still be transferability issues arising from changes in these factors between the time
22 period when the C-R functions were estimated and the time period of this risk analysis.
23 • Procedures for adjusting air quality to simulate alternate standard levels: There is
24 uncertainty in developing the method for adjusting current ambient PM2.5 levels (at
25 individual monitors used in the risk assessment) to simulate just attaining alternative
26 standard (methods available are likely to include both retrospective empirical monitor-
27 based trend analysis and forward-looking model -based predictions - see section 3.2. 1 and
28 section 2.3 for additional detail).
29 • The impact of historical air quality on estimates of health risk from long-term
30 exposures (i.e., the amount of time that a population experiences new lower ambient
3 1 PM2.5 levels before there is a noticeable reduction in health effect incidence): Some
32 studies of long-term mortality provide effect estimates differentiated by consecutive,
33 multi-year time periods (e.g., Pope et al., 2002). These studies may provide insights into
34 this issue and the degree to which it could impact risk estimates (by providing different
35 effect estimates).
3-24
-------
1 • Estimates of policy-relevant background PMi.s levels in a particular location. There
2 is uncertainty associated with characterizing PRB for individual locations (see Section
3 3.2.1 for additional detail).
4 3.3.4 Uncertainty Characterization - Quantitative Analysis
5 The quantitative uncertainty analysis planed for the PM2.5 risk assessment will utilize a
6 deterministic sensitivity analysis-based approach, designed to provide the decision maker with a
7 reasonable alternative set of risk estimates to supplement the set of core risk estimates that are
8 generated. This set of additional risk estimates will provide insights into the impact of
9 uncertainty on the initial set of core risk estimates (the deterministic uncertainty approach is
10 illustrated in Figure 3-2).
11 Ideally, a 2-dimensional probabilistic simulation would be used as the basis for a
12 quantitative analysis of uncertainty, where one dimension of the simulation reflected variability
13 in risk and the other the impact of multiple sources of uncertainty. However, this type of
14 probabilistic simulation requires that we have: (a) clearly-defined uncertainty distributions for
15 inputs to the risk model, (b) any existing correlations between those input parameters clearly
16 defined, and (c) confidence levels assigned to each model for those analysis steps where we have
17 multiple competing modeling options. At this point, we do not have sufficient information to
18 meet these requirements for many of the key modeling elements associated with the risk
19 assessment.
20 In the absence of these types of data required to support a 2-dimensional probabilistic
21 uncertainty analysis, deterministic sensitivity analysis-based methods can be used to support
22 quantitative characterization of uncertainty. Specifically, alternate modeling options for key
23 modeling elements can be considered (either one at a time, or in combination) to determine the
24 potential impact on risk estimates. The option of considering these factors in combination (i.e., a
25 multi-factor sensitivity analysis) can be used to derive a set of alternative plausible risk estimates
26 which provides insights into the impact of uncertainty on risk estimates. This application of
27 multi-factor sensitivity analysis forms the basis for the deterministic uncertainty analysis
28 approach developed for this risk assessment.
3-25
-------
Figure 3-2. Overview of Uncertainty Analysis Approach Developed for the PM NAAQS Risk Assessment
Risk Assessment Modeling Elements
Air Quality:
Ambient Monitoring Data for Selected
Urban Areas (specification of composite
monitor)
Modeled Background Concentrations
(PRB)
Air Quality Adjustment (roll-back)
Procedures
Concentration-Response:
Selection of Human Epidemiological
Studies to Provide Concentration-
Response Functions
Specification of Concentration -
Response Relationships - consider:
• single- v.s. multi-chemical
• single- v.s. multi-city
• lag
• seasonally-differentiated
• slope of CR function (threshold)
Baseline Health Effects Incidence Rates
and Demographics:
Estimates of City-specific Baseline
Health Effects Incidence Rates
City-specific Demographic Data
Primary Analysis Generating Core Set of Risk Estimates
Apply criteria for
selecting subset of
modeling element
options for use in
the core analysis
Apply single-factor
sensitivity analysis
to identify those
modeling elements
with a significant
impact on risk
results
Generate core set of risk
estimates for each
combination of urban study
area and air quality scenario
Uncertainty Analysis
Producing Additional Set of
Plausible Risk Estimates
Identify additional plausible
modeling options (distinct from
those used in the core analysis)
for this set of key modeling
elements
Conduct multi-factor sensitivity
analysis (for this subset of key
modeling options) to generate
alternative sets of reasonable risk
estimates for a subset of urban
study areas and air quality
scenarios
3-26
-------
1 The step-wise procedure for conducting the deterministic uncertainty analysis is
2 illustrated in Figure 3-2. It is important to point out that we plan to generate a core set of risk
3 estimates prior to conducting the uncertainty analysis. This core set of risk estimates would be
4 derived by first applying the criteria discussed in preceding sections (sections 3.2.1 through
5 3.2.5) to identify those options for key modeling elements which have the strongest scientific
6 support (with these determinations being based primarily on the evaluation provided in the
7 ISA).15 The core set of risk estimates will be generated for each combination of urban study area
8 and air quality scenario.
9 Once the core set of risk estimates has been generated, the uncertainty analysis will begin
10 with a single-factor sensitivity analysis intended to identify those modeling elements (comprising
11 the PM2.5 risk assessment framework) that have the potential to significantly impact risk
12 estimates. This set of key modeling elements would form the basis for the uncertainty analysis.
13 Next, plausible modeling options (distinct from those used in the core analysis) would be
14 specified for each of these key modeling elements. In identifying these plausible modeling
15 options, we plan to place emphasis on identifying input factors or modeling approaches which,
16 while representing alternatives to those used in the core simulation, still have some degree of
17 scientific support in the literature. Consequently, while we may have less confidence in risk
18 estimates generated using these alternate modeling options relative to the core risk estimates,
19 they could still considered reasonable and consequently may be interpreted as providing
20 additional perspective on overall uncertainty associated with the core set of risk estimates.
21 Once the set of plausible modeling options is specified for the key modeling elements, we
22 plan to use a multi-factor sensitivity analysis to generate a set of reasonable alternative risk
23 estimates. Specifically, various combinations of these alternative modeling options would be
15 For example, as noted in section 3.2.3, if a study provides both single-day and distributed lag models, the
distributed lag model would be used in the core analysis, while the individual day lags, if retained in the risk
assessment, would be included in the uncertainty analysis. With regard to non-linearity in functions, including the
potential for thresholds, generally non-threshold models will be used in the core analysis (based on information
provided in the ISA) and thresholds models, if they are considered at all, would be reserved for the uncertainty
analysis. It is also important to point out that for some of the modeling elements, multiple options may be included
as part of the core simulation (e.g., both multi- and single-component models may be used in core simulations for
specific health endpoints).
3-27
-------
1 used to generate risk estimates, each representing an uncertainty simulation.16 We plan to
2 generate this set of alternative risk estimates for a subset of the urban study areas and air quality
3 scenarios.
4 The combined sets of core results and alternative risk estimates (for a combination of
5 urban study area and air quality scenario) could be interpreted as representing an initial
6 characterization of risk for that combination of urban study area and air quality scenario,
7 reflecting recognized sources of uncertainty in risk modeling. However, this interpretation needs
8 to be tempered by consideration of several factors: (a) this does not represent a characterization
9 of a distribution of uncertainty around the core set of risk estimates, it merely represents several
10 point estimates likely falling within that uncertainty distribution and (b) the set of modeled risk
11 estimates may not contain actual upper-bound and lower-bound risk estimates given
12 scientifically defensible modeling options. Despite these caveats, the risk estimates defined by
13 the sets of core and alternative risk estimates should be useful to characterize confidence
14 associated with the results of the application of the PM2.5 risk assessment model.
15 3.4 PRESENTATION OF RISK ESTIMATES TO INFORM
16 CONSIDERATION OF STANDARDS
17 This section discusses the nature of the risk estimates that we plan to generate as part of
18 the review of the PM NAAQS. We plan to conduct the risk assessment in two phases. Phase 1
19 would include analysis of risk associated with recent air quality and simulating air quality to just
20 meet the current NAAQS. Phase 2 would focus on evaluating risk associated with simulating air
21 quality that just meets alternative NAAQS under consideration.
22 We plan to present risk estimates in two ways: (1) total (absolute) health effects
23 incidence (above PRB) for recent air quality and simulations of air quality just meeting the
24 current and alternative NAAQS under consideration, and (2) risk reduction estimates, reflecting
25 the difference between (a) risks associated with recent air quality compared to risks associated
26 with just meeting the current NAAQS and (b) reflecting the difference between risks associated
16 Note, that care would be taken in linking these modeling options together to insure that they are compatible and do
not represent combinations that are scientifically not defensible.
3-28
-------
1 with just meeting the current compared to risks associated with just meeting alternative NAAQS
2 under consideration.
3 In presenting risk estimates, we plan to emphasize the core (base-case) estimates given
4 that these would include risk estimates with greater overall confidence. We plan to also present
5 additional risk estimates generated as part of the uncertainty analyses in order to provide
6 additional context for understanding the potential impact of uncertainty on the risk estimates and
7 particularly on the core estimates of risk.
8 To further support interpretation of risk estimates generated in this analysis, we may also
9 consider the urban study areas with respect to the degree to which they represent the range of key
10 PM2.5 risk-related attributes that are spatially differentiated across the nation. In general, we plan
11 to consider the degree to which the urban study areas provide coverage for different regions of
12 the country, defined by similarities in PM2.5-related parameters. Alternatively, we are
13 consideration a more detailed analysis that would evaluate a set of spatially-distributed PM2 5-
14 related parameters (e.g., PM2 5 composition, air-conditioning use, demographics including socio-
15 economic status (SES), baseline health incidence rates). This analysis would allow us to
16 determine whether the selected urban study areas reflect national-level variability in these key
17 PM2 s-related parameters, or whether they are more concentrated in terms of their coverage.
18 Based on generally available data, e.g. from the 2000 Census, CDC, or other sources,
19 distributions for risk-related parameters across U.S. counties would be generated. The specific
20 values of these parameters for the selected urban study areas would then be plotted on these
21 distributions, and an evaluation of how representative the selected study areas are of the
22 individual parameters, relative to the national distributions, could be done.
23 The specific choices of parameters for which we would examine the representativeness of
24 the selected urban study areas would be informed through an assessment of the epidemiology
25 literature. We plan to particularly focus on meta-analyses and multi-city studies which have
26 identified parameters that influence heterogeneity in PM2 5 effect estimates, and exposure studies
27 which have explored determinants of differences in personal exposures to ambient PM2 5. While
28 personal exposure is not generally incorporated directly into epidemiology studies evaluating
29 ambient PM2.5-related effects, differences in the PM2.5 effect estimates between cities clearly is
3-29
-------
1 impacted by differing levels of those exposure determinants. Once we have identified these
2 parameters, we plan to develop national distributions for those parameters (or reasonable
3 surrogates) based on readily available data sources. Formal comparisons of parameter
4 distributions for the set of urban study areas and the national parameter distributions would be
5 conducted using standard statistical tests, e.g. the Kolmogorov-Smirnov non-parametric test for
6 equality of distributions. In addition, we plan to consider visual comparisons using probability
7 density functions, cumulative distribution functions, and boxplots.
8 3.5 NATIONAL SCALE HEALTH IMPACT ASSESSMENT FOR LONG-
9 TERM EXPOSURE MORTALITY RELATED TO PM2 5 EXPOSURES
10 While the risk analysis focused on selected urban study areas is designed to provide
11 information on risks in specific areas with a range of attributes reflecting different characteristics
12 of susceptibility and vulnerability, this assessment would not provide a sense of the overall
13 magnitude of the public health burden imposed by recent levels of ambient PM2.5. As noted
14 earlier, to address this broader question, we are considering conducting a national assessment of
15 the mortality impacts in the U.S. population associated with long-term exposures to ambient
16 PM2.5. This estimate could provide context for the magnitude of the long-term PM2.5 exposure-
17 related mortality estimated in the risk analysis for the selected urban study areas, as well as some
18 indication of the significance of the public health impacts associated with recent ambient PM2.5
19 levels. We are not considering an evaluation of all health impacts in this national assessment,
20 but rather, we would focus our analyses on mortality associated with long-term PM2.5 exposures
21 because of the significance of this endpoint and because of the strength of the available C-R
22 information.17 In addition, we would not consider it appropriate to compare the results of this
23 assessment to the results from the selected urban area case studies, due to differences in the
24 methods used to estimate air quality concentrations.
25 We are considering using EPA's peer-reviewed environmental Benefits Mapping
26 Analysis Program (BenMAP; Abt Associates, 2008) to estimate the total incidence of premature
17 Modeling of short-term (i.e., 24-hour average) PM2 5- related health impacts is subject to greater uncertainty than
modeling of long-term exposure-related mortality, reflecting the requirement (in modeling short-term endpoints) to
extrapolate city-specific studies to other locations in order to generate national-scale estimates. Due to the greater
3-30
-------
1 mortality associated with recent ambient PM2 5 concentrations. Similar to the urban study areas
2 analyses, the health impact functions for long-term exposure mortality require three elements: (1)
3 the effect estimate relating ambient PM2.5 concentrations to mortality, obtained from the
4 epidemiology literature, (2) baseline incidence of mortality, and (3) annual average ambient
5 PM2.s concentrations.
6 In the national-scale mortality impact analysis, we would consider using the same C-R
7 function(s) for mortality associated with long-term PM2 5 exposures as we plan to use in
8 modeling the selected urban study areas (e.g., those based on national cohort studies such as the
9 American Cancer Society study (Pope et al., 2004), and the Harvard Six-city Study (Laden et al.,
10 2006)). We would also obtain baseline incidence of mortality from the same sources as the
11 urban study areas risk assessment (i.e., the CDC Wonder database).
12 The national-scale mortality assessment would differ from the urban study areas analyses
13 in the methodology used to estimate ambient PM2.5. We would use a data fusion approach,
14 which combines monitored ambient PM2 5 concentrations with modeled PM2 5 concentrations
15 based on the Community Model for Air Quality (CMAQ). The data fusion approach is known as
16 enhanced Voronoi Neighbor Averaging (eVNA), which is essentially an inverse distance
17 weighted interpolation of monitored PM2 5 values, scaled to reflect the ratios of monitored to
18 modeled PM2 5 values. We would calculate ambient PM2 5 concentrations at a 12 km grid
19 resolution throughout the continental U.S. A modeled ambient PM2.5 field would be generated
20 based on recent U.S. conditions for a single year (e.g., 2005). This analysis would estimate the
21 health impact associated with ambient PM2.5 above the lowest observed annual average levels
22 measured in the cohort studies, recognizing that there is more empirical support for the C-R
23 function in the range above the lowest measured level for the time period examined in the
24 epidemiological studies.
25 There are a number of important uncertainties that enter into the national scale
26 assessment beyond those identified for the urban study areas assessment. These include
27 uncertainties regarding the estimation of PM2.5 air quality in areas away from monitors, which
degree of uncertainty associated with modeling short-term endpoints, we have decided to focus this national-scale
analysis on long-term exposure-related mortality.
3-31
-------
1 include both interpolation uncertainties and uncertainties in the CMAQ modeling (which include
2 uncertainties in both the CMAQ modeling structure, and in the inputs to the CMAQ modeling,
3 e.g., meteorology and emissions). Also, adding to the uncertainty is the application of the C-R
4 functions based on a limited number of urban areas to populations throughout the U.S. We
5 would need to include a qualitative discussion of key uncertainties in the assessment. This
6 qualitative assessment would, in part, be informed by the characterization of uncertainties
7 provided by the expert elicitation addressing PM2.s-related mortality (Roman et al., 2008).
8 Results from the national scale mortality impact assessment described above would
9 include national incidence estimates, as well as percent of total incidence attributed to exposure
10 toPM2.5. Analyses would include results based on multiple C-R functions. Results would also
11 include a limited quantitative uncertainty assessment based on the statistical error reported in the
12 epidemiological studies.
13 In addition to providing an overall estimate of the mortality impact associated with recent
14 long-term ambient PM2.5 concentrations, the national scale mortality impact assessment would
15 estimate the cumulative distribution of mortality risk from long-term PM2.5 concentrations across
16 the U.S. population. This would allow us to assess the degree to which our selected urban study
17 areas analysis characterizes long-term mortality risk for the more vulnerable and susceptible
18 populations in the U.S. In other words, do the urban study area risk estimates include the upper
19 end of the national risk distribution, or are they more representative of a different part of the
20 distribution? This analysis would not involve a direct comparison of risk estimates for the urban
21 study areas compared with risk levels generated for those same urban areas within the national -
22 scale analysis, but rather would focus on comparisons of the urban area risk estimates with
23 percentiles of the national risk distribution.
24 3.6 OVERVIEW OF HEALTH RISK ASSESSMENT APPROACH
25 DEVELOPED FOR PMi0.2 5
26 As noted in Section 3.1, we plan to build on the risk assessment completed in the last PM
27 NAAQS review to conduct a risk assessment for PMio-2.5. While the first draft ISA does not
28 identify any of the health effect endpoint categories evaluated for PMio-2.5 to have sufficient
29 support to be classified as causal or likely causal, we believe that the available evidence is
3-32
-------
1 sufficiently suggestive of a causal association to support a limited risk assessment for some
2 health endpoints related to short-term (daily) exposures to PMio-2.5- The prior PMio-2.5 risk
3 assessment included hospital admissions for cardiovascular and/or respiratory causes in two
4 locations (Detroit and Seattle) and respiratory symptoms in St. Louis. For the current review, we
5 plan to review the scientific evidence for the health endpoints included in the prior risk
6 assessment and to consider whether there are additional C-R relationships and locations that
7 should be included in a PMio-2.5 risk assessment for these health outcomes. Further, we also plan
8 to consider whether there is sufficient information to warrant inclusion of an expanded set of
9 health endpoints such as additional respiratory morbidity outcomes (e.g., asthma-related
10 outpatient and emergency department visits and hospitalization) as well as mortality. Our
11 judgments about which endpoints and locations to include in the risk assessment will be
12 informed by the continuing scientific assessment as well as comments provided by the CAS AC
13 panel and general public. A provisional set of studies that we would consider in identifying
14 health effects endpoints to potentially model for PMio-2.5 (based on information provided in the
15 first draft ISA) is provided in Table 3-3.
16 We recognize that there will be significantly greater uncertainties associated with PMio-2.5
17 risk estimates relative to the planned PM2 5 risk assessment due to a number of factors including:
18 (1) much greater uncertainty in whether an observed association reflects a causal relationship for
19 PMio-2.5 and various health outcomes; (2) more limited PMio-2.5 ambient air quality data which
20 often is obtained by calculation from measurements for co-located PMio and PM2.5 monitors; (3)
21 far fewer health studies and locations with more mixed effect estimates for PMio-2.5; and (4)
22 greater concern about exposure measurement error for PMio-2.5 compared to PM2.5 in large part
23 due to the greater spatial variability in ambient PMio-2.5 concentrations and exposures.
24 The basic approach we plan to use to estimate risks for PMio-2.5 is identical to the
25 approach used for PM2.5 as illustrated in Figure 3-1. In order to estimate the incidence of a
26 particular health effect associated with recent conditions in a specific county or set of counties
27 attributable to PMio-2.5 daily exposures in excess of PRB, as well as the change in incidence of
28 the health effect in that county or set of counties corresponding to a given change in
3-33
-------
1 Table 3-3 Provisional List of Studies to be Considered for PMi0-2.s Risk Assessment:
2 Cardiovascular and Respiratory Morbidity Associated with Short-term PMi0_2.5 Exposure
Urban
Location
Detroit, MI
Seattle, WA
St. Louis,
MO
Spokane,
WA
Atlanta, GA
108 U.S.
Counties
Cardiovascular Hospital
Admissions
Lippmann et al. (2000)1
Congestive heart disease,
Ischemic heart disease
Dysrhythmias
Respiratory Hospital
Admissions (and
Emergency Room Visits*)
Lippmann et al. (2000)1
Pneumonia,
COPD+
Sheppard etal. (1999)2
Asthma
Slaughter et al. (2005)*
All respiratory
Acute asthma
COPD
Peel et al. (2005)*
All respiratory
Upper respiratory infection
Asthma
Pneumonia
COPD
Peng et al. (2008)
Cardiovascular disease
Peripheral vascular disease
Respiratory Symptoms
Schwartz and Neas (2000)
Lower respiratory symptoms
1 Reanalyzed in Ito (2003); COPD+ is indicated here because the authors included asthma in their definition of
COPD. 2 Reanalyzed in Sheppard (2003).
4 ambient PMio-2.5 levels resulting from just meeting a specified set of alternative PMio-2.5
5 standards under consideration, the following three elements are required:
6 • Air quality information including: (1) recent air quality data for PMio-2.5 from
7 ambient monitors for the selected locations, (2) estimates of PRB appropriate for
8 these locations, and (3) a method for adjusting the recent data to reflect the
9 patterns of air quality estimated to occur when an area just meets a alternative
10 PMio-2.5 standards under consideration;
11 • Relative risk-based C-R functions (preferably derived from epidemiological
12 studies conducted in the assessment location) which provide an estimate of the
13 relationship between the health endpoints of interest and ambient PMio-2.5
14 concentrations;
15 Annual or seasonal baseline health effects incidence rates and population data are needed to
16 provide an estimate of the annual or seasonal baseline incidence of health effects in an area
17 before any changes related to PMio-2.5 air quality could be considered.
3-34
-------
i 4 SCOPE AND APPROACH FOR POPULATION
2 EXPOSURE ANALYSIS
3 4.1 INTRODUCTION
4
5 As part of the last PMNAAQS review, EPA did not conduct an exposure assessment.
6 There is much new scientific information available since the last review on human health effects,
7 exposure, and PM air quality, which provides the basis for conducting a quantitative exposure
8 assessment in this review. This assessment is planned to focus on evaluating fine particle
9 exposures (PM^.s) and will build upon the information presented in the ISA. This assessment
10 will include discussions of factors that affect exposure to ambient PM2.5 and the use of fixed site
1 1 measurements of ambient PM2 5 concentrations as a surrogate for population (or community)
12 average exposure to ambient PM in epidemiologic studies. There are two specific purposes that
13 such an assessment could serve: (1) providing insight on population exposures with respect to
14 informing the interpretation of available epidemiologic studies; and (2) assessing population
1 5 exposures above benchmark levels of concern, and providing input to quantitative risk
16 assessments based on evidence from clinical studies. At this time, based upon the first draft ISA
17 (U.S., EPA, 2008b), we are unaware of any results from human clinical studies that would
18 provide the basis for exposure-response functions that could inform a quantitative risk
19 assessment or benchmark levels of concern; therefore, the planned assessment focuses on the
20 first purpose. The available monitoring data for PMio-2.5 are much more limited than ambient
21 monitoring data available for PM2.5 and do not provide enough spatial coverage for exposure
22 modeling to be credible. Therefore, we do not plan on conducting an exposure assessment for
23 thoracic coarse particles
24 Performing an exposure analysis will be helpful for identifying the various personal and
25 building-related factors which may be responsible for some of the differences observed in
26 epidemiologic studies of ambient PM2.5. Exposure-related factors may contribute to city -to-city
27 differences (mostly seen in time-series studies) in the reported PM2 5 concentration-response
28 (C-R) functions or in the results from intra-urban studies (e.g., cohort studies of long-term
4-1
-------
1 exposures to PM2 5). Thus, an exposure assessment for PM2 5 has the potential to shed some light
2 on these and other issues related to uncertainties in the existing PM2.5 epidemiology literature.
3 We plan to model population exposures to ambient PM2.5 in approximately 10 urban
4 areas across the U.S. where epidemiological studies have been conducted that are the basis of C-
5 R functions used in the planned quantitative risk assessment described in Chapter 3. The basis
6 for the selection of areas to model is described in section 4.4 below. The exposure periods we
7 plan to model will be the periods studied in the corresponding epidemiologic studies.
8 The primary measures of interest are estimated population distributions of 24-hour
9 average exposures. Additional quantitative measures of exposure may take into account factors
10 including the magnitude and duration of PM2.5 exposures and the frequency of repeated peak
11 exposures. We plan to develop estimates for population exposures associated with historical
12 PM2.5 levels only. We do not plan to estimate population exposures for scenarios that simulate
13 just meeting the current PM2.5 standards or potential alternative primary PM2.5 standards under
14 consideration, as these are not directly relevant to interpreting currently available epidemiologic
15 studies based on past environmental conditions.
16 4.2 THE APEX POPULATION EXPOSURE MODEL
17 As stated in the Integrated Review Plan (U.S. EPA, 2008a, section 5.4), EPA considered
18 using the Air Pollutants Exposure (APEX) model (Richmond et al., 2002; U.S. EPA, 2008 e,f)
19 and/or the Stochastic Human Exposure and Dose Simulation model for PM (SHEDS-PM) (Burke
20 and Vedantham, 2005) in conducting an exposure assessment for this review. 18 APEX, also
21 referred to as the Total Risk Integrated Methodology/Exposure (TREVI.Expo) model, has its
22 origins in the NAAQS Exposure Model (NEM), which was developed in the early 1980's (Biller
18 A third model considered for this application is the MENTOR-1A (Modeling ENvironment for TOtal Risk studies
in a "One Atmosphere" setting) model. MENTOR-1 A includes both gas and aerosol phase microenvironmental
processes (Georgopoulos & Lioy, 2006; Georgopoulos et al., 2005). MENTOR-1A also includes optional selection
of Models-3/CMAQ compatible modules for microenvironmental transformations, both physical (e.g. condensation
to form secondary aerosol) and chemical (with CB4 and RADM2 based alternative mechanisms). MENTOR-1A
performs exposure calculations for multiple pollutants and allows for direct linkage to compute the uptakes, and
links with physiologically based toxicokinetic models (PBTK) for assessing distributions of target tissue doses for
populations, which can be used for evaluation with population-based biomarker studies. Since the underlying
exposure model of MENTOR-1A is similar to SHEDS-PM and the additional complexity involving chemistry and
dose are not required for this application, this model was not selected for use.
4-2
-------
1 et al., 1981; McCurdy, 1994, 1995). SHEDS-PM was developed by EPA/ORD's National
2 Exposure Research Laboratory (NERL) (Burke et al., 2001; Ozkaynak et al 2009). APEX and
3 SHEDS-PM are Monte Carlo simulation models that simulate a large number of randomly
4 sampled individuals within a metropolitan area to represent area-wide population exposures.
5 Although these models were developed independently, they are both based on the current state of
6 knowledge of inhalation exposure modeling and are fundamentally similar. Both models
7 simulate the movements of individuals through time and space and their exposure to a given
8 pollutant in indoor, outdoor, and in-vehicle microenvironments. The models stochastically
9 generate simulated individuals using census-derived probability distributions for demographic
10 characteristics. A large number of simulated individuals are modeled and collectively they
11 represent a random sample of the study area population. Research conducted by NERL has been
12 used in the development of the APEX model in a number of areas including: incorporation of the
13 excess post-exercise oxygen consumption model (McCurdy, 2000; Isaacs et al., 2008); inclusion
14 of an improved method for constructing longitudinal diaries (Glen et al., 2008); continuation of
15 updates to the Consolidated Human Activity Database (CHAD) (McCurdy et al., 2000); and
16 incorporation of a model for calculating deposition of PM to lungs.
17 EPA staff has considered the features of these two models for the specific purpose of
18 conducting an exposure assessment to inform this review of the PM NAAQS and has concluded
19 that APEX is the most appropriate model for this particular regulatory application. The rationale
20 for selecting the APEX model includes consideration of the following factors:
21 • APEX is available to the public with non-proprietary code and documentation,
22 whereas, at this time, SHEDS-PM is not. This allows interested members of the
23 public access to the model and, if desired, the ability to recreate the technical analyses
24 supporting the NAAQS review. Transparency and availability of the methods and
25 data used in the technical analyses is an integral part of the NAAQS review process
26 and encouraged by the provisions of the Information Quality Act.
27 • APEX was designed to allow users the ability to model any airborne pollutant. This
28 model has been primarily used for regulatory purposes, with significant applications
29 to inform the recent reviews of the ozone, sulfur dioxide, and nitrogen dioxide
30 NAAQS. Periodic reviews by CASAC and the public and application of APEX in
4-3
-------
1 other NAAQS reviews have resulted in an ongoing and chronologically documented
2 process of refinement of the model.
3 • APEX can be run using a batch mode, which is particularly important for this
4 application, to allow for the large number of simulations that are performed for
5 NAAQS assessments.
6 • Quantitative assessment of uncertainty and variability, a critical component of this
7 analysis, is integrated into SHEDS-PM so that a single model run yields both
8 variability and uncertainty results (Burke et al., 2001). In order to analyze uncertainty
9 with APEX, we plan to use external software that modifies APEX inputs according to
10 distributions reflecting uncertainty and runs APEX numerous times for the outer
11 uncertainty loop (see section 4.8). This would result in a 2-dimensional Monte Carlo
12 treatment that tracks variability and uncertainty separately. While this approach
13 requires more effort than an uncertainty analysis with SHEDS-PM, the flexibility of
14 this method is deemed important for performing the planned assessment of
15 uncertainty.
16 APEX simulates the movement of individuals through time and space and their exposure
17 to a given pollutant in indoor, outdoor, and in-vehicle microenvironments. Figure 4-1 provides a
18 schematic overview of the APEX model. The model stochastically generates simulated
19 individuals using census-derived probability distributions for demographic characteristics
20 (Figure 4-1, steps 1-3). The population demographics are from the 2000 Census data at the tract
21 or block level, and a national commuting database based on 2000 Census data provides home-to-
22 work commuting flows between tracts. A large number of simulated individuals are modeled,
23 and collectively, they represent a random sample of the study area population.
24 Diary-derived time activity data are used to construct a sequence of activity events for
25 each simulated individual consistent with the individual's demographic characteristics and
26 accounting for effects of day type (e.g., weekday, weekend) and outdoor temperature on daily
27 activities (Figure 4-1, step 4). APEX calculates the concentration in the microenvironment
28 associated with each event in an individual's activity pattern and sums the event-specific
29 exposures within each hour to obtain a continuous time series of hourly exposures spanning the
30 time period of interest (Figure 4-1, steps 5 and 6). From these exposure estimates, APEX
31 calculates exposures for averaging times greater than one hour.
4-4
-------
Figure 4-1. Overview of the APEX Model
1. Characterize study area
2. Characterize study population
3. Generate N number of
simulated individuals (profiles)
2000 Census tract-level data for the entire U.S. (sectors=tracts for the NAAQS ozone exposure application)
Sector location data f
(latitude, longitude) 1
Defined study area (sectors
within a city radius and with air
quality and meteorological data
within their radii of influence)
Locations of air quality and
meteorological measurements;
radii of influence
Sector population data
(age/gender/race)
Commuting flow data
(origin/destination sectors)
-^
-N
i
>
r
1^
Population within
the study area
Age/gender/tract-specific
employment probabilities
physiological
distribution data (body
weight, height, etc)
Stochastic
profile generator
Distribution functions for
profile variables
(e.g, probability of air
conditioning)
Distribution functions
for seasonal and daily
varying profile variables
(e.g., window status, car
speed)
( 0
- National
database
(^ J) - Simulation <^
step ^"~~~--.
- Area-specific
input data
j> - Data processor
- Intermediate step
or data
±\ - Output data
A simulated individual with the
following profile:
• Home sector
• Work sector (if employed)
•Age
• Gender
• Race
• Employment status
• Home gas stove
• Home gas pilot
• Home air conditioner
• Car air conditioner
• Physiological parameters
(height, weight, etc.)
4-5
-------
Figure 4-1. Overview of the APEX Model, continued
4. Construct sequence of activity events
for each simulated individual
Diary events/activities
and personal information
(e.g., from CHAD)
Activity diary pools by day
type/temperature category
Profile for an
individual
Stochastic diary
selector using
age, gender, and
employment
Selected diary records for each day in the simulation
period, resulting in a sequence of events
(microenvironments visited, minutes spent, and
activity) in the simulation period, for an individual
Each day in the simulation
period is assigned to an
activity pool based on day type
and temperature category
Maximum/mean daily
temperature data
Stochastic
calculation of energy
expended per event (adjusted for
hysiological limits and EPOC)
and ventilation
rates
Physiological
parameters from
profile
Sequence of events for an
individual
4-6
-------
Figure 4-1. Overview of the APEX Model, concluded
5. Calculate concentrations in
microenvironments for all events for
each simulated individual
6. Calculate hourly
exposures for each
simulated individual
7. Calculate
population exposure
statistics
Microenvironments defined by
grouping of CHAD location codes
Hourly air quality
data for all sectors
Select calculation method for
each microenvironment:
• Factors
• Mass balance
Calculate
concentrations in all
microenvironments
I
Average exposures
for simulated person,
stratified by ventilation
rate:
• Hourly
• Daily 1-hour max
• Daily 8-hour max
Daily
Population exposure
indicators for:
• Total population
• Children
• Asthmatic children
Hourly concentrations and
minutes spent in each
microenvironment visited by
the simulated individual
Concentrations for all events
for each simulated individual
Sequence of events for
each simulated individual
Calculate hourly
concentrations in
microenvironments
visited
4-7
-------
1
2
3
4
6
7
8
9
10
11
12
13
APEX employs a flexible approach for simulating microenvironmental concentrations,
where the user can define any number of microenvironments to be modeled and their
characteristics. For this modeling application, we propose modeling the microenvironments
listed in Table 4-1.
Table 4-1 Microenvironments to be Modeled
Microenvironment
Method
Indoors - residence
Indoors - restaurants, bars
Indoors - schools
Indoors - day care centers (commercial)
Indoors - offices
Indoors - shopping malls
Indoors - other (e.g., stores not in malls)
Outdoors - bus stop
Outdoors - near road
Outdoors - other (e.g., playgrounds, parks)
In vehicle - cars and light trucks
In vehicle - heavy trucks
In vehicle - school buses
In vehicle - mass transit vehicles - buses and trolleys
In vehicle - mass transit vehicles - underground (subways)
mass balance
mass balance
mass balance
mass balance
mass balance
mass balance
or factors
mass balance
or factors
factors
factors
factors
mass balance
or factors
mass balance
or factors
mass balance
or factors
factors
factors
We plan to calculate the concentrations in each microenvironment using either a factors
or mass-balance approach19, depending upon data availability, with probability distributions for
the parameters that enter into the calculations (e.g., indoor-outdoor air exchange rates) supplied
as inputs to the model. These distributions represent the variability of parameters, and can vary
spatially and can be set up to depend on the values of other variables in the model. For example,
the distribution of air exchange rates in a home, office, or car depends on the ambient
temperature and the type of heating and air conditioning present. The user can choose to keep
the value of a stochastic parameter constant for an individual for the entire simulation (e.g.,
19
The factors and mass-balance approaches are described in section 4.5.5.
4-8
-------
1 house volume), or can specify that a new value shall be drawn hourly, daily, or seasonally from
2 specified distributions. APEX also allows the user to specify diurnal, weekly, and seasonal
3 patterns for microenvironmental parameters.
4 4.3 MEASURES OF EXPOSURE
5 EPA plans to estimate distributions of 24-hour average exposures using APEX for
6 population groups represented in epidemiologic studies. We are considering whether shorter
7 averaging times would also be informative for differentiating exposure characteristics in
8 different cities. We may consider additional indicators of exposure that incorporate factors such
9 as relationships between magnitude and duration of exposures, frequency of repeated high
10 exposures, and ventilation rate (i.e., breathing rate) of the individual at the time of exposure. In
11 addition to these measures, we plan to summarize distributions of personal exposure factors,
12 which we define as ratios of personal exposures to ambient concentrations. To support
13 interpretation of multi-city epidemiologic studies, we plan to conduct an analysis of the
14 dependence of these distributions and other exposure measures on city-specific attributes.
15 4.4 SELECTION OF URBAN AREAS AND TIME PERIODS
16 EPA plans to conduct exposure analyses for about 10 metropolitan areas that have been
17 considered in multi-city epidemiologic studies. We plan to use multi-city studies instead of
18 individual city studies because some sources of inter-city heterogeneity may be avoided using
19 these data, such as the use of different statistical methods, different treatment of confounders, the
20 use of different types of data, and publication bias. The time periods and geographic areas to be
21 modeled will coincide with those of the corresponding epidemiologic studies conducted in these
22 areas. The selection of urban study areas to include in the exposure analysis will also take into
23 consideration the availability of ambient PM2.5 data, and the desire to represent a range of
24 geographic areas, population demographics, PM2.5 climatology, PM2.5 species and air pollution
25 composition, and other variables that could account for variation in population exposures
26 between cities.
4-9
-------
1 4.5 DEVELOPMENT OF MODEL INPUTS
2 4.5.1 Population Demographics
3 We plan to obtain tract-level population counts from the 2000 Census of Population and
4 Housing Summary File 120. Summary File 1 (SF 1) contains the 100-percent data, which is the
5 information compiled from the questions asked of all people and about every housing unit.
6 In the 2000 U. S. Census, estimates of employment were developed by census tract21.
7 The file input to APEX will be broken down by gender and age group, so that each gender/age
8 group combination is given an employment probability fraction (ranging from 0 to 1) within each
9 census tract. The age groupings in this file are: 16-19, 20-21, 22-24, 25-29, 30-34, 35-44, 45-54,
10 55-59, 60-61, 62-64, 65-69, 70-74, and greater than 75 years of age. Children under 16 years of
11 age will be assumed to be not employed.
12 4.5.2 Commuting
13 As part of the population demographics inputs, it is important to integrate working
14 patterns into the assessment. In addition to using estimates of employment by tract, APEX also
15 incorporates home-to-work commuting data. We plan to use the national commuting database
16 provided with APEX in this analysis. Commuting data were derived from the 2000 Census and
17 were collected as part of the Census Transportation Planning Package (CTPP) (U.S. DOT,
18 2000)22. The data used to generate APEX inputs were taken from the "Part 3-The Journey To
19 Work" files. These files contain counts of individuals commuting from home to work locations
20 at a number of geographic scales. These data have been processed to calculate fractions for each
21 tract-to-tract flow to create the national commuting data distributed with APEX. This database
22 contains commuting data for each of the 50 states and Washington, D.C. This data set does not
23 differentiate people that work at home from those that commute within their home tract.
20 http://www.census.gov/prod/cen2000/doc/sfl.pdf
21 Employment data from the 2000 Census can be found on the U. S. Census web site:
http://www.census.gov/population/www/cen2000/phc-t28.html (Employment Status: 2000- Supplemental Tables).
22 These data are available from the U.S. DOT Bureau of Transportation Statistics (BTS) at the web site:
http://transtats.bts.gov/.
4-10
-------
1 4.5.3 Ambient PM2.s Concentrations
2 We plan to conduct exposure modeling based on PM2 5 concentrations measured at
3 ambient air monitors in and near the areas being modeled. Sources for these data include the
4 daily and hourly concentration measurements from the monitoring data maintained in EPA's Air
5 Quality System (AQS), the Interagency Monitoring of Protected Visual Environments network
6 (IMPROVE, 2008), and the Canadian National Air Pollution Surveillance (NAPS) network
7 (Environment Canada, 2008). We plan to input hourly ambient concentrations into APEX, and
8 allocate the daily data to hourly values according to the diurnal profiles of the hourly monitors.
9 Methods for spatial interpolation and for estimating PM2.5 concentrations near roadways are
10 being developed.
11 4.5.4 Meteorological Data
12 Surface meteorological observations will be obtained from the National Climatic Data
13 Center23 to provide hourly temperatures for input to APEX. We plan to use all meteorological
14 stations within and nearby each selected urban study area.
15 4.5.5 Specification of Microenvironments
16 Parameters defining each microenvironment will be specified by distributions which
17 reflect the variability of these parameters. The parameters needed depend on whether a
18 microenvironment is modeled using the factors model or the mass balance model.
19 We plan to use the factors model to model simple environments, like outdoor areas, that
20 do not contain pollutant sources, or microenvironments for which data are not available to use
21 the mass-balance model. Two parameters affect the pollutant concentration calculation in the
22 factors method, the proximity and infiltration factors. The proximity factor (FPR) is a unitless
23 parameter that represents the relationship of the ambient concentration outside of the
24 microenvironment (Co) to the concentration at a monitoring station (d) by the equation Co =
25 FPR CA. The infiltration factor (Finf) is a unitless parameter that represents the equilibrium
26 fraction of pollutant entering a microenvironment from outside the microenvironment and
' See http://www.ncdc.noaa.gov/oa/ncdc.html
4-11
-------
1 remaining suspended. The concentration inside the microenvironment (C/) is estimated by the
2 equation C/ = FinfCo. The infiltration factor in the factors model is often expressed as:
F
J
,
a + k
4 where Pisa penetration coefficient, a is an air exchange rate, and k is a particle loss rate (ISA,
5 page 3-130). APEX draws values of these parameters from distributions specified by the user, to
6 model the stochastic nature of these factors.
7 The mass balance model is more appropriate for complex environments. The mass
8 balance method assumes that an enclosed microenvironment (e.g., a room in a residence) is a
9 single well-mixed volume in which the air concentration is approximately spatially uniform.
10 APEX estimates the concentration of an air pollutant in such a microenvironment by using the
1 1 following four processes (as illustrated in Figure 4-2):
12
13 • Inflow of air into the microenvironment;
14 • Outflow of air from the microenvironment;
15 • Removal of a pollutant from the microenvironment due to deposition, filtration, and chemical
16 degradation; and
17 • Emissions from sources of a pollutant inside the microenvironment.
4-12
-------
Microenvironment
Air
outflow
7
Indoorsources
Air
inflow
Removal due to:
•Chemical reactions
•Deposition
•Filtration
1
2 Figure 4-2. The Mass Balance Model
3
4 Considering the microenvironment as a distinct, well-mixed volume of air, the mass
5 balance equation for a pollutant can be described by:
6
7
10
11
12
13
dC(t) dCm(t} dCout(t) dCloss(t)
dt dt dt dt dt
where:
C(t) = Concentration in the microenvironment at time t (|ig/m )
dt
= Rate of change in C(t) due to air entering the micro
— — — = Rate of change in C(t) due to air leaving the micro
dt
— ' = Rate of change in C(t) due to all removal processes
dt
dt
= Rate of change in C(t) due to all source terms
4-13
-------
1 In addition to proximity factors, this method supports parameter distributions for time
2 varying emissions sources, decay rate, air exchange rate, volume, and removal rate. We plan to
3 estimate the distributions of these parameters based on available data and a review of the
4 literature.
5 4.5.6 Indoor Sources
6 We do not plan to model indoor sources of PM2.5 in this analysis, since our focus is on
7 exposure to PM2.5 of ambient origin. Differences in the characteristics of indoor sources for
8 different cities may be a factor contributing to inter-city heterogeneity of C-R functions;
9 however, staff feels that there is not sufficient city-specific information in indoor source types
10 and usage to be informative in this analysis, with the possible exception of wood stoves. The
11 formation of ultrafme PM via chemical reactions of air fresheners and ozone (Weschler et al.,
12 2003) is considered an indoor source, and will not be modeled.
13 The "personal cloud" is the assemblage of particles adjacent to a person that is generated
14 by personal activities such as resuspension of particles from floors, furniture, or clothes, and not
15 by other indoor sources of PM. In order to be able to model this in the framework of a mass-
16 balance model, the personal cloud is defined in terms of processes that generate the PM personal
17 cloud, and not as the difference between predicted and measured personal PM exposures. The
18 extent to which the particles comprising personal clouds are of ambient origin is not known, and
19 will depend heavily on household and behavioral characteristics. Wallace et al. (2006) report
20 estimates from various studies of the indoor personal cloud ranging from 2-4 ug/m3. It seems
21 reasonable to assume that most of this is not of ambient origin and that the personal cloud is not
22 likely to be a significant contributor to the highest 24-hour or annual average exposures of the
23 population. Therefore, we do not plan to model the personal cloud.
24 4.5.7 Activity Patterns
25 Exposure models use human activity pattern data to predict and estimate exposure to
26 pollutants. Different human activities, such as outdoor exercise, indoor reading, or driving, have
27 different pollutant exposure characteristics. In addition, different human activities require
4-14
-------
1 different metabolic rates, and higher rates lead to higher doses. To accurately model individuals
2 and their exposure to pollutants, it is critical to have a firm understanding of their daily activities.
3 The Consolidated Human Activity Database (CHAD) provides data on human activities
4 through a database system of collected human diaries, or daily activity logs (McCurdy et al.,
5 2000; US EPA, 2002; Graham and McCurdy, 2004). The purpose of CHAD is to provide a basis
6 for conducting multi-route, multi-media exposure assessments (McCurdy et al., 2000). The data
7 contained within CHAD come from multiple surveys with varied structures (Table 4-2). In
8 general, the surveys have a data foundation based on daily diaries of human activity. Individuals
9 filled out diaries of their daily activities and this information was entered and stored in CHAD.
10 Relevant data for these individuals, such as age, are included as well. In addition, CHAD
11 contains activity-specific metabolic distributions developed from literature-derived data, which
12 are used to provide an estimate of metabolic rates of respondents through their various activities.
13 The locations used in the CHAD diaries must be assigned appropriately to the APEX
14 microenvironments listed in Table 4-1. Each of the microenvironments is designed to simulate
15 an environment in which people spend time during the day. There are many more CHAD
16 locations than microenvironments being modeled (there are over 100 CHAD locations and 15
17 proposed microenvironments modeled in this assessment) thus, most of the microenvironments
18 have multiple CHAD locations mapped to them.
4-15
-------
1 Table 4-2 Studies in CHAD used in this analysis
Study name
Baltimore
California Adults
(CARB)
California Children
(CARB)
California
Adolescents (CARB)
Cincinnati (EPRI)
Denver (EPA)
Los Angeles:
Elementary School
Los Angeles: High
School
NHAPS2-Air
Geographic
coverage
One building
in Baltimore
California
California
California
Cincinnati
metro, area
Denver
metro, area
Los Angeles
Los Angeles
National
Study time
period
1/1997-2/1997,
7/1998-8/1998
10/1987-9/1988
04/1989-2/1990
10/1987-9/1988
3/1985, 8/1985
11/1982-2/1983
10/1989
10/1990
9/1992-10/1994
Subject
ages
72-93
18-94
<1-11
12- 17
-------
NHAPS-Water
PSID CDS3 1
PSID CDS II
Seattle
RTP Panel
Washington, D.C.
Totals
National
National
National
Seattle, WA
Chapel Hill,
Raleigh, NC
Wash., D.C.
metro, area
9/1992-10/1994
3/1997-6/1997,
9/1997-12/1997
10/2002-6/2003
12/2000-5/2001
6/2000-5/2001
11/1982-2/1983
-------
1 4.6 PILOT EXPOSURE ASSESSMENT
2 We are conducting PM2.5 exposure modeling for one city (Detroit, Michigan) for two
3 months (January and July 2002) to explore ways of approaching some exposure modeling issues
4 before we start on the multi-city, multi-year exposure modeling we plan to conduct for this
5 assessment. This pilot exposure assessment is planned to provide additional information to
6 inform our proposed exposure modeling approach. This urban area and time period were selected
7 for the pilot assessment since air quality modeling results using AERMOD24, CAMx25, and
8 CMAQ26 at a detailed spatial resolution were readily available. At this time, the pilot exposure
9 modeling is not far along enough to give insights into the issues that we are trying to address.27
10 In this section, we discuss how we plan to approach this analysis. The preparation of inputs to
11 the exposure model (APEX) for this pilot assessment is described in Appendix A.
12 Estimation of ambient concentrations between monitoring sites
13 An evaluation of methods for spatially interpolating concentrations is being conducted
14 using air quality modeling as well as monitoring data, and a method will be selected taking into
15 account the performance of methods and the effort required to apply the methods.
16 Development of a parametric model for estimation of ambient concentrations at or close to
17 roadways
18 It is likely that the use of air quality models in conjunction with monitoring data will
19 result in more accurate estimates of ambient concentrations near roadways than interpolation
20 methods relying solely on monitored concentrations, because the air quality models have the
21 potential to be able to resolve near-roadway concentration gradients that the monitors cannot.
22 Air quality modeling using link-based roadway emissions can be used for estimating
23 ambient concentrations close to roadways (US EPA, 2008c). However, the time and resources
24 required to prepare the link-based emissions inventories and the meteorological model inputs to
24 AERMOD is an EPA regulatory air quality dispersion model that replaces ISC3. See
http://www.epa.gov/scram001/dispersion_prefrec.htm
25 See http://www.camx.com
26 See http://www.epa.gov/asmdnerl/CMAQ/cmaq_model.html
27 To the extent that results are available in advance of the upcoming CASAC consultation meeting on this planning
document, we plan to provide a summary of results at that time.
4-18
-------
1 run air quality models for an urban area preclude their use for more than a few cities in the
2 context of this review. We plan to perform air quality modeling for a small number of urban
3 areas and develop a stochastic model for estimating near-road concentrations based on the air
4 quality modeling results. If this modeling effort is successful, this algorithm would be used in
5 exposure modeling for the full set of cities to be modeled.
6 The dependent variables in this model are the near-road concentrations estimated by the
7 air dispersion modeling. A set of independent variables to be considered is being developed,
8 including block-level Census data and measures of local roadway intensity which incorporate
9 summations of nearby roadways, weighted by a surrogate for traffic volume and inverse distance
10 from roadway. These variables will need to be readily available for large urban areas in the U.S.
11 Estimation of ambient concentrations for days with no measurements at sites that provide
12 measurements every three or every six days
13 We will consider a method similar to that used for hourly data (described in Appendix A)
14 as well as methods of spatial interpolation for filling in missing daily concentrations.
15 Estimation of hourly concentrations from daily monitoring data
16 One approach to this issue could be to apply the diurnal profiles from nearby hourly
17 monitors to the 24-hour average PM2.5 concentrations at a daily monitor (not changing the 24-
18 hour averages). Air quality modeling may be used to estimate hourly concentrations from 24-
19 hour monitoring data by applying the modeled diurnal profile to the 24-hour monitored PM2.5
20 concentrations. Alternatively, we are considering using some combination of air quality
21 modeling and hourly measurements. An analysis of these options is planned
22 Evaluation of the utility of the refinement of PM2.s into size classes for the
23 microenvironmental mass-balance calculations
24 We plan to perform exposure model simulations with and without resolution into PM2.5
25 size classes. If we have alternative ways of apportioning PM2.5 into size classes, we will
26 consider conducting simulations using this information. We plan to assess these results to see
27 whether or not the refinement of microenvironmental parameters into size classes of PM2.5
28 makes any significant difference in the measures of exposure of interest.
4-19
-------
1 Modeling people's near-road and in-vehicle activity patterns
2 Near-road and in-vehicle activities are important components of exposure to PM; staff are
3 working to develop an improved methodology for characterizing these activities.
4 4.7 EXPOSURE MODELING ISSUES
5 In this section, we highlight some aspects of the proposed exposure modeling that have
6 the potential to significantly contribute to uncertainties in the exposure analysis. These aspects
7 of people's exposures are either not modeled or are based on limited information. The
8 uncertainty analysis will attempt to quantitatively estimate the effect of these aspects on the
9 modeled exposures.
10 • Representativeness of Personal Activity Patterns
11 The human activity data will be drawn from CHAD. The CHAD includes data from
12 several surveys covering specific time periods at city, state, and national levels, with
13 varying degrees of representativeness. The extent to which the human activity database
14 provides a balanced representation of the population being modeled varies across areas.
15 Although the algorithm that constructs activity sequences attempts to account for the
16 effects of population demographics and local climate on activity, this adjustment
17 procedure does not fully account for all intercity differences in people's activities.
18 Activity patterns are affected by many local factors, including topography, land use,
19 traffic patterns, mass transit systems, and recreational opportunities.
20 • Longitudinal Personal Activity Patterns
21 The average subject in the CHAD time/activity studies provided less than two days of
22 diary data. For this reason, the construction of a year-long activity sequence for each
23 individual requires some combination of repeating data from one subject and using data
24 from multiple subjects. A key issue in this assessment is the development of an approach
25 for creating year-long activity sequences for individuals based on a cross-sectional
26 activity data base that includes 24-hour records. We believe an appropriate approach
27 should adequately account for the day-to-day and week-to-week repetition of activities
28 common to individuals while maintaining realistic variability between individuals.
29 • Averting Behavior
30 Behavior changes in response to PM2.5 pollution or in response to air quality index (AQI)
31 notification ("averting behavior") is not being taken into account in our planned exposure
32 modeling (Mansfield and Corey, 2003; Wen et al., 2009). We do not feel that this is a
33 relatively influential uncertainty at this time.
4-20
-------
1 • Modeling Near-Traffic Outdoor Environments and Public Transportation
2 Modeling activities such as walking next to roads, waiting at bus stops, bicycling, and
3 riding motorcycles, buses, subways and trains is difficult due to the limited information
4 available about these activities. It is also difficult to estimate the ambient concentrations
5 in these environments. As mentioned in the previous section, staff are working on
6 methods to characterize these exposures. An important part of this effort will be to
7 characterize the uncertainties of these methods that we plan to incorporate into the
8 quantitative uncertainty analysis.
9 4.8 UNCERTAINTY AND VARIABILITY
10 The primary difficulty in performing an exposure modeling uncertainty analysis is the
11 quantitative characterization of the uncertainties of the model inputs and model formulation.
12 Information about the variability of model inputs or the variability and uncertainty combined is
13 often available, but it is usually difficult to estimate the uncertainty separately from the
14 variability. In considering the use of APEX for a PM2.5 exposure assessment, EPA has
15 considered the availability of information to provide plausible distributions or ranges for the
16 uncertainties of all of the model inputs. EPA plans to build upon the APEX exposure modeling
17 uncertainty analysis conducted in support of the recent review of the ozone NAAQS (Langstaff,
18 2007). We plan to improve on these distributions of variability and uncertainly, where data are
19 available to do so, and to extend the analysis of model formulation uncertainty.
20 Once estimates of the uncertainty of the model inputs have been developed, we plan to
21 propagate these uncertainties through the model to quantify the resultant uncertainty of the
22 model predictions. The APEX uncertainty analysis methodology incorporates a 2-dimensional
23 Monte Carlo sampling approach that explicitly characterizes and models the variability and
24 uncertainty in inputs and outputs. Essentially, this approach entails performing thousands of
25 model runs with model inputs randomly sampled from specified distributions reflecting
26 uncertainty of the model inputs, while each single APEX run simulates distributions of
27 variability. This 2-dimensional Monte Carlo method allows for the separate characterization of
28 the variability and uncertainty in the model results (Morgan and Henri on, 1990; Cullen and Frey,
29 1999). This approach allows for great flexibility in specifying uncertainty distributions for any
30 of the model inputs and parameters that are supplied to APEX by input files. Furthermore, this
31 allows us to specify conditional distributions and joint distributions between parameters for
4-21
-------
1 which we have data, which can be critically important in modeling uncertainty (Haas, 1997;
2 Haas, 1999; Wu and Tsang, 2004).
3 Uncertainties are inherent in modeled representations of physical reality due to
4 simplifying assumptions and other aspects of model formulation. The methods for assessing
5 input parameter uncertainty and model formulation or structure uncertainty are different. It is
6 difficult to incorporate the uncertainties due to the model formulation into a quantitative
7 assessment of uncertainty in a straightforward manner. The preferred way to assess model
8 formulation uncertainty is by comparing model predictions with measured values, while having
9 fairly complete knowledge of the uncertainty due to input parameters. EPA plans to ascertain
10 whether sufficient data are available to perform such an evaluation. For example, we will
11 consider using the data collected in the Detroit Study (DEARS28) for this purpose. In the
12 absence of measurements that can be used to estimate model uncertainty, our planned approach
13 to assessing model formulation uncertainty will be to partition this uncertainty into that of the
14 components, or sub-models, of APEX. For each of the sub-models within APEX, we plan to
15 discuss the simplifying assumptions and those uncertainties associated with the sub-models
16 which are distinct from the input data uncertainties. Where possible, we plan to evaluate these
17 sub-models by comparing their predictions with measured data. Alternatively, we may formulate
18 an informed judgment as to a range of plausible uncertainties for the sub-models. We plan to
19 quantitatively assemble the different types of uncertainties and variability to present an
20 integrated analysis of uncertainty and variability.
See http://www.epa.gov/dears
4-22
-------
i 5 SCHEDULE AND MILESTONES
2 Table 5-1 lists the key milestones for the Risk and Exposure Assessment (REA) that are
3 planned as part of the current PM NAAQS review. Consultation with the CAS AC PM Panel is
4 scheduled for April 1-2, 2009 to obtain review of the first draft Integrated Science Assessment
5 (ISA) and to obtain input on the plans to conduct quantitative assessments. EPA staff will then
6 proceed to develop health risk estimates associated with recent PM ambient concentrations and
7 levels representing just meeting the current PM standards and exposure estimates associated with
8 historical ambient PM2.5 concentrations. These estimates and the methodologies used will be
9 presented in the first draft PM REA. CASAC and public comments on this plan will be taken
10 into consideration in the development of the first draft REA, the preparation of which will
11 coincide and draw from the second draft ISA. The first draft report is scheduled to be released
12 for CASAC and public review in August 2009. EPA will receive comments on this draft
13 document from the CASAC and the general public at a meeting planned for September 2009.
14 The second draft REA will draw on the final ISA and will reflect consideration of CASAC and
15 public comments on the first draft REA. The second draft REA will include assessments for just
16 meeting potential alternative standards. We plan to release the second draft REA in March 2010
17 for review by CASAC and the general public at a meeting that is planned for April 2010. Staff
18 will consider these review comments and prepare a final REA, currently planned to be completed
19 in July 2010. The final REA will reflect consideration of CASAC and public comments on the
20 second draft REA. The final ISA and final REA will inform the policy assessment and
21 rulemaking steps that will lead to a final decision of the PM NAAQS. Our current schedule
22 includes plans for issuing a proposed rule in January 2011 and a final rule in October 2011.
5-1
-------
1
2
Table 5-1 Key Milestones for the Risk and Exposure Assessment
Milestone
Release first draft PM ISA
Release draft PM REA Scope and Methods Plans
CASAC/public review and meeting on first draft PM ISA
CASAC consultation on draft PM REA Scope and Methods Plans
Release second draft PM ISA
Release first draft of the PM REA
CASAC/public review and meeting on second draft PM ISA and first
draft REA
Final PM ISA
Release second draft of the PM REA
CASAC/public review and meeting on second draft of the PM REA
Final PM REA
Date
December 2008
February 2009
April 1, 2009
April 2, 2009
July 2009
August 2009
September 2009
December 2009
March 20 10
April 2010
July 2010
5-2
-------
6 REFERENCES
2 Abt Associates Inc. (1996). "A Particulate Matter Risk Assessment for Philadelphia and Los
3 Angeles." Bethesda, MD. Prepared for the Office of Air Quality Planning and Standards,
4 U.S. Environmental Protection Agency, Contract No. 68-W4-0029. July 3 (revised
5 November). Available: http://www.epa.gov/ttn/naaqs/standards/pm/s_pm_pr_td.html.
6 Abt Associates Inc. (2002). Proposed Methodology for Particulate Matter Risk Analyses for
7 Selected Urban Areas: Draft Report. Bethesda, MD. Prepared for the Office of Air Quality
8 Planning and Standards, U.S. Environmental Protection Agency, Contract No. 68-D-03-002.
9 Available: http://www.epa.gov/ttn/naaqs/standards/pm/s_pm_cr_td.html.
10 Abt Associates (2005). Particulate Matter Health Risk Assessment for Selected Urban Areas.
11 Final Report. Bethesda, MD. Prepared for the Office of Air Quality Planning and
12 Standards, U.S. Environmental Protection Agency, Contract No. 68-D-03-002. EPA 452/R-
13 05-007A. Available: http://www.epa.gov/ttn/naaqs/standards/pm/s_pm_cr_td.html.
14 Abt Associates (2008). Environmental Benefits Mapping and Analysis Program (Version 3.0).
15 Bethesda, MD. Prepared for Environmental Protection Agency, Office of Air Quality
16 Planning and Standards, Air Benefits and Cost Group. Research Triangle Park, NC.
17 AHS (2003). U.S. Census Bureau, Current Housing Reports, Series 11170/03-5, American
18 Housing Survey for the Detroit Metropolitan Area: 2003.
19 Akland, G. G., Hartwell, T. D., Johnson, T. R., Whitmore, R. W. (1985). Measuring human
20 exposure to carbon monoxide in Washington, D. C. and Denver, Colorado during the winter
21 of 1982-83. Environ Sci Technol. 19:911-918.
22 Babin SM; Burkom HS; Holtry RS; Tabernero NR; Stokes LD; Davies-Cole JO; DeHaan K; Lee
23 DH. (2007). Pediatric patient asthma-related emergency department visits and admissions in
24 Washington, DC, from 2001-2004, and associations with air quality, socio-economic status
25 and age group. Environ Health 6: 9.
26 Bell ML; Ebisu K; Peng RD; Walker J; Samet JM; Zeger SL; Dominic F. (2008). Seasonal and
27 regional short-term effects of fine particles on hospital admissions in 202 U.S. counties,
28 1999-2005. Am JEpidemiol 168(11): 1301-1310.
29 Biller, W., Feagans, T., Johnson, T., Duggan, G., Paul, R., McCurdy, T. and Thomas, H. (1981).
30 A general model for estimating exposure associated with alternative NAAQS. 74th
31 annual meeting of the APCA, Philadeplphia, PA.
32 Burke, J and Vedantham, R. (2005). SHEDS-PM: Stochastic Human Exposure and Dose
33 Simulation for Particulate Matter User Guide.
34 Burke, J. pers. comm., 3/21/08. Updated SHEDS-AirToxics parameters for the New Haven
35 simulations.
36 Burke, J., Zufall, M., Ozkaynak, H. (2001). A population exposure model for particulate matter:
37 Case study results forPM2.s in Philadelphia, PA. J Expos Anal Environ Epidemiol. 11, 470-
38 489.
6-1
-------
1 Byun D; Schere KL. (2006). Review of the Governing Equations, Computational Algorithms,
2 and Other Components of the Models-3 Community Multiscale Air Quality (CMAQ)
3 Modeling System. Applied Mechanics Reviews 59: 51.
4 Byun DW; Ching JKS. (1999). Science Algorithms of the EPA Models-3 Community Multiscale
5 Air Quality (CMAQ) Modeling System: US Environmental Protection Agency, Office of
6 Research and Development.
7 Chen, Y.; Yang, Q.; Krewski, D.; Burnett, R.T.; Shi, Y.; McGrail, K. (2005) The effect of coarse
8 ambient particulate matter on first, second, and overall hospital admissions for respiratory
9 disease among the elderly. Inh. Toxicol. 17: 649-655.
10 Chimonas MA; Gessner BD. (2007). Airborne particulate matter from primarily geologic, non-
11 industrial sources at levels below National Ambient Air Quality Standards is associated with
12 outpatient visits for asthma and quick-relief medication prescriptions among children less
13 than 20 years old enrolled in Medicaid in Anchorage, Alaska. Environ Res 103: 397-404.
14 Chock, D. P.; Winkler, S.; Chen, C. (2000). A study of the association between daily mortality
15 and ambient air pollutant concentrations in Pittsburgh, Pennsylvania. J. Air Waste Manage.
16 Assoc. 50:1481-1500.
17 Cullen, A.C. and Frey, H.C. (1999). Probabilistic Techniques in Exposure Assessment: A
18 Handbook for Dealing with Variability and Uncertainty in Models and Inputs. Springer.
19 Dockery DW; Pope CA, 3rd; Xu X; Spengler JD; Ware JH; Fay ME; Ferris BG, Jr.; Speizer FE.
20 (1993). An association between air pollution and mortality in six U.S. cities. The New
21 England journal of medicine 329: 1753-1759.
22 Dominici F; Peng RD; Bell ML; Pham L; McDermott A; Zeger SL; Samet JMCEH. (2006). Fine
23 particulate air pollution and hospital admission for cardiovascular and respiratory diseases.
24 JAMA 295: 1127-1134.
25 Eftim SE; Samet JM; Janes H; McDermott A; Dominici F. (2008). Fine particulate matter and
26 mortality: a comparison of the six cities and American Cancer Society cohorts with a
27 medicare cohort. Epidemiology 19: 209-216.
28 Enstrom, J. E. (2005). Fine particulate air pollution and total mortality among elderly
29 Californians, 1873-2002. Inh. Tox. 17:803-816.
30 Environment Canada. (2008). National Air Pollution Surveillance (NAPS) Network.
31 Environmental Technology Advancement Directorate, Environmental Protection Service,
32 Environment Canada. Available : http://www.etc-cte.ec.gc.ca/NAPS/index_e.html
33 Fairley, D. (1999). Daily mortality and air pollution in Santa Clara County, 1989-1996. EHP.
34 107:637-641.
35 Fairley, D. (2003). Mortality and air pollution for Santa Clara County, California, 1989-1996. In:
36 Revised analyses of time-series studies of air pollution and health. Special report. Boston,
37 MA: Health Effects Institute; pp. 97-106. Available:
38 http://www.healtheffects.org/Pubs/TimeSeries.pdf. October 18, 2004.
6-2
-------
1 Fiore, A.M., DJ. Jacob, H. Liu, R.M. Yantosca, T.D. Fairlie, and Q.Li (2003). Variability in
2 surface ozone background over the United States: Implications for air quality policy.
3 Journal of Geophysical Research. Vol. 108 (D24), 4787.
4 Franklin M; Zeka A; Schwartz J. (2007). Association between PM2.5 and all-cause and specific-
5 cause mortality in 27 US communities. Journal of Exposure Science and Environmental
6 Epidemiology 17: 279-287.
7 Franklin M; Koutrakis P; Schwartz P. (2008). The role of particle composition on the association
8 between PM2.5 and mortality. Epidemiology 19: 680-689.
9 Georgopoulos, P. and Lioy, P. (2006). From a theoretical framework of human exposure and
10 dose assessment to computational system implementation: the Modeling ENvironment for
11 TOtal Risk Studies (MENTOR). J Toxicol Environ Health B Crit Rev. 9(6):457-83.
12 Georgopoulos, P., Wang, S., Vyas, V., Sun, Q., Burke, J., Vedantham, R., McCurdy, T.,
13 Ozkaynak, H. (2005) A source-to-dose assessment of population exposures to fine PM and
14 ozone in Philadelphia, PA, during a summer 1999 episode. J Expos Anal Environ
15 Epidemiol. 15, 439-457.
16 Glen, G., Smith, L., Isaacs, K., McCurdy, T. and Langstaff, J. (2008). A new method of
17 longitudinal diary assembly for human exposure modeling. Journal of Exposure Science
18 and Environmental Epidemiology 18, 299-311.
19 Goss CH; Newsom SA; Schildcrout JS; Sheppard L; Kaufman JD. (2004). Effect of ambient air
20 pollution on pulmonary exacerbations and lung function in cystic fibrosis. American journal
21 of respiratory and critical care medicine 169: 816-821 Graham, S. E. and McCurdy, T.
22 (2004). Developing meaningful cohorts for human exposure models. J Expo Anal Environ
23 Epidemiol. 14: 23-43.
24 Haas, C. (1997). Importance of Distributional Form in Characterizing Inputs to Monte Carlo
25 Risk Assessments. Risk Analysis., 17:107-113.
26 Haas, C. (1999). On Modeling Correlated Random Variables in Risk Assessment. Risk
27 Analysis, 19:1205-1214.
28 Hartwell, T. D., Clayton, C. A., Ritchie, R. M., Whitmore, R. W., Zelon, H. S., Jones, S. M.,
29 Whitehurst, D. A. (1984). Study of Carbon Monoxide Exposure of Residents of
30 Washington, DC and Denver, Colorado. Research Triangle Park, NC: U.S. Environmental
31 Protection Agency, Office of Research and Development, Environmental Monitoring
32 Systems Laboratory. EPA-600/4-84-031.
33 Hofferth, S., Davis-Kean, P.E., Davis, J., & Finkelstein, J. (1999). The Child Development
34 Supplement to the Panel Study of Income Dynamics: 1997 User Guide. Available at:
35 http://psidonline.isr.umich.edu/CDS/usergd.html
36 Hopke. P. (2002). ). Letter from Dr. Phil Hopke, Chair, Clean Air Scientific Advisory
37 Committee (CASAC) to Honorable Christine Todd Whitman, Administrator, U.S. EPA.
38 Final advisory review report by the CASAC Particulate Matter Review Panel on the
39 proposed particulate matter risk assessment. May 23, 2002. Available:
40 http://www.epa.gov/sab/pdf/casacadv02002.pdf
6-3
-------
1 IMPROVE (2008). Data from the Interagency Monitoring of Protected Visual Environments,
2 available at: http://vista.cira.colostate.edu/improve/default.htm
3 Isaacs, K., Glen, G., McCurdy, T. and Smith, L. (2008). Modeling energy expenditure and
4 oxygen consumption in human exposure models: accounting for fatigue and EPOC. Journal
5 of Exposure Science and Environmental Epidemiology 18, 289-298.
6 Ito, K. (2003). Associations of particulate matter components with daily mortality and morbidity
7 in Detroit, Michigan. In: Revised analyses of time-series studies of air pollution and
8 health. Special report. Boston, MA: Health Effects Institute; pp. 143-156. Available:
9 http://www.healtheffects.org/Pubs/TimeSeries.pdf. May 12, 2004.
10 Ito K; Thurston GD; Silverman RA. (2007). Characterization of PM2.5, gaseous pollutants, and
11 meteorological interactions in the context of time-series health effects models. Journal of
12 Exposure Science and Environmental Epidemiology 17 Suppl 2: S45-60.
13 Jerrett, M.; Burnett, R. T.; Ma, R.; Pope, C. A, III; Krewski, D. ; Newbold, K. B.; Thurston, G.;
14 Shi, Y.; Finkelstein, M.; Calle, E. E.; Thun, M. J. (2005) Spatial analysis of air pollution
15 and mortality in Los Angeles. Epidemiology 16(6):727-736.
16 Johnson, T. (1989). Human Activity Patterns in Cincinnati, Ohio. Palo Alto, CA: Electric
17 Power Research Institute. EPRIEN-6204.
18 Klemm, R.J.; Mason, R.M.; Heilig, C.M.; Neas, L.M.; and Dockery, D.W. (2000). Daily
19 mortality associated specifically with fine particles? Data reconstruction and replication of
20 analysis. J Air Waste Manage Assoc. 50: 1215-1222.
21 Klemm, R. J.; Mason, R. (2003). Replication of reanalysis of Harvard Six-City mortality study.
22 In: Revised analyses of time-series studies of air pollution and health. Special report.
23 Boston, MA: Health Effects Institute; pp. 165-172. Available:
24 http://www.healtheffects.org/Pubs/TimeSeries.pdf. May 12, 2004.
25 Klepeis, N. E., Tsang, A. M., Behar, J. V. (1996). Analysis of the National Human Activity
26 Pattern Survey (NHAPS) Respondents from a Standpoint of Exposure Assessment.
27 Washington, DC: U.S. Environmental Protection Agency, Office of Research and
28 Development. EPA/600/R-96/074.
29 Krewski, D.; Burnett, R. T.; Goldberg, M. S.; Hoover, K.; Siemiatycki, J,; Jerrett, M.;
30 Abrahamowicz, M.; White, W. H. (2000). Reanalysis of the Harvard Six Cities Study and
31 the American Cancer Society Study of parti culate air pollution and mortality. A special
32 report of the Institute's particle epidemiology reanalysis project. Cambridge, MA: Health
33 Effects Institute.
34 Laden, F.; Schwartz, J.; Speizer, F.E.; Dockery, D.W. (2006). Reduction in fine particulate air
35 pollution and mortality: extended follow-up of the Harvard Six Cities Study. Am. J. Respir.
36 Crit. Care. Med. 173: 667-672.
37 Langstaff, J. (2004). OAQPS Staff Memorandum to PM NAAQS Review Docket (OAR-2001-
38 0017). Subject: A Methodology for Incorporating Short-term Variable Background
39 Concentrations in Risk Assessment. December 17, 2004. Available at:
40 http://www.epa.gov/ttn/naaqs/standards/pm/s_pm_cr_td.html.
6-4
-------
1 Langstaff, J. (2005). OAQPS Staff Memorandum to PM NAAQS Review Docket (OAR-2001-
2 0017). Subject: Estimation of Policy-Relevant Background Concentrations of Parti culate
3 Matter. January 27, 2005. Available at:
4 http://www.epa.gov/ttn/naaqs/standards/pm/s_pm_cr_td.html
5 Langstaff, J. E. (2007). OAQPS Staff Memorandum to Ozone NAAQS Review Docket (EPA-
6 HQ-OAR-2005-0172). Subject: Analysis of Uncertainty in Ozone Population Exposure
7 Modeling. [January 31,2007]. Available at:
8 http://www.epa.gov/ttn/naaqs/standards/ozone/s_o3_cr_td.html
9 Lipfert F; Wyzga R; Baty J; Miller J. (2006a). Traffic density as a surrogate measure of
10 environmental exposures in studies of air pollution health effects: Long-term mortality in a
11 cohort of US veterans. Atmospheric Environment 40: 154-169.
12 Lipfert FW; Baty JD; Miller JP; Wyzga RE. (2006b). PM2.5 constituents and related air quality
13 variables as predictors of survival in a cohort of U.S. military veterans. Inhal Toxicol 18:
14 645-657.
15 Lippmann M; Ito K; Nadas A; Burnett RT. (2000). Association of parti culate matter components
16 with daily mortality and morbidity in urban populations. Res Rep Health Eff Inst: 5-72,
17 discussion 73-82.
18 Liu, L., Box, M., Kalman, D., Kaufman, J., Koenig, J., Larson, T., Lumley, T., Sheppard, L.,
19 Wallace, L. (2003). Exposure assessment of paniculate matter for susceptible populations
20 in Seattle. Environ Health Perspect. 111(7):909-18.
21 Long, C.M., Suh, H.H., Catalano, P.J., Koutrakis, P. (2001). Using Time- and Size-Resolved
22 Particulate Data to Quantify Indoor Penetration and Deposition Behavior, Environmental
23 Science & Technology, 35: 2089-2099
24 Long, C., Suh, H., Koutrakis, P. (2000). Characterization of indoor particle sources using
25 continuous, mass and size monitors, J. Air Waste Manag. Assoc., 50:1236-1250
26 Mainieri, T., Blackburn, Z., Stafford, F., Eccles, J., Schoeni, R., Yeung, W., and McGonagle, K.
27 (2004). The Panel Study of Income Dynamics Child Development Supplement: User
28 Guide for CDS-II. The Institute for Social Research, Ann Arbor, MI. Available at:
29 http://psidonline.isr.umich.edu/CDS
30 Mansfield, C. and Corey, C. (2003). Analysis of Survey Data on Ozone Alert Days, Final
31 Report, September 2003. Prepared for the U.S. Environmental Protection Agency by RTI
32 International, Research Triangle Park, NC.
33 Mar, T.F.; Norris, G.A.; Koenig, J.Q.; Larson, T.V. (2000) Associations between air pollution
34 and mortality in Phoenix, 1995-1997. Env. Health Perspect. 108(4): 347-353.
35 Mar, T. F.; Norris, G. A.; Larson, T. V.; Wilson, W. E.; Koenig, J. Q. (2003). Air pollution and
36 cardiovascular mortality in Phoenix, 1995-1997. In: Revised analyses of time-series studies
37 of air pollution and health. Special report. Boston, MA: Health Effects Institute; pp. 177-
38 182. Available: http://www.healtheffects.org/Pubs/TimeSeries.pdf October 18, 2004.
39 McCurdy, T. (1994). "Human exposure to ambient ozone," pp. 85-127 in: D.J. McKee (ed.)
40 Tropospheric Ozone. Ann Arbor MI: Lewis Publishers.
6-5
-------
1 McCurdy, T. (1995). Estimating Human Exposure to Selected Motor Vehicle Pollutants Using
2 the NEM Series of Models: Lessons To Be Learned. J. Expo. Anal, and Env. Epidemiol.
3 5(4):533-550.
4 McCurdy, T. (2000). Conceptual basis for multi-route intake dose modeling using an energy
5 expenditure approach. JExpo Anal Environ Epidemiol. 10: 1-12.
6 McCurdy, T., Glen, G., Smith, L., Lakkadi, Y. (2000). The National Exposure Research
7 Laboratory's Consolidated Human Activity Database. J Expo Anal Environ Epidemiol. 10:
8 566-578.
9 McDonnell WF; Nishino-Ishikawa N; Petersen FF; Chen LH; Abbey DE. (2000). Relationships
10 of mortality with the fine and coarse fractions of long-term ambient PMio concentrations in
11 nonsmokers. Journal of exposure analysis and environmental epidemiology 10: 427-436.
12 Miller KA; Siscovick DS; Sheppard L; Shepherd K; Sullivan JH; Anderson GL; Kaufman JD.
13 (2007). Long-term exposure to air pollution and incidence of cardiovascular events in
14 women. The New England journal of medicine 356: 447-458.
15 Moolgavkar, S.H. (2000a). Air pollution and daily mortality in three US counties. Env. Health
16 Perspect. 108: 777-784.
17 Moolgavkar, S.H. (2003). Air pollution and daily deaths and hospital admissions in Los Angeles
18 and Cook Counties. In: Revised analyses of time-series studies of air pollution and health.
19 Special report. Boston, MA: Health Effects Institute; pp. 183-198. Available:
20 http://www.healtheffects.org/Pubs/TimeSeries.pdf. October 18, 2004.
21 Morgan and Henri on (1990). Uncertainty: A Guide to Dealing with Uncertainty in Qualitative
22 Risk and Policy Analysis. Cambridge University Press.
23 Ostro B; Broadwin R; Green S; Feng WY; Lipsett M. (2006). Fine particulate air pollution and
24 mortality in nine California counties: results from CALFINE. Environ Health Perspect 114:
25 29-33.
26 Ostro B; Feng WY; Broadwin R; Green S; Lipsett M. (2007). The effects of components of fine
27 particulate air pollution on mortality in Clifornia: results from CALFINE. Environ Health
28 Perspect 115: 13-19.
29 Ozkaynak H., Frey H.C., Burke, J., and Finder, R.W. (2009). Analysis of coupled model
30 uncertainties in source-to-dose modeling of human exposures to ambient air pollution: a
31 PM2.5 case-study. Atmospheric Environment, Vol. 43, pp. 1641-1649.
32 Peel JL; Tolbert PE; Klein M; Metzger KB; Flanders WD; Todd K; Mulholland JA; Ryan PB;
33 Frumkin H. (2005). Ambient air pollution and respiratory emergency department visits.
34 Epidemiology 16: 164-174.
35 Peng RD; Chang HH; Bell ML; McDermott A; Zeger SL; Samet JM; Dominici F. (2008). Coarse
36 particulate matter air pollution and hospital admissions for cardiovascular and respiratory
37 diseases among Medicare patients. JAMA 299: 2172-2179.
6-6
-------
1 Persily, A. and Gorfain, J. (2004). Analysis of Ventilation Data from the U.S. Environmental
2 Protection Agency Building Assessment Survey and Evaluation (BASE) Study. National
3 Institute of Standards and Technology, NISTIR 7145, December 2004.
4 Persily, A., J. Gorfain, G. Brunner (2005). Ventilation Design and Performance in U.S. Office
5 Buildings. ASHRAE Journal. April 2005, 30-35.
6 Pope CA, 3rd; Thun MJ; Namboodiri MM; Dockery DW; Evans JS; Speizer FE; Heath CW, Jr.
7 (1995). Parti culate air pollution as a predictor of mortality in a prospective study of U.S.
8 adults. American journal of respiratory and critical care medicine 151: 669-674.
9 Pope, C. A., Ill; Burnett, R. T.; Thun, M. J.; Calle, E. E.; Krewski, D.; Ito, K.; Thurston, G. D.
10 (2002). Lung cancer, cardiopulmonary mortality, and long-term exposure to fine parti culate
11 air pollution. J. Am. Med. Assoc. 287:1132-1141.
12 Pope CA, 3rd; Burnett RT; Thurston GD; Thun MJ; Calle EE; Krewski D; Godleski JJ. (2004).
13 Cardiovascular mortality and long-term exposure to parti culate air pollution:
14 epidemiological evidence of general pathophysiological pathways of disease. Circulation
15 109: 71-77.
16 Richmond H., Palma, T, Langstaff, J., McCurdy, T., Glenn, G., Smith, L. (2002). Further
17 Refinements and Testing of APEX (3.0): EPA's population exposure model for criteria and
18 air toxic inhalation exposures. Joint meeting of the International Society of Exposure
19 Analysis and International Society of Environmental Epidemiology, Vancouver, Canada.
20 August 11-15, 2002.
21 Rizzo, M. (2008). Memo from Michael Rizzo to the Nitrogen Dioxide NAAQS Review Docket
22 (EPA-HQ-OAR-2006-0922). Investigation of How Distributions of Hourly NO2
23 Concentrations Have Changed Over Time in Six Cities. November 20, 2008. Available:
24 http://www.epa.gov/ttn/naaqs/standards/nox/s_nox_cr_rea.html.
25 Robinson, J. P., Wiley, J. A., Piazza, T., Garrett, K., and Cirksena, K. (1989). Activity Patterns
26 of California Residents and their Implications for Potential Exposure to Pollution.
27 California Air Resources Board, Sacramento, CA. CARB-A6-177-33.
28 Roman, H.A., Walker, K.D., Gettleman, T, Conner, L., Richmond, H.M., Hubbell, B.J., Kinney,
29 P.L. (2008). An Expert Judgment Assessment of the Mortality Impact of Changes in
30 Ambient Fine Parti culate Matter in the U.S. Industrial Economics, Inc., Cambridge, MA;
31 U.S. Environmental Protection Agency, Office of Air Quality Planning and Standards;
32 Columbia University Mailman School of Public Health.
33
34 Schwartz, J. (2003). Daily deaths associated with air pollution in six US cities and short-term
35 mortality displacement in Boston. In: Revised analyses of time-series studies of air pollution
36 and health. Special report. Boston, MA: Health Effects Institute; pp. 219-226. Available:
37 http://www.healtheffects.org/Pubs/TimeSeries.pdf. October 18, 2004.
38 Schwartz, J.; Dockery, D. W.; Neas, L. M. (1996). Is daily mortality associated specifically with
39 fine particles? J. Air Waste Manage. Assoc. 46:927-939.
6-7
-------
1 Schwartz J; Neas LM. (2000). Fine particles are more strongly associated than coarse particles
2 with acute respiratory health effects in schoolchildren. Epidemiology 11: 6-10.
3 Sheppard, L.; Levy, D.; Norris, G.; Larson, TV; Koenig, J.Q. (1999). Effects of ambient air
4 pollution on nonelderly asthma hospital admissions in Seattle, Washington, 1987-1994.
5 Epi. 10:23-30.
6 Sheppard, L. (2003). Ambient air pollution and nonelderly asthma hospital admissions in Seattle,
7 Washington, 1987-1994. In: Revised analyses of time-series studies of air pollution and
8 health. Special report. Boston, MA: Health Effects Institute; pp. 227-230. Available:
9 http://www.healtheffects.org/Pubs/TimeSeries.pdf. October 18, 2004.
10 Slaughter JC; Kim E; Sheppard L; Sullivan JH; Larson TV; Claiborn C. (2005). Association
11 between paniculate matter and emergency room visits, hospital admissions and mortality
12 in Spokane, Washington. Journal of exposure analysis and environmental epidemiology
13 15: 153-159.
14 Spier, C. E., Little, D. E., Trim, S. C., Johnson, T. R., Linn, W. S., Hackney, J. D. (1992).
15 Activity patterns in elementary and high school students exposed to oxidant pollution. J
16 Expo Anal Environ Epidemiol. 2: 277-293.
17 Svendsen, E., Yeatts, K., Neas, L., Orton, S., Alexis, N., Creason, J., Williams, R., Peden, D.
18 Circulating neutrophil CD 14 expression counteracts the inverse association of ambient
19 PM2.5 and PM10-2.5 on lung function in asthmatic children. Annals of Allergy, Asthma,
20 and Immunology 99:244-253 (2007)
21 Tsang A. M., and Klepeis, N. E. (1996). Descriptive Statistics Tables from a Detailed Analysis
22 of the National Human Activity Pattern Survey (NHAPS) Data. U.S. Environmental
23 Protection Agency. EPA/600/R-96/148.
24 Turk, B. H., Grimsrud, D. T., Brown, J. T., Geisling-Sobotka, K. L., Harrison, J., Prill, R. J.
25 (1989). Commercial Building Ventilation Rates and Particle Concentrations. ASHRAE No.
26 3248.
27 U.S. Department of Transportation and U.S. Census Bureau (2000). Census of Population and
28 Housing, 2000 long-form (sample) data, Census Transportation Planning Package (CTPP)
29 2000. Available at: http://transtats.bts.gov/
30 U.S. EPA (2001a). Review of the National Ambient Air Quality Standards for Particulate Matter:
31 Policy Assessment of Scientific and Technical Information, Preliminary Draft OAQPS Staff
32 Paper. Research Triangle Park, NC 27711: Office of Air Quality Planning and Standards.
33 June 2001.
34 U.S. Environmental Protection Agency (200Ib). Particulate Matter NAAQS Risk Analysis
35 Scoping Plan, Draft. Research Triangle Park, NC; Office of Air Quality Planning and
36 Standards. Available: http://www.epa.gov/ttn/naaqs/standards/pm/s_pm cr td.html.
37 U.S. EPA (2002). Consolidated Human Activities Database (CHAD) Users Guide. Database
38 and documentation available at: http://www.epa.gov/chadnetl/
6-8
-------
1 U.S. EPA (2004). Air Quality Criteria for Particulate Matter. National Center for Environmental
2 Assessment, Office of Research and Development, U.S. Environmental Protection Agency,
3 Research Triangle Park, NC 27711; report no. EPA/600/P-99/002aF and EP A/600/P-
4 99/002bF. October 2004. Available:
5 http ://www. epa.gov/ttn/naaqs/standards/pm/s_pm_cr_cd.html
6 U.S. EPA (2005). Review of the National Ambient Air Quality Standards for Particulate Matter:
7 Policy Assessment of Scientific and Technical Information, OAQPS Staff Paper. Research
8 Triangle Park, NC 27711: Office of Air Quality Planning and Standards; report no. EPA
9 EPA-452/R-05-005a. December 2005. Available:
10 http://www.epa.gov/ttn/naaqs/standards/pm/s_pm_cr_sp.html
11 U.S. EPA (2007). Ozone Population Exposure Analysis for Selected Urban Areas. Office of Air
12 Quality Planning and Standards, U.S. Environmental Protection Agency, Research Triangle
13 Park,NC. EPA-452/R-07-010. July 2007. Available:
14 http://www.epa.gov/ttn/naaqs/standards/ozone/s_o3_cr_td.html.
15 U. S. EPA (2008a). Integrated Review Plan for the National Ambient Air Quality Standards fro
16 Particulate Matter. National Center for Environmental Assessment and Office of Air
17 Quality Planning and Standards, U.S. Environmental Protection Agency, Research Triangle
18 Park, NC. Report No. EPA 452/R-08-004. March 2008. Available at:
19 http://www.epa.gov/ttn/naaqs/standards/pm/data/2008_03_fmal_integrated_review_plan.pdf
20 U.S. EPA (2008b). Integrated Science Assessment for Particulate Matter: First External Review
21 Draft. National Center for Environmental Assessment-RTF Division, Office of Air Quality
22 Planning and Standards, Research Triangle Park, NC. EPA/600/R-08/139 and 139A.
23 December 2008. Available:
24 http://www.epa.gov/ttn/naaqs/standards/pm/s_pm_2007_isa.html.
25 U.S. EPA (2008c). Risk and Exposure Assessment to Support the Review of the NO2 Primary
26 National Ambient Air Quality Standard. Office of Air Quality Planning and Standards, U.S.
27 Environmental Protection Agency, Research Triangle Park, NC. EPA-452/R-08-008a and b.
28 November 2008. Available:
29 http://www.epa.gov/ttn/naaqs/standards/nox/s_nox_cr_rea.html.
30 U.S. EPA(2008d). National Air Quality: Status and Trends Through 2007. Office of Air
31 Quality Planning and Standards, U.S. Environmental Protection Agency, Research Triangle
32 Park,NC. EPA-454/R-08-006. November 2008. Available:
33 http://www.epa.gov/airtrends/2008/index.html.
34 U.S. EPA (2008e). EPA's Report on the Environment. U.S. Environmental Protection Agency
35 Washington, DC EPA/600/R-07/045F. May 2008. Available:
36 http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=l 90806
37 U.S. EPA (2008e). Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
38 Documentation (TRIM.Expo / APEX, Version 4) Volume I: User's Guide. Office of Air
39 Quality Planning and Standards, U.S. Environmental Protection Agency, Research Triangle
40 Park, NC. EPA-452/B-08-001a. Available: http://www.epa.gov/ttn/fera/human_apex.html.
6-9
-------
1 U. S. EPA (2008f). Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model
2 Documentation (TRIM.Expo / APEX, Version 4) Volume II: Technical Support Document.
3 Office of Air Quality Planning and Standards, U.S. Environmental Protection Agency,
4 Research Triangle Park, NC. EPA-452/B-08-001b. Available:
5 http ://www. epa.gov/ttn/fera/human_apex.html.
6 U. S. EPA (2009). Particulate Matter National Ambient Air Quality Standards: Scope and
7 Methods Plan for Urban Visibility Impact Assessment. Office of Air Quality Planning and
8 Standards, U.S. Environmental Protection Agency, Research Triangle Park, NC. Available:
9 http://www.epa.gov/ttn/naaqs/standards/pm/s_pm_2007_pd.html.
10 Villeneuve, P.J., M.S. Goldberg, D. Krewski, R.T. Burnett, Y. Chen (2002). Fine particulate air
11 pollution and all-cause mortality within the Harvard Six-Cities Study: variations in risk by
12 period of exposure. Ann Epidemol 12(8): 568-576.
13 Wallace, L. and Williams, R. Use of personal-indoor-outdoor sulfur concentrations to estimate
14 the infiltration factor, outdoor exposure factor, penetration coefficient, and deposition rate
15 for individual homes. Environmental Science and Technology, 39, 1707-1714 (2005)
16 Wallace, L., Williams, R., Rea, A., Croghan, C. (2006). Continuous weeklong measurements of
17 personal exposures and indoor concentrations of fine particles for 37 health-impaired
18 North Carolina residents for up to four seasons. Atmospheric Environment 40:399-414.
19 Wallace, L., Williams, R., Suggs, J. Validation of a method for estimating long-term exposures
20 based on short-term measurements. Risk Analysis, 25, 687-694 (2005)
21 Wallace, L., Williams, R., Suggs, J., Jones, P. Estimating contributions of outdoor fine particles
22 to indoor concentrations and personal exposures. EPA 600/R-06/023. (2006)
23 Wen X., Balluz L., and Mokdad A. (2009). "Association between media alerts of Air Quality
24 Index and change of outdoor activity among adult asthma in six states, BRFSS, 2005." J.
25 Comm. Health 34: 40-46.
26 Weschler, C., (2003). Indoor chemistry as a source of particles. In: Morawska, L., Salthammer,
27 T. (Eds.), Indoor Environment: Airborne Particles and Settled Dust. Wiley-VCH, Weinheim,
28 pp. 167-189.
29 Wiley, J. A., Robinson, J. P., Piazza, T., Garrett, K., Cirksena, K., Cheng, Y.-T., Martin, G.
30 (1991a). Activity Patterns of California Residents: Final Report. California Air Resources
31 Board, Sacramento, CA. ARB/R93/487. Available from: NTIS, Springfield, VA., PB94-
32 108719.
33 Wiley, J. A., Robinson, J. P., Cheng, Y.-T., Piazza, T., Stork, L., Pladsen, K. (1991b). Study of
34 Children's Activity Patterns: Final Report. California Air Resources Board, Sacramento, CA.
35 ARB-R-93/489.
36 Williams, R. (2005). EPA's Detroit Exposure and Aerosol Research Study. EPA Research
37 Highlights, AWMA Environmental Manager, October 2005, pp43.
38 Williams, R., Suggs, J., Creason, J., Rodes, C., Lawless, P., Kwok, R., Zweidinger, R., Sheldon,
39 L. (2000). The 1998 Baltimore particulate matter epidemiology-exposure study: Part 2.
6-10
-------
1 Personal exposure associated with an elderly population. J Expo Anal Environ Epidemiol.
2 10(6): 533-543.
3 Williams, R., Suggs,!., Rea, A., et al. (2003a). The Research Triangle Park particulate matter
4 panel study: PM mass concentration relationships. Atmos Environ. 37: 5349-5363.
5 Williams, R., Suggs, J., Rea, A., et al. (2003b). The Research Triangle Park particulate panel
6 study: modeling ambient source contribution to personal and residential PM mass
7 concentrations. Atmos Environ 37: 5365-5378.
8 Wilson WE; Mar TF; Koenig JQ. (2007). Influence of exposure error and effect modification by
9 socioeconomic status on the association of acute cardiovascular mortality with particulate
10 matter in Phoenix. Journal of Exposure Science and Environmental Epidemiology 17: SI 1.
11 Wu, F. and Tsang, Y. (2004). Second-order Monte Carlo uncertainty/variability analysis using
12 correlated model parameters: application to salmonid embryo survival risk assessment.
13 Ecological Modeling 177, 393-414.
6-11
-------
i APPENDIX A: PREPARATION OF MODEL INPUTS
2 FOR THE PILOT EXPOSURE ASSESSMENT
3 This appendix describes the APEX model inputs and the data used to create the APEX
4 input files for the pilot exposure assessment. Note that these model inputs are preliminary and
5 ongoing efforts to compile data to better characterize the distributions of parameters used for
6 APEX inputs will be described in the first draft REA. For example, recent air exchange rate
7 (AER) data will be used to update the distributions of AERs.
8 The geographic extent of the modeling region consists of the census tracts in and
9 overlapping the Detroit domain for recent photochemical model simulations on a 1 km grid.
10 This domain is bounded by longitudes from -83.8410 to -82.7875 degrees and latitudes from
11 41.8039 to 42.8552 degrees, and has 72 columns and 108 rows. Figure A- 1 shows this domain
12 as well as the locations of the PM2.5 ambient monitors in this area in 2002. The population,
13 commuting, activity, and meteorological data used are described in section 4.6 above.
14 The microenvironments that we plan to model are those listed in Table 4-1 in Chapter 4.
15 There are some particle size dependent parameters used in the mass-balance model. We plan to
16 model PM2.5 in four size ranges to be able to capture size dependencies. The size categories of
17 PM2.5 modeled in this simulation are 0-0.03, 0.03-0.1, 0.1-1.0 and 1.0-2.5 urn. It is likely that we
18 will revise these ranges after additional studies are reviewed. Although exposures will be
19 modeled for these size categories, we plan to combine them in order to obtain PM2.5 exposures
20 and do not plan to report the size categories separately. If the estimated exposures are not
21 sensitive to this refinement in this pilot application, we plan to reconsider the appropriateness of
22 modeling separate size classes in the additional cities analyzed.
23 The microenvironment parameters that must be specified for the mass-balance model
24 include air exchange rates, penetration factors, and decay and deposition rates for indoor areas.
25 The parameters for the factors model calculations include proximity factors for outdoor
26 microenvironments and vehicles and penetration values for vehicles.
27
A-l
-------
2
3
4
5
6
7
yv
Flushing3 Flint •^J <
(jeriese^es Davis
Swartz Creek Burton°
rand Grand-Blanc
Fenton Holly
ston gp ^
\^/^^
Lapeer \
^ LaPe6^ gLU
St. Clair Port Huron ft/
— Marysville
Oxford — ^P DJ^Km^\nH Saint Clair
WAIUIU Romeo Kicnmona
H^ke Orion Macomb .c ry
r CCv, D ,..-** Marine City
Oakland ^>i.f><''b^> New Baltimore—, /
uaKiana Rochester r ^> < ^ . A ^, Z/ J
^^. , < uy --^) ipaint oia IT; heaven y*
m L^ike^ 7 ^r^J^^r^y^ Algonac
merce /\> /'
Noyi^jiD.etroit-Metro-Ai'ea ^ /
viN^ K /
N J -fc§-''
• «-^*^
MAfeyne
V CP r-~$ A
Romulus t^C-J
-------
1 residences with central or room air conditioning in Detroit is estimated by the 2003 American
2 Housing Survey to be 0.81 (AHS, 2003).
3
4
Table A-l. Lognormal Distribution Parameters for Residential Air Exchange Rates (hr"1)
5
6
9
10
11
12
13
14
15
16
17
18
19
20
Temp (F)
<50
50-68
68-77
77-86
>86
<50
50-68
68-77
77-86
>86
A/C
Yes
Yes
Yes
Yes
Yes
No
No
No
No
No
Geometric
mean
0.71
1.14
1.14
1.24
1.24
1.02
0.79
1.61
1.61
1.61
Geometric
standard
deviation
2.02
2.68
2.68
2.18
2.18
2.14
2.04
2.12
2.12
2.12
minimum = 0.1, maximum =10
The deposition values used in this analysis are presented in Table A- 2 . These were
estimated using data from Long et al. (2001), who estimated average decay rates based on their
measurements of 31 homes (Long et al. 2000). The data from the study were remapped from
smaller size intervals onto the four ranges selected for this simulation. The less than 0.03 bin is
represented by the 0.02-0.03 Long et al. bin; the 0.03-0.10 bin is the sum of four Long et al. bins:
0.03-0.04, 0.04-0.06, 0.06-0.08, and 0.08-1.0. The 0.1-1.0 bin is the sum of the six Long et al.
bins: 0.1-0.15, 0.15-0.2, 0.2-0.3, 0.3-0.4, 0.4-0.5, and 0.7-1.0. The 1.0-2.5 bin is the sum of the
1-2 and 2-3 Long et al. bins. Uncertainties were converted from standard errors to standard
deviations and added in quadrature. The Long et al. (2001) study differentiates summer and
winter because the houses included in the study were located in the Northeast U.S. and residents
typically opened windows during the summer months for air ventilation. We define the summer
months as the six warmest months of the year, May through October.
A-3
-------
1
2
Table A- 2 Normal Distribution Parameters for Depositional/Decay Rates (hr
4
5
6
7
10
11
12
13
14
15
16
Size (urn)
<0.03
0.03 -0.1
0.1-1.0
1.0-2.5
<0.03
0.03-0.1
0.1-1.0
1.0-2.5
Season
Summer
Summer
Summer
Summer
Winter
Winter
Winter
Winter
Mean
0.59
0.35
0.20
0.72
0.13
0.11
0.22
0.43
Standard
deviation
0.84
0.30
0.23
1.63
0.64
0.27
0.26
0.41
minimum = 0, maximum = 5
Penetration rates for residences (Table A- 3) are also estimated using data presented by
Long et al. (2001), who measured penetration rates of 41 homes. The data from the study were
remapped from smaller size intervals onto the four ranges selected for this simulation. The effect
of this remapping is reflected in the larger standard deviations. As described above, the study
divided the year into summer and winter months.
Table A- 3 Normal Distribution Parameters for Penetration Rates
Size (um)
<0.03
0.03-0.1
0.1-1.0
1.0-2.5
<0.03
0.03-0.1
0.1-1.0
1.0-2.5
Season
Summer
Summer
Summer
Summer
Winter
Winter
Winter
Winter
Mean
0.90
0.98
0.94
0.63
0.55
0.66
0.75
0.57
Standard
deviation
0.89
0.38
0.32
1.28
0.57
0.29
0.26
0.32
minimum = 0, maximum = 1
Non-residential Indoor Microenvironments
Air exchange rates for non-residential microenvironments are specified as lognormal
distributions with geometric mean of 1.109 hr"1, and geometric standard deviation of 3.015.
Minimum and maximum bounds of 0.07 to 13.8 are used, with resampling if a sample is outside
of these bounds. This distribution is described in (U.S. EPA, 2007) and is based on two studies:
A-4
-------
1 the Building Assessment Survey and Evaluation (BASE) Study (Persily and Gorfain 2004;
2 Persily et al. 2005) and a study presented in Turk et al. (1989).
3 Outdoor Microenvironments and Vehicles
4 The parameters for modeling these microenvironments are currently being developed.
5 For this analysis, we are planning to use as preliminary distributions the values used in an
6 exposure modeling study of New Haven, CT conducted by ORD using SHEDS-PM (pers.
7 comm., Janet Burke, 3/21/08). The proximity factors for outdoors near-road, public garages and
8 parking lot, and all in-vehicle microenvironments are assigned lognormal distributions with
9 geometric mean 3.320 and geometric standard deviation 0.902 on weekdays, and geometric
10 mean 2.075 and geometric standard deviation 1.498 on weekends. These factors are intended to
11 account for the typically higher ambient concentrations in these areas than the concentrations at
12 monitoring locations. If we use near-road dispersion modeling in estimating the spatial
13 concentration fields, then we will use proximity factors centered around 1.0. Penetration factors
14 for vehicles are assigned normal distributions with mean 0.3 and standard deviation 0.232. For
15 this APEX application, we plan to assign minimum and maximum values of 0.2 and 20 for the
16 proximity factors, and 0.1 and 1.0 for penetration factors, and resample if these bounds are
17 exceeded. We incorporate bounds on the distributions so that unrealistic values will not be
18 generated.
19 Ambient PM2.s Concentrations
20 There were six hourly and 12 daily PM2.5 monitors in and near the greater Detroit
21 metropolitan area in 2002. Three of the hourly monitors are in Canada and three are in
22 Michigan; the daily monitors are all in Michigan (for these counts collocated monitors are
23 counted as single monitors). The monitors in Canada are in the NAPS network, and the
24 Michigan monitors are all from AQS. Table A- 4 lists the averaging time, period, number of
25 days with measurements, and the monitoring method for each monitor during the period July 1-
26 31, 2002. Collocated monitors are grouped within rows. Figure A- 1 shows the locations of
27 these monitors using the map location codes in Table A- 4, except for location B, which is to the
28 west in Lansing.
A-5
-------
Table A- 4 PM2.5 Monitors in Detroit Region, July 1-31, 2002
4
5
6
Monitor id 1
2604900211
2604900213
2606500122
2609900091
2611500051
2611500055
2612500011
2614700051
2614700052
2616100051
2616100081
2616100082
2616100083
2616300011
2616300012
2616300013
2616300015
2616300151
2616300161
2616300191
2616300251
2616300331
2616300335
2616300361
3909500241
3909500251
3909500261
3909500265
NAPS060204
NAPS060211
NAPS061004
Averaging
time (hours)
24
1
24
24
24
24
24
24
24
24
24
24
1
24
24
1
24
24
24
24
24
24
24
24
24
24
24
24
1
1
1
Period
(days)
3
6
o
J
o
J
6
o
J
o
J
6
o
J
3
6
1
6
3
3
1
3
3
o
J
6
o
J
1
o
J
1
6
-
-
-
#Days
with data
11
31
6
10
10
6
8
10
5
10
10
3
31
19
3
31
10
8
28
11
11
11
6
9
31
11
30
6
31
31
31
Monitor
method2
118
701
118
118
118
810
118
118
118
118
118
118
701
118
118
701
810
118
118
118
118
118
810
118
120
120
120
810
cont
cont
cont
Map
location
A
B
C
D
E
F
G
H
I
J
K
L
M
N
0
P
Q
R
S
T
U
Collocated monitors are grouped within rows
2 Method 118: FRM - R & P MODEL 2025 PM2.5 SEQUENTIAL w/WINS / Gravimetric
Method 120: FRM - ANDERSEN RAAS2.5-300 PM2.5 SEQ w/WINS / Gravimetric
Method 701: TEOM Gravimetric 50° C with PM25 SCC and no correction factor.
Method 810: SASS / Gravimetric
We plan to estimate missing air quality data from hourly monitors using the following
procedure: where there are consecutive strings of missing values (data gaps) of less than six
A-6
-------
1 hours, missing values will be estimated by linear interpolation between the valid values at the
2 ends of the gap and any remaining missing values at a monitor will be estimated by fitting linear
3 regression models for each hour of the day with each of the other monitors, and for each hour of
4 the day choosing the model which maximizes R2, subject to the constraints that R2 must be
5 greater than 0.5 and the number of regression data values is at least 50 (the maximum number of
6 values possible is 365 when one year is modeled). If there are any remaining missing values at
7 this point, for gaps of less than nine hours, we plan to estimate missing values by linear
8 interpolation between the valid values at the ends of the gap. Any remaining missing values
9 after that would be replaced with the region-wide mean concentration for that hour. We plan to
10 continue to investigate methods for estimating missing data from daily monitors.
11 One of the major difficulties to be resolved is how to estimate the hourly spatially-
12 varying PM2.5 air quality concentration fields, which are inputs to the APEX exposure model.
13 PM2.5 air quality monitors are spatially sparse and often 24-hour average measurements are only
14 reported every three or six days. We plan to evaluate the use of air quality models to augment
15 the monitored ambient PM2.5 concentrations to better estimate ambient concentrations at
16 locations between monitors and for days with no measurements, as discussed in section 4.6.
A-7
-------
3
4
5
United States
Environmental Protection
Agency
Office of Air Quality Planning and Standards
Health and Environmental Impacts Division
Research Triangle Park, NC
Publication No. EPA-452/P-09-002
February 2009
------- |