*
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
OFFICE OF THE ADMINISTRATOR
SCIENCE ADVISORY BOARD
August 22, 2008
EPA-CASAC-08-019
The Honorable Stephen L. Johnson
Administrator
U.S. Environmental Protection Agency
1200 Pennsylvania Avenue, N.W.
Washington, D.C. 20460
Subject: Clean Air Scientific Advisory Committee's (CAS AC) Peer Review of EPA's
Risk and Exposure Assessment to Support the Review of the SO 2 Primary
National Ambient Air Quality Standards (First Draft, July 2008)
Dear Administrator Johnson:
The Clean Air Scientific Advisory Committee (CASAC), augmented by subject-matter-
experts to form the CASAC Sulfur Oxides Primary NAAQS Review Panel conducted its review
of EPA's Risk and Exposure Assessment to Support the Review of the SO 2 Primary National
Ambient Air Quality Standards: First Draft (REA) on July 30-31, 2008. The REA was prepared
by the EPA Office of Air Quality Planning and Standards (OAQPS) staff as part of EPA's
ongoing review of the primary national ambient air quality standards (NAAQS) for sulfur
dioxide (802). The CASAC held a subsequent teleconference on August 12, 2008, to discuss its
draft advisory letter. This letter presents CASAC's advice on the first draft REA.
The CASAC found the draft provides an appropriate framework for a risk and exposure
assessment. However, it is incomplete and lacks certain details as discussed in the Panel's
answers to the Agency charge questions that follow. In addition, a major concern is that the
benchmark values being considered for SO2 effects are based on the benchmark values (0.4 to
0.6 ppm) suggested by the Integrated Science Assessment (ISA) for Sulfur Oxides - Health
Criteria (Second External Review Draft, May 2008) to which CASAC raised objections in our
letter to you dated August 8, 2008. The CASAC believes strongly that the weight of clinical and
epidemiology evidence indicates there are detectable clinically relevant health effects in sensitive
subpopulations down to a level at least as low as 0.2 ppm SO2. These sensitive subpopulations
represent a substantial segment of the at-risk population. The benchmark values should be
adjusted downward in both the ISA and the REA documents. A second major concern is that the
overall approach to risk characterization remains to be specified.
-------
CASAC's response to EPA's charge questions are summarized below. Individual
recommendations from CASAC Panel members to strengthen the final REA are appended in
Enclosure B.
Air Quality Information and Analyses:
Charge Question 1: We have evaluated SO2 air quality throughout the United States, using all
available 5-minute and 1-hour ambient monitoring data for years 1997 through 2007. To what
extent are the air quality characterizations and analyses technically sound, clearly
communicated, appropriately characterized, and relevant to the review of the primary SO2
NAAQS?
Chapter 6, at present, provides a good overview of existing air quality data, with
particular focus on the limited 5-minute average data and how to extend the more widely
available 1-hr average observations. At present, the chapter is often difficult to follow, and the
need for specific analyses is not always readily apparent. In the beginning, the chapter should
provide a road map for what is done and why, and the material in the chapter should be better
focused. In regard to the analyses, EPA should consider describing the observations using log
normal distributions. Several figures need to be clarified as recommended in the individual
comments.
Charge Question 2: To what extent are the properties of ambient SO2 appropriately
characterized, including ambient levels, spatial and temporal patterns, relationships between
various averaging times, and the relationship between ambient SO2 and human exposure?
EPA staff uses the limited data available for 5-min peak concentrations by developing a
model of 5-min peak concentration to the related 1-hr average concentration, then applies that
model to the more widely available 1-hr data to generate estimated peak concentrations. At
present, the approach chosen, in part, uses calculated coefficients of variation (COVs) to classify
monitors, and develops cumulative density functions (CDFs) for the peak-to-mean ratios from
which to apply Monte Carlo simulations. EPA should assess whether more general parametric
relationships (possibly assuming log-normal distributions) between the 5-min and 1-hr average
concentrations can be derived, and thus minimize the occurrence or extent of differences
between the predicted and observed numbers of exceedances. For those cases where very large
differences occur, more careful analysis is needed. In addition to addressing outliers at the high
end of the distribution, more attention to the distribution at lower levels is warranted, particularly
considering the CAS AC's recommendation to consider potential exposures below 0.4 ppm. The
current analysis is confusing because of changes in the monitoring network that affect raw
temporal trends and inference about the relationship of ambient concentration to human exposure.
EPA should develop analytical approaches that facilitate intended interpretations, such as
presentation of concentration trends not biased by network modifications.
Charge Question 3: Twenty locations were selected for detailed analyses, using ambient SO2
monitoring data for years 2002-2006. What are the views of the panel regarding the
appropriateness of these locations, the time period of analysis, and the approach used to select
them?
11
-------
While the choice of the first three locations is very solid, the REA needs to better justify
its choice of the remaining 17 locations for detailed analysis. Restricting the analysis to counties
where 3 or more monitors exist may exclude locations where peak concentrations are most likely.
Although the 3-monitors-per-county criterion is reasonable, other criteria are equally justified
(e.g., assessing where peak concentrations are most likely to occur).
Charge Question 4: In order to simulate just meeting either the current 24-hour or annual
standards, staff adjusted SO 2 air quality levels for the years 2002-2006 upwards in all but one
location. Ambient monitoring data in North Hampton County PA were above the 24-hour
standard in the year 2006 and were therefore adjusted downward. To what extent is the
approach taken technically sound, clearly communicated, and appropriately characterized?
The approach used is reasonable, although it could be more clearly communicated.
While the panel recognizes that this approach assumes, somewhat unrealistically, that all source
emissions would increase similarly, this simple approach is adequate for the purposes of the
REA.
Charge Question 5: What are the views of the Panel regarding the adequacy of the assessment
of uncertainty and variability?
The uncertainty and variability discussion, at present, is mostly qualitative and needs
work. It should be more quantitative in regards to the development and use of peak-to-mean
ratios (PMRs) with particular attention to the uncertainty in the model's ability to simulate
exceedances over a wider range of levels. Further, the influence of monitor siting (e.g., terrain,
source location, magnitude of emissions) and "upset" emission conditions on measured
concentrations and distributions needs to be more thoroughly considered. The discussion of
ambient monitor to exposure representation recognizes the many limitations of how observations
may represent actual exposures, but should be more definitive as to the degree of uncertainty and
possible bias that is introduced. With regard to the assessment of uncertainty introduced by the
statistical model (Section 6.5.7), the REA should be explicit as to the use of duplicate or
independent data, and should consider the use of data-withholding approaches.
Exposure Analysis:
Charge Question 1: To what extent is the assessment, interpretation, and presentation of the
initial results of the exposure analysis technically sound, clearly communicated, and
appropriately characterized?
The presentation in Chapters 2 and 7 should be revised to better frame the exposure
analysis. The committee suggests the text describe how the different components of the
exposure assessment fit together, with perhaps a schematic diagram displaying the steps of
calculation. The two intended applications of the exposure predictions should be discussed along
* The REA should be explicit in stating that the analysis was limited to the primary standards and did not consider
the three-hour secondary SO2 NAAQS.
ill
-------
with an overview of the features of the simulated exposure data that are most important for each
particular application. There is currently no discussion of the health risk assessment application
of the exposure modeling or the key features of the exposure predictions that will influence the
risk assessment.
While the exposure analysis was generally technically sound, there were several specific
concerns. The estimation of the additional eleven 5-minute concentrations in an hour forces all
the other values near the hourly average. This artificially reduces the variability of the 5-minute
data and effectively constrains the predicted number of 5-minute exceedances in any given hour
to be one or less. Furthermore, given the inertia effect of the indoor air diluting the peak levels,
the temporal correlation between outdoor 5-minute levels may be critical to correctly calculating
the distribution of 5-minute average levels indoors. In addition, the assumed removal rate
distributions may strongly affect conclusions since they are represented as uniform and with high
rates. The data underlying these removal rate distributions must be fully described, in addition to
adoption of reasonable methodology to derive distributional inputs for this factor for use in
current modeling approaches such as APEX. Because the indicated ranges will lead to very large
reductions in expected 862 indoor concentrations, this is a key issue for the modeling of indoor
exposures and it is likely to be a large source of uncertainty.
Charge Question 2: The draft risk and exposure assessment evaluates exposures in selected
locations encompassing a variety of SO 2 emission source types in the state of Missouri; these
areas were chosen as an initial case study since 1) air quality measurements indicated numerous
exceedances of 5-minute benchmark values, 2) there are multiple stationary source emissions
above 1,000 tons per year, and 3) there are numerous ambient monitors measuring 5-minute and
1-hour SO2 concentrations. The second draft may also evaluate exposures in the remainder of
Missouri and also include areas of Pennsylvania, West Virginia, and other locations with large
SO2 emission sources. What are the views of the panel regarding the appropriateness of these
proposed additional locations and on the approach used to select them?
The initial case study location is a reasonable choice. The committee had diverse
opinions regarding inclusion of additional locations. Since the current results are derived from a
single scenario based on a set of assumptions and local conditions, the relative value of varying
the local conditions vs. assessing the sensitivity to other model input parameters and distributions
should be evaluated. Additional sensitivity analyses in the current location could be an
alternative and the additional effort in these analyses may be lower; however this should not be
the only consideration. Towards selecting additional locations, an assessment of the national
diversity of 862 concentrations as driven by multiple sources (i.e. in areas with overlapping
plumes) may help with the location decision. Several committee members were interested in
seeing an urban area in the northeast or an area near Midwest power plants included.
The text should reflect that the scenarios investigated are for a small subset of the US
population near relatively large SC>2 sources and state that the numbers produced should be
interpreted accordingly.
Charge Question 3: Do Panel members have comments on the appropriateness and/or
relevance of the populations evaluated in the exposure assessment?
IV
-------
The populations were viewed to be generally appropriate and relevant. The most highly
exposed populations will live near sources; it will be helpful to quantify the size of these
populations since many sources are in rural areas. With respect to the table of age-specific
asthma prevalence rates, the committee observed discontinuities and a few rates that were higher
for females than males; these data should be checked. There is no provision for medication use
in the modeling of the asthmatic population. Since the proportion of the asthmatic population
that is routinely medicated could have implications for the health risk assessment, it may be
worth addressing this point in the document.
Charge Question 4: To what extent are the approaches taken to model SO2 emission sources
technically sound and clearly communicated?
AERMOD is the appropriate concentration prediction tool if emissions-based modeling is
determined to be the best option for obtaining the spatio-temporal concentration distribution.
The model depends on a source inventory that may be incomplete and also omits many small and
far away sources. This potential downward bias should be evaluated and an adjustment
approach provided. Furthermore, there should be evaluation of whether upset conditions (high,
short term emissions due to unusual events) from nearby sources are affecting predictions.
Other alternatives to capturing the concentration distribution should also be considered.
The key consideration for assessing the quality of predictions for the purpose of this
analysis is whether the predictions are comparable to real-world concentrations in level and
variability. This analysis needs to capture the entire distribution (for the health risk assessment)
as well as the peaks (i.e. the tails of the distribution, for the exposure risk assessment).
Additional work is needed to assess the quality of the predictions.
Charge Question 5: Human exposures were modeled using APEX to simulate the movement of
individuals through different microenvironments. Do panelists have comments on the
microenvironments modeled?
APEX is an appropriate tool for the exposure model, and the approach represents current
best practice. Generally the micro-environments and parameters chosen seem appropriate.
Given the need to capture 5-minute peaks, the approaches to predicting 5-minute outdoor
concentrations and to modeling infiltration are critical; these should be reassessed and in addition
they should be given high priority for evaluation in sensitivity analyses. Description of the
simulated exposure data should identify the times and locations where most of the peak
exposures occur. If the higher exposures are occurring indoors, the assumptions for the air
exchange rate and air conditioning prevalence may be particularly influential.
Given the continued reliance on APEX, EPA should ensure there is further evaluation
and improvement of this model across a range of conditions and pollutants. Ultimately, as one
component of this effort, human activity and representative population databases should be
updated and expanded, including consideration of any unique features of activity patterns
associated with asthmatics.
-------
Health Risks:
Charge Question 1: What are the views of the Panel on the overall characterization of the health
risk evidence for SO 2?
The draft REA draws on the ISA in selecting the outcomes and exposure-response
relationships to be used. The principal reliance on the clinical studies of persons with asthma is
appropriate for the quantitative analysis. There is a clear documentation of a causal association,
and the exposure-response relationship has been characterized with reasonable certainty. The
Panel also recommends that the Agency give consideration to the size and susceptibility of the
subpopulations of persons with asthma susceptible to SO2 at various concentrations. This
sensitivity analysis should be carried out in a framework based around the context set by the
epidemiological studies.
Charge Question 2: The characterization of health risks focuses on potential health benchmark
values identified from the experimental SO 2 human exposure literature on lung function with
accompanying respiratory symptoms. What are the views of the Panel on using potential health
benchmarks from this literature to characterize health risks?
The Panel concurs with the use of the clinical studies to derive benchmarks for
characterizing the health risks of SC>2 exposure. In using these studies, the Agency needs to
acknowledge the highly selective nature of the volunteers included in these studies, who may not
adequately represent the full range of sensitivity present in susceptible populations. Additionally,
these studies addressed 862 alone and hence do not replicate the general circumstances of
exposure to ambient SC>2, which is a component of a complex mixture having particulate and
gaseous components that may influence dose.
Charge Question 3: Do Panel members have comments on the range of potential health effects
bench mark values chosen to characterize risks associated with 5-minute SO2 exposures?
The Panel strongly concurs that the range of values to be considered should be extended
lower than the proposed range of 0.4 - 0.6 ppm. The range of exposures emphasized in the ISA
(0.4 - 0.6 ppm) has clearly carried over to the thinking used in the REA. Adoption of this range
might leave substantial numbers of exercising mild asthmatics at considerable risk. Studies
(such as those listed in Table 5-1 of the ISA) have shown that 5 - 20% of mild to moderate
asthmatics experience moderate or greater decrements in lung function at 862 concentrations as
low as 0.2 - 0.3 ppm. For ethical reasons severe asthmatics were not part of these clinical
studies, but it is not unreasonable to presume that they would have responded to even a greater
degree, although routine use of medication among this group could influence response. In
addition, the epidemiological evidence shows emergency room visits and hospitalizations for
respiratory illnesses associated with 24-hour SC>2 levels below the current standard (0.14 ppm
averaged over at 24-hour period). Collectively, this evidence should lead to a conclusion that 0.2
ppm or even a lower level of short-term exposure is an appropriate lower bound value for EPA's
benchmark analysis.
VI
-------
Charge Question 4: To what extent is the assessment, interpretation, and presentation of initial
risk characterization results technically sound, clearly communicated, and appropriately
characterized?
The assessment, interpretation, and presentation of initial risk characterization results are
technically sound, clearly communicated, and reasonable for a first draft. Attention has been
directed at potentially susceptible subgroups. Additionally the basis for selecting the counties for
the substantive characterization for benchmark health risks for 5-minute peak SO2 exposure
should be clarified. In particular, consideration of the representativeness of the locations needs
to be included along with development of the approach that will be used to extend the results of
the risk characterization nationally. Are there enough urban sites included? In characterizing
risks, the Agency should give consideration to the possibility that SO2 not only has effects of
clinical significance for individuals but also has population-level effects that may be relevant to
public health. Additionally, the Panel recommends an exploration of the sensitivity of the risk
characterization to the model used for the concentration-response relationship. A variety of
forms for this relationship can be assumed and different statistical models might be used to
estimate the form of this relationship. Several members elaborate these possibilities in their
comments (e.g. see comments of Drs. Hattis and Sheppard).
The discussion of uncertainty and variability remains completely generic. At this point,
while there is extensive discussion of these matters with regard to exposure, and a probabilistic
approach is described for addressing uncertainty in health estimates, the overall approach in the
risk characterization remains to be specified.
Charge Question 5: The epidemiology literature will be used to qualitatively characterize SO 2-
related health risks for health outcomes such as respiratory symptoms and emergency
department visits and hospital admissions for respiratory-related causes. However, staff has
judged that it is not appropriate to use the available SO 2 epidemiological studies as the basis for
a quantitative risk assessment in this review. Do panel members have comments on this
judgment and/or on the rationale presented to support it?
The Panel members had a range of views on use of the epidemiological data, ranging
from considering the epidemiological evidence as qualitatively indicative of clinical morbidity
resulting from the physiological responses documented in the clinical studies, on the one hand, to
proposing that the epidemiological evidence should be used to develop a quantitative risk model,
on the other hand. The Panel acknowledges that time constraints and resources may limit the
approaches that can be taken by EPA for risk estimation. Regardless, the Panel members were
not certain as to the nature of the qualitative risk characterization that will be performed. The
epidemiological studies cited in the ISA address SO2 as a component of a complex mixture and
under exposure circumstances in which it may be a surrogate for other components of the air
pollution mixture.
vn
-------
The CAS AC Sulfur Oxide Panel was pleased to review the first draft of the REA and
provide advice early in the development of this important document. We look forward to
receiving the Agency's response and reviewing the second draft on December 17-18, 2008.
Sincerely,
Dr. Rbgene Henderson
Chair
Clean Air Scientific Advisory Committee
Enclosure A: Roster of CASAC Sulfur Oxides Primary NAAQS Review Panel
Enclosure B: Compilation of Individual Panel Member Comments on EPA's Risk and Exposure
Assessment to Support the Review of the SO2 Primary National Ambient Air
Quality Standards, July 2008)
Vlll
-------
Enclosure A: U.S. Environmental Protection Agency
Clean Air Scientific Advisory Committee (CASAC)
Sulfur Oxides Primary NAAQS Review Panel
CASAC MEMBERS
Dr. Rogene Henderson (Chair), Scientist Emeritus, Lovelace Respiratory
Research Institute, Albuquerque, NM
Dr. Ellis B. Cowling, Emeritus Professor,, Colleges of Natural Resources and
Agriculture and Life Sciences, North Carolina State University, Raleigh, NC
Dr. James Crapo,* Professor of Medicine, Department of Medicine , National
Jewish Medical and Research Center, Denver, CO
Dr. Douglas Crawford-Brown, Professor and Director, Department of
Environmental Sciences and Engineering, Carolina Environmental Program,
University of North Carolina at Chapel Hill, Chapel Hill, NC
Dr. Donna Kenski, Director of Data Analysis, Lake Michigan Air Directors
Consortium, Des Plaines, IL
Dr. Armistead (Ted) Russell, Professor, Department of Civil and Environmental
Engineering , Georgia Institute of Technology, Atlanta, GA
Dr. Jonathan M. Samet, Professor and Chair of the Department of Epidemiology,
Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD
PANEL MEMBERS
Mr. Ed Avol, Professor, Preventive Medicine, Keck School of Medicine,
University of Southern California, Los Angeles, CA
Dr. John R. Balmes, Professor, Department of Medicine, Division of
Occupational and Environmental Medicine, University of California, San
Francisco, CA
Dr. Terry Gordon, Professor, Environmental Medicine, NYU School of
Medicine, Tuxedo, NY
Did not participate in the July 30-31, 2008 review of the first draft REA.
-------
Dr. Dale Hattis, Research Professor, Center for Technology, Environment, and
Development, George Perkins Marsh Institute, Clark University, Worcester, MA
Dr. Patrick Kinney, Associate Professor, Department of Environmental Health
Sciences, Mailman School of Public Health , Columbia University, New York,
NY
Dr. Steven Kleeberger, Professor, Lab Chief, Laboratory of Respiratory Biology,
National Institute of Environmental Health Sciences, National Institutes of Health,
Research Triangle Park, NC
Dr. Timothy V. Larson, Professor, Department of Civil and Environmental
Engineering, University of Washington, Seattle, WA
Dr. Kent Pinkerton, Professor, Regents of the University of California, Center
for Health and the Environment, University of California, Davis, CA
Dr. Edward Postlethwait, Professor and Chair, Department of Environmental
Health Sciences, School of Public Health, University of Alabama at Birmingham,
Birmingham, AL
Dr. Richard Schlesinger, Associate Dean, Department of Biology, Dyson
College, Pace University, New York, NY
Dr. Christian Seigneur, Vice President, Atmospheric & Environmental Research,
Inc., San Ramon, CA
Dr. Elizabeth A. (Lianne) Sheppard, Research Professor, Biostatistics and
Environmental & Occupational Health Sciences, Public Health and Community
Medicine, University of Washington, Seattle, WA
Dr. Frank Speizer, Edward Kass Professor of Medicine, Channing Laboratory,
Harvard Medical School, Boston, MA
Dr. George Thurston, Professor, Environmental Medicine, NYU School of
Medicine, New York University, Tuxedo, NY
Dr. James Ultman, Professor, Chemical Engineering, Bioengineering Program,
Pennsylvania State University, University Park, PA
11
-------
Dr. Ronald Wyzga, Technical Executive, Air Quality Health and Risk, Electric
Power Research Institute, Palo Alto, CA
SCIENCE ADVISORY BOARD STAFF
Dr. Holly Stallworth, Designated Federal Officer, Science Advisory Board Staff
Office, Washington, D.C.
in
-------
Enclosure B: Preliminary Individual Comments on the Risk and Exposure
Assessment to Support the Review of the SO2 Primary National Ambient Air
Quality Standards (First Draft) from the Clean Air Scientific Advisory
Committee (CASAC) Sulfur Oxides Primary National Ambient Air Quality
Standards (NAAQS) Review Panel
Mr. EdAvol 2
Dr. John Balmes 6
Dr. Douglas Crawford-Brown 9
Dr. Terry Gordon 13
Dr. Dale Hattis 15
Dr. Donna Kenski 28
Dr. Patrick Kinney 32
Dr. Steven Kleeburger 33
Dr. Timothy Larson 34
Dr. Kent E. Pinkerton 37
Dr. Jonathan Samet 39
Dr. Richard Schlesinger 41
Dr. Christian Seigneur 42
Dr. Lianne Sheppard 44
Dr. Frank Speizer 51
Dr. George Thurston 54
Dr. James Ultman 59
Dr. Ronald Wyzga 62
-------
Mr. Ed Avol
Comments on SOx REA - 1st Draft
The compilation of current thinking regarding the performance of a risk and
exposure assessment to support the review of the SCh Primary National Ambient
Air Quality Standard (NAAQS) should be descriptive, informative, logical, and
clear. The first draft document provided for review demonstrates most, but not all,
of these attributes. There is a great deal of information described within the
document (although some of it seemed like unnecessary duplication of the ISA to
me), and the general format is somewhat logical. However, I had trouble at times
with the written clarity of the document. Several sections became fully immersed
in detailed discussions or documentation of procedural steps to operate a model or
perform an analysis, rather than describing the general approach. In my judgment,
the more complete operational details of the model or application should have
been relegated to an appendix or archive, and not be a part of the main text.
In my opinion, the pages and pages of so much detail (in the way of operational
aspects, such as the text provided in Chapter 6) tended to blur the overall flow and
logic of the document.
Moreover, there still appears to be a provincial perspective of accepting US and
Canadian epidemiologic studies for consideration, but relegating other
international studies (regardless of pedigree) as "...supportive evidence..." If the
studies have withstood critical peer review and are published in well-recognized
and respected journals, why should they not be considered equally?
With regard to the exposure perspectives presented in the document and the air
quality analytical decisions (locations, data, approaches, etc), one question might
be whether we are looking forward or backward in thinking about public health
and exposure potential. The selection process for station data usage and exposure
orientation is almost wholly guided by proximity to or downwind trajectory from
power generation plants. In general, this is probably appropriate, given the
currently understood sources and source strengths. However, these are typically
located in rural areas away from major metropolitan areas and populations. It
might be insightful to also consider other sources — for example, seaport
operations, where bunker fuels containing tens of thousands of ppm sulfur are
routinely emitted, or rail, where fuels can contain hundreds or thousands of ppm
sulfur, or large cumulative concentrations of traffic emissions (on-road vehicle
sources individually may be quite low in sulfur emissions, but collectively may be
a substantial line or area source). These activities and exposures tend to be in
more urban and heavily populated areas.
Characterization of Health Risks
(1) (comments on the overall characterization of health evidence for SO2)
-------
I generally found the presentation and characterization of the health evidence to
be reasonable and appropriate. However, I am a little concerned that the
threshold for consideration of data for the risk assessment may be unduly high or
restrictive. While I agree with the general conclusion that the data regarding SO2
exposure for endpoints other than respiratory morbidity are not sufficient to infer
a causal relationship, one wonders if the combined weight of multiple "not quites"
or "almosts" should somehow count for something in the aggregate risk
assessment. The controlled-exposure chamber data clearly support consideration
of observable and clinically significant health outcomes at or below 0.2ppm SO2,
and the epi data, in toto, paint a picture of clinically relevant detrimental health
outcomes in a appreciable portion of the sensitive population.
(2) (use of clinical exposure data on SO2 to characterize health risks)
The use of clinical exposure data on SO2 to characterize health risks seems
appropriate. Admittedly, the population studied in clinical research is small,
somewhat self-selected, and generally biased towards increased
interest/motivation in health and reduced severity of existing disease. Regardless,
the observational data available from these numerous investigations are
invaluable in establishing the potential for actual manifestation of specific health
outcomes. The observed effects in these respiratory-compromised individuals
legitimately raises the potential that other more individuals in the general
population, with more severely compromised respiratory function who would not
or could not participate in controlled chamber studies, are likely to be at risk to
low-level (less than or equal to 0.2 ppm) and short-term (minutes to an hour)
exposures to SO2.
(3) (comments on the range of potential health effects benchmark values chosen
to characterize risks associated with 5-min SO2 exposures)
A number of controlled-exposure (clinical chamber) studies from the 1980s
(primarily from US researchers at UCSF, Rancho Los Amigos Medical Center in
Los Angeles, and the USEPA in Chapel Hill) demonstrated and confirmed the
almost-immediate bronchoconstrictive effects of inhaled SO2 at levels in the 0.4-
0.6ppm range, and also documents some responders down to and below 0.2 ppm.
Subsequent clinical studies in the ensuing decades, though fewer in number and
scope, have generally re-confirmed or extended these findings. Thus, the
underlying evidence for a proposed range of health effects benchmark values
seems available, corroboratory, and sufficient to support the consideration of a
range of values for risks, in the 0.2 ppm vicinity, associated with 5-min SO2
exposures.
The more difficult issue to assess is the health implication of such short-term
exposures, since many of the observed effects seemed to have declined, partially
reversed, or resolved within a half hour or so of initial clinical exposure, even in
the face of continuing exposures. Epidemiologic information regarding a range of
health outcomes (including asthma-related symptoms, ED visits, and
hospitalizations) as the result of short-term (5min to 24hr) exposures are less
-------
consistent and convincing, but in the aggregate, do suggest an excess toll on
human respiratory health. Therefore, it does appear that based on the available
evidence, there is a susceptible and vulnerable population of people at risk from
short-term exposure to SO2.
(4) (judgment on assessment, interpretation, and presentation of risk
characterization results)
The assessment and presentation of the risk characterization results, based on
ambient air quality and various permutations of peaks, peak-to-means, and other
indices of exposure, seems extensive. In all of the presented detail, however, the
clarity and summary points of interpretation are lost or rarely made. The Chapter
6 presentation is detailed and extensive, with page after page of plots and tables,
but what is missing is a clear and succinct summary of what has been established
by virtue of all the presentation by the end of the chapter.
Since the evolution of this format of ISA and REA is still in its infancy, it might
be worth considering a slightly different format for presentation of the individual
sections. The summary conclusion of each section could be stated in underline or
bold format at the outset of each chapter section, and then the supporting material
for the stated claim could be provided. This would have the advantage of clearly
showing and stating the point of the ensuing presentation, discussion, or data.
Alternatively, there needs to be additional effort made in the document to clearly
state the summary conclusions in an accessible manner for document users.
(5) (comments on staff determination that SO2 epi data is not appropriate for
quantitative risk assessment)
The staff recommendation that the available epi data is not sufficient to make a
quantitative risk assessment is based, in part, on the determination that"... staff
recommends primarily relying on US studies." (line 21, pi67). The basis for this
decision (to primarily use US studies only) is one that merits additional
consideration, scrutiny, and potential reversal. Well-designed and executed
studies are not limited to (or necessarily a boundary condition of) US-based
studies. Unadjusted confounding variables and confounding exposures, lack of
complete and precise study details, and well-constructed and appropriately
performed statistical analyses challenge both American and foreign researchers.
More useful activities would be to (a) identify specific gaps in available
information needed for critical public health decisions, and (b) move aggressively
to provide the necessary funding to obtain that information.
A separate question to be addressed is the relative level of staff comfort with
regard to weight of evidence providable by epi data per se, compared to more
controllable (but more artificial) exposure scenarios such as those utilized in
clinical chamber work or animal toxicology. The trade-offs between real-world
exposures of unrestrained mobile populations and lack of control or more
complete understanding of those exposures have been noted and discussed on
several occasions, but how to effectively and appropriately exploit the full value
-------
of community or population-based studies to assess and protect public health is a
critically important issue that should be explicitly resolved by staff so that the
most appropriate judgments can be reached using the widest possible range of
available, credible, and relevant data.
-------
Dr. John Balmes
Comments on SOx REA - 1st Draft
2. The characterization of health risks focuses on potential health benchmark
values identified from the experimental SCh human exposure literature on lung
function with accompanying respiratory symptoms. What are the views of the
Panel on using potential health benchmarks from this literature to characterize
health risks?
In contrast to my opinion regarding the NOx Risk and Exposure Assessment, I
support the staff decision to use the experimental 862 human exposure literature
on lung response and respiratory symptom responses in subjects with asthma.
This literature is sufficiently extensive to provide the basis for a quantitative risk
assessment. I concur with staff s judgment that while the epidemiological
literature shows relatively consistent associations with asthma outcomes
(respiratory symptoms in children, emergency department (ED) visits and
hospitalizations in children and adults), this literature is not sufficiently robust to
support a quantitative risk assessment. That said, I endorse the staffs plan to use
the data from recent U.S. and Canadian epidemiological studies of 862 and ED
visits/hospitalizations to "qualitatively assess the range of SC>2 air quality levels
that are associated with these endpoints."
Staff has reviewed the relevant controlled human exposure studies and selected
health benchmark exposure values from those studies that reflect the potential for
adverse effects in most asthmatic patients. Symptomatic bronchoconstriction will
occur in a substantial proportion of such individuals when exposed for 5-10
minutes to concentrations of SC>2 between 0.4-0.6 ppm during exercise. However,
it is also evident in the data from several of the controlled human exposure studies
that some individuals responded with symptomatic bronchoconstriction from
short-term exposure to 0.2 ppm. Thus, I recommend that 0.2 ppm also be
considered as a health benchmark exposure value. As noted in the draft REA
document such effects of SC>2 in controlled human exposure studies are coherent
with the associations between ambient SC>2 and asthma outcomes reported in the
epidemiological literature.
Specific Comments
p. 14, linelS should be "their" instead of "there".
p. 16, lines 14-17 The study by Winterton et al. to which this sentence refers
found an association between the homozygous wild-type allele for a common
polymorphism in the promoter region of TNFa (-308 G/A). The homozygous
wild-type would be AA. This sentence should specify the specific polymorphism
studied because there are other polymorphisms for TNFa.
-------
p. 27, line 11 should be "... for boys in a Toronto, ON study (mean 24-
h..."
p. 27, line 12 should be "to these hospitalization studies..."
p. 71, line 3 should be "...less than 1%..."
p. 135, Table 7-7 title should be ".. .children in the Midwestern U.S."
p. 136, Table 7-8 title should be ".. .adults in Missouri"
p. 147, line 4 should be "... dispersion modeled concentrations were..."
pp. 155-157 There is no discussion in this Uncertainty Analysis section
of the uncertainties related to using National Health Interview Survey (NHIS)
data for the prevalence of asthma in children of different ages or Missouri
Department of Health data for the prevalence of asthma in adults from different
regions of the state. For example, NHIS data are representative of the country as
a whole, but do not have sufficient geographic resolution to be used at the state
level. That is why Table 7-7 gives prevalence data for the Midwestern U.S. rather
than Missouri.
p. 169, line 2 should be ".. .or retrieve ...
-------
Dr. Douglas Crawford-Brown
Comments on SOx REA - 1st Draft
These comments focus on Chapters 5 through 8 of the Risk and Exposure
Assessment Draft, referring to earlier chapters only as they are needed.
I compared the conclusions in the early chapters to those in the ISA. The authors
have been faithful to the primary conclusions from that earlier document. The
same health effects, exposure durations (short-term) and sensitive subpopulations
(asthmatics) are considered. It also places the same strengths and limitations, and
hence sources of uncertainty, on personal exposure estimates found in the ISA.
In previous reviews of NAAQS assessments, including the recent one on NOx
which uses similar methodologies, I have approved the proportional roll-up or
roll-down methods. I support, therefore, the use of this method in the current
document. The authors should state, however, any assumptions implicit in this
approach, such as whether regulated sources and non-regulated sources are
equally affected by any change in the NAAQS.
As in the draft of the NOx REA, I agree that the adjustment of the benchmarks
produces the same result mathematically as adjusting the air concentrations. But it
makes no sense scientifically, and is likely to be attacked as such. The savings in
processing time don't appear to me sufficient to justify a method that people will
fail to understand as mathematically equivalent, and will make it appear that the
EPA staff is willing to make calculations based on an assumption that effects
occur at levels below the benchmarks.
I support what is essentially a hazard quotient in the assessment (although the
term is not used, the procedure is identical to one using an HQ calculation). The
one issue I would raise here is that the hazard quotient approach usually has a
margin of safety built in through uncertainty factors, and the current assessment
does not appear to have this margin built in. The next draft should at least make
mention of this issue.
The authors have done a much better job than in the first NOx draft of describing
the relationship between the three approaches examined in the report. They seem
to have learned from the NOx reviews.
I support the use of APEX and CHAD for the purpose of performing the
stochastic calculations for the Chapter 7 analysis. These models contain
assumptions that are routine in EPA assessments and have found application in a
wide range of settings. They have been fully vetted for the kinds of assessments
performed here. There remains, however, the problematic relationship between
ambient levels as measured at monitors and ambient levels at or near the points of
exposure for populations. I realize there is not much that can be done about that
-------
issue, because the monitors are located where they are and can't be changed for
the purposes of this assessment. But I would like to see a slightly better
description of the implications of this problem for overall uncertainty.
As my expertise does not extend to air quality modeling, I can't comment on the
adequacy of AERMOD for these purposes. It is a modeling package that has been
used extensively in past EPA assessments, including the NOx assessment, and so
I will assume here that it has been vetted. But I leave further vetting to other
members of CASAC.
Assuming the air modelling can be performed adequately (and again, I will leave
it to other CASAC members to comment on this in a more informed way), then
the subsequent steps in Chapters 7 and 8 are reasonable. The development of the
longitudinal activity sequences is a sophisticated piece of work, being state-of-
the-science. The stochastic sampling methodology is reasonable and employed
commonly at the EPA. The assumptions going into the sampling are adequately
described. The microenvironments are both the correct ones to model given
current data and well executed in the assessment steps (with a caveat about
whether they correctly model the activities of asthmatics, which I note later in this
review).
I found the characterization of results throughout informative and simple to
follow. They walk the reader through the relevant findings. The one thing that
continues to concern me is that I don't know how the results are to be used in any
policy decision. For example, how many individuals, with how many exceedences,
would count as acceptable or unacceptable in any decisions? I suppose it will be
argued that those are policy concerns, not scientific ones, and that the only job of
the REA to present these numbers. But I still expected to see at least some
mention of this issue rather than leaving it entirely in the hands of policy staff and
administrators.
I found it difficult to follow the uncertainty analyses, or at least to understand the
magnitude and implications of any one source of uncertainty. I expected to see
some statements, even if qualitative, about the uncertainty in the various risk
results (e.g. uncertainty in number of people above a benchmark, percent of
asthmatics experiencing a high exposure day one or more times). This aspect can
be greatly improved.
I end with a comment I have made in other settings of CASAC, including in my
review of similar methodologies for the NOx REA. The modelling performed
here starting with Chapter 7 is impressive and represents state-of-the-science. But
I worry that it may be hide a false sense of confidence in these results, which I
take to be quite uncertain. There are many, many assumptions built into the
assessment. At the moment, I think of the results as a kind of scenario analysis,
and not necessarily an accurate reflection of actual exposures and risks in the US
population. The methods may be pushing the current analytic ability too far.
10
-------
It is for this reason that I believe the results of Chapter 9 will be quite important
once they are produced. I realize the problems with epidemiology studies, but it
seems to me there are equal uncertainties in the exposure assessments in Chapters
7 and 8.1 think of the relationship between the epidemiology and clinical studies
as one between Exact Questions, Approximate Answers, and Approximate
Questions, Exact Answers. By this, I mean that the clinical studies ask a question
(how do people respond when in a clinical setting?) that only approximates the
one we want to ask (how do people respond in the natural setting?), but give a
rather precise answer to that approximate question. Epidemiological studies
address exactly the question we want, but provide only an approximate answer. I
am not sure which approximation I prefer. In the end, perhaps the current results
of this REA and those of the Chapter 9 assessment will need to be used as
bounding answers. We will need to discuss this in more detail at the CASAC
meeting.
Some Specific Comments:
Page 10-1 don't see how the section Scenarios for the Current Assessment
actually specifies scenarios. I was looking form greater detail here.
Chapters 2, 3 and 4 need headings, or at least introductory paragraphs, stating that
these are reviews of the ISA conclusions. The Introduction says they are, but the
reader may not remember that when reading the subsequent chapters..
The conclusions of Chapters 2, 3 and 4 are consistent with the draft ISA. Short
term exposures and morbidity is the only association judged sufficiently strong in
both documents.
0.4-0.6 ppm is identified in clinical trials to result in a substantial fraction of
exercising asthmatics to have significant decrements in lung function, for 5-10
minute peaks. Why not just set the standard in that range, then? What is the
purpose of the rest of the assessment? Is it only to explore the answers within that
range? And if so, do the answers developed really allow us to differentiate the
acceptability of a 0.4 ppm standard from one at 0.6?
I didn't review the part of Chapter 6 associated with air quality monitoring. I
agree that application of PMR values is OK, but I can't comment on the empirical
validity of these. I also am not convinced that the variability distributions used are
valid in the tails of the distributions, which I suspect will affect the results.
On Page 71,1 am not sure what is intended by the analysis of the impact of
reducing the number of monitors. Why was this assessment done? I'm sure there
is a reason, and suspect my inability to see it is related to my lack of
understanding of this area, but some explanation would be good.
11
-------
The uncertainty analysis in Chapter 6 has at least identified the major issues of
uncertainty. Given that it is so qualitative, and doesn't involve any formal
uncertainty analyses, it is hard to understand what the reader is to take away from
this. Why were the results not run several times with at least some changes in
parameter inputs, to at least get a sense of sensitivity? Still, there is probably no
way to do a regular quantitative uncertainty analysis given the complexity of the
calculations.
Chapter 7: APEX is the correct model for exposure. Not convinced it can model
asthmatics well, however, so the assumption seems to be that they behave as the
rest of the population in the CHAD database. Am I correct that this is assumed,
and what are the implications on uncertainty if this assumed? I suspect asthmatics
are less likely to go outdoors and play, especially during bad air quality days.
I applaud the use of decision trees on page 121. This kind of tree helps the reader
understand the process used here. There are many places in the document where a
similar tree would have been useful.
Is the assumption that one individual with N exceedences is the same (in terms of
degree of adversity) as N individuals with 1 exceedence?
I'm not sure what to make of Tables such as 7-14. What would constitute large or
small numbers? What is the criterion for this judgment? Or is the intent just to
provide the numbers and let someone else decide in the policy branch? And how
do we interpret a table which is both number of people and the number of
exceedances per individual? There just seems to me to be too much flexibility in
interpreting these tables.
The sensitivity analyses in Chapter 7 are better than in Chapter 6, although again
this is not a full uncertainty analysis.
In Chapter 8,1 am generally supportive of the approach. However, in the end, one
will still be left with looking at the number of people above a given decrement of
lung function or other metric, and so the logic will be the same as in Chapters 6 or
7. The only difference is that a 0.4-0.6 threshold will be replaced with a threshold
based on level of decrement. I don't see this adds anything.
12
-------
Dr. Terry Gordon
Comments on SOx REA - 1st Draft
Characterization of Health Risks (Chapters 3, 4, 5, 6, 7, 8, 9):
1. What are the views of the Panel on the overall characterization of the health
evidence for SO 2? Is this presentation clear and appropriately balanced?
The characterization of the health evidence was presented in a clear and balanced
approach. The document is improved in style and clarity from the previous
development document and is better in many respects than the first draft of the
NOx REA.
2. The characterization of health risks focuses on potential health benchmark
values identified from the experimental SO 2 human exposure literature on lung
function with accompanying respiratory symptoms. What are the views of the
Panel on using potential health benchmarks from this literature to characterize
health risks?
The choice of these benchmarks is appropriate and the uncertainty factors
surrounding this data base were appropriately described with one possible
exception. It must be noted that for health and ethical reasons, the clinical studies
which form the basis of this assessment did not utilize moderate to severe
asthmatics in the 5-10 minute exposure protocols. Therefore, the severity of
pulmonary function decrements and asthmatic symptoms may be underestimated
for the more severe asthma phenotype. EPA should present information regarding
the relative numbers of mild, moderate, and severe asthmatics that make up the
population of the U.S. and consider how these potentially more susceptible severe
asthmatics may be affected by short term ambient exposure to SO2. Admittedly,
the majority of the clinical studies were conducted in the mid-1980's. The subject
criteria, medications, and disease severity classifications have changed since that
time and, therefore, the uncertainty discussion on how well these subjects
represent today's asthmatic population in the U.S. could be expanded.
3. Do panel members have comments on the range of potential health effects
benchmark values chosen to characterize risks associated with 5-minute SO2
exposures?
The range of benchmark values in the exceedance calculations for exposed
asthmatics utilized 0.4 ppm as the cut-off for health effects and, although this is
out of my area of expertise, it was unclear why this was done in light of the health
risk assessment which, utilizing probabilistic math, goes down to 0.2 ppm.
4. To what extent is the assessment, interpretation, and presentation of initial risk
characterization results technically sound, clearly communicated, and
13
-------
appropriately characterized?
The risk characterization is quite clear and technically sound.
5. The epidemiology literature will be used to qualitatively characterize SO2-
related health risks for health outcomes such as respiratory symptoms and
emergency department visits and hospital admissions for respiratory-related
causes. However, staff has judged that it is not appropriate to use the available
SO2 epidemiological studies as the basis for a quantitative risk assessment in this
review. Do panel members have comments on this judgment and/or on the
rationale presented to support it?
Although the epidemiology studies may not lend themselves to easy quantitative
risk assessment, they are quite important despite the potential confounding by co-
pollutants. In light of the positive findings in children and older adults, staff
should make every effort to seriously consider these epidemiology data in a
quantitative assessment, particularly if the qualitative assessment warrants such a
step.
Minor Comments:
Page 14, line 15 - substitute 'their' for 'there'
Page 16, line 16 - There are many different alleles/polymorphisms for TNF so it
is unclear which 'wild-type allele' is being referred to here (I believe it's the -308
polymorphism).
Page 71, line 3 - change 'thank' to 'than'
Page 161, line 10 - Should 'for' be added after 'model'?
Page 161, lines 20-22 - The refractory period has not been shown to last for a
significant amount of time. Because Sheppard et al (1983) only looked at sulfur
dioxide tolerance up to 90 minutes, a repeat exposure at 10 hours after the first 0.4
ppm exposure, for example, could cause a response. Although not identical
challenges, it has been shown that tolerance to exercise-induced asthma is lost by
4 hours after the primary exercise challenge (Edmunds, 1987).
14
-------
Dr. Dale Hattis
Comments on SOx REA - 1st Draft
Air Quality Information and Analyses (Chapter 6):
1. We have evaluated SCh air quality throughout the United States, using all
available 5-minute and 1-hour ambient monitoring data for years 1997 through
2007. To what extent are the air quality characterizations and analyses
technically sound, clearly communicated, appropriately characterized, and
relevant to the review of the primary SCh NAAQS?
The basic approach of doing a set of empirical distributions of peak to median
ratios based on a large database stratified by coefficient of variation (3 strata) and
average SO2 level (5 strata) is reasonable. The only quibble is whether the
number of strata selected for the two variables is the best choice. This could be
tested by running a parallel analysis or two with greater numbers of strata of each
type and comparing the bias and variability of the predictions vs the observations
of peak levels with the base case analysis provided in the current document.
My major problem with Chapter 6 is its exclusive focus on quantifying
exceedances of the very high health benchmark values (400 ppb and above). As I
illustrated in my comment on the plan for the REA in the previous CASAC SO2
meeting, the problem of SO2 asthma responses is not well summarized by looking
at a few localized sites where there are simultaneously very high concentrations
(from local sources) and members of a sensitive subgroup known to react to those
concentrations by direct clinical observation. In fact the problem needs to be
analyzed as a combination of geographic/temporal variability in exposure levels
combined with interindividual variability in sensitivity. In fact, based on the
lognormal distribution of 1-hour ambient SO2 levels and the distribution of
individual sensitivity thresholds observed by Horstman et al. (1986) my earlier
calculations indicated that only about 22% of the total events causing asthmatics
to endure a 100% increase in specific airway resistance would occur at
concentrations of 400 ppm and above. Therefore it would be more reasonable for
any subsequent version of the REA to include exceedances of at least a few lower
SO2 levels. I have recently updated this analysis to factor in the smaller ED50
and slightly greater interindividual variability indicated by the Linn et al. papers
included in the updated ISA. This revised analysis indicates that only 11% of the
overall population asthma-exacerbation effect could be expected to occur at over
400 ppb. About 50% of the expected effect is likely to occur at concentrations of
160 ppb and below.
As a further step in this analysis I have fit lognormal distributions to the exposure
levels derived in the new REA. It can be seen in Figures 1 and 2 that lognormal
distributions do not fit perfectly to these results—if anything the lognormal fits
tend to underestimate the frequency of very high exposure levels. Despite this,
15
-------
using the lognormal distributions of 5 minute maximum exposures to exercising
asthmatics for current emission sources in Missouri, less than 2% of the overall
asthma exacerbation effect is expected to occur below 400 ppb. About 50% of
the expected to occur at exposure levels of 120 ppb and below. The further
diminished importance of vary high exposure levels results from a lower overall
variability and higher geometric mean exposure in these 5-minute exposure
estimates relative to the previous estimates for the national distributions of 1 hour
concentration levels at ambient monitors.
Figure 1
Lognormal Plots of the Distributions of 5-minute Maximum
Exposures for Exercising Asthmatics as Modeled by APEX
—Current SO2 Emissions
o
a.
£•§
IE
SV,
cs
81
«! C
I!
Asthmatic child.--as is SO2 y = 1.24+ 0.366x R'2=/).96
All asthmatics-as is SO2 y = 1.12 + 0.375x W/= 0/55
Z-Score
16
-------
Figure 2
Lognormal Plots of the Distributions of 5-minute Maximum
Exposures for Exercising Asthmatics as Modeled by APEX--
Emissions Adjusted to "Just Meet" Current Standards
o
te
03-a
S.2
3-8
II
0-3
!« b
-a
s
Asthmatic child-just meet SO2 y = 1.46 + 0.447x RA2 = 0.982
All asthmatics-just meet SO2 y = 1.26 + 0.493x RA2 = 0.978
Z-Score
2. To what extent are the properties of ambient SCh appropriately characterized,
including ambient levels, spatial and temporal patterns, relationships between
various averaging times, and the relationship between ambient SCh and human
exposure?
3. Twenty locations were selected for detailed analyses, using ambient SCh
monitoring data for years 2002-2006. What are the views of the panel regarding
the appropriateness of these locations, the time period of analysis, and the
approach used to select them?
These seem reasonable to me.
4. In order to simulate just meeting either the current 24-hour or annual standards,
staff adjusted SCh air quality levels for the years 2002-2006 upwards in all but
one location. Ambient monitoring data in North Hampton County PA were above
17
-------
the 24-hour standard in the year 2006 and were therefore adjusted downward. To
what extent is the approach taken technically sound, clearly communicated, and
appropriately characterized?
These seem reasonable to me.
5. What are the views of the Panel regarding the adequacy of the assessment of
uncertainty and variability?
The document appears to do a reasonable job at this.
Exposure Analysis (Chapters 2, 7):
1. To what extent is the assessment, interpretation, and presentation of the initial
results of the exposure analysis technically sound, clearly communicated, and
appropriately characterized?
I have a number of problems with the analysis and its summarization.
Specifically:
p. 142—the method for assessing indoor exposures assumes there is only one
peak outdoor level per hour of exposure—all the rest of the 5 minute periods will
have an average level assigned, calculated after excluding the peak 5 minutes.
This will mean that indoor exposures will have much lower peak levels as slow
air exchange rates will effectively dilute the 5 minute peaks toward the hourly
averages.
p. 145—removal rate distributions are represented as uniform and rather high—
"Resulting estimates were as follows; morning: 4.9 - 19.8 h-i and afternoon: 3.4 -
9.8 h-i " How are these derived from the data? Reproduce some summary of the
data and analysis from the cited paper of Grontoft and Raychaudhuri, 2004.
Grontoft T and MR Raychaudhuri. 2004. Compilation of Tables of Surface
Deposition Velocities for O3, NO2 and SO2 to a Range of Indoor Surfaces.
Atmos Environ. 38:533-544.27
In general I disapprove of the use of uniform distributions because they imply
zero probability of occurrence of values outside the designated range. There is
usually no good reason to assume this. The data underlying these distributions
must be fully described, in addition to reasonable methodology to derive
distributional inputs for this factor for use in APEX. Because the indicated ranges
will lead to very large reductions in expected SO2 indoor concentrations, this is a
key issue for the modeling of indoor exposures.
18
-------
p. 157—the uncertainty section does not discuss the uncertainties in the indoor
removal rate—likely a very influential variable, at least for indoor exposures
2. The draft risk and exposure assessment evaluates exposures in selected
locations encompassing a variety of SCh emission source types in the state of
Missouri; these areas were chosen as an initial case study since 1) air quality
measurements indicated numerous exceedances of 5-minute benchmark values, 2)
there are multiple stationary source emissions above 1,000 tons per year, and 3)
there are numerous ambient monitors measuring 5-minute and 1-hour SCh
concentrations. The second draft may also evaluate exposures in the remainder of
Missouri and also include areas of Pennsylvania, West Virginia, and other
locations with large SCh emission sources. What are the views of the panel
regarding the appropriateness of these proposed additional locations and on the
approach used to select them?
These seem reasonable to me.
3. Do Panel members have comments on the appropriateness and/or relevance of
the populations evaluated in the exposure assessment?
These seem reasonable to me.
4. To what extent are the approaches taken to model SCh emission sources
technically sound and clearly communicated?
P. 125 discusses the approach for using Aeromod for deriving outdoor
concentrations for input into the exposure model as follows:
"As discussed above, as a first approximation point sources at major facilities
were assumed to represent the SCh emissions throughout Missouri2o, where
major facilities were defined as those with SCh emissions totals exceeding 1,000
tpy. Nationwide, there are 918 major facilities and 10,651 associated stacks,
according to the 2002 NEI. Within Missouri, 281 major facility stacks were
identified, but only 115 of these stacks have greater than or equal to 1.0 tpy SCh
emissions in the 2002 NEI. Each of these stacks was paired to a surface
meteorological station, defining its modeling domain. These are the final list of
stacks identified in Table 7-1, above. "
It seems to me this guarantees an underestimation of emissions as the
concentrations resulting from many smaller sources within and outside 20 km of
the major sources will be omitted. The document as it stands does not seem to
provide an approach for adjusting the estimated ambient outdoor concentrations
upward to reflect this source of systematic bias.
19
-------
Some comparison between predicted and measured concentration distributions for
a few monitors n Green County, Mo. is provided in Figure 7-3. The figure does
not provide the detail needed for quantitative comparison that a tabular
comparison would; and the discussion is vague and qualitative—saying mainly
that the distributions seen at the monitors are "bounded" by the modeled values.
The EPA should develop a procedure to quantitatively adjust the modeled
distributions to distributions observed at some reasonably representative set of
monitors, as was previously suggested for the NO2 analysis.
5. Human exposures were modeled using APEX to simulate the movement of
individuals through different microenvironments. Do Panel members have
comments on the microenvironments modeled?
Characterization of Health Risks (Chapters 3, 4, 5, 6, 7, 8, 9):
1. What are the views of the Panel on the overall characterization of the health
evidence for SCh? Is this presentation clear and appropriately balanced?
2. The characterization of health risks focuses on potential health benchmark
values identified from the experimental SCh human exposure literature on lung
function with accompanying respiratory symptoms. What are the views of the
Panel on using potential health benchmarks from this literature to characterize
health risks?
As mentioned in my response above to question #1 of the air quality analysis, the
distribution of individual sensitivities among asthmatics mean that there are likely
to be appreciable numbers of asthmatics who respond to 5 minute exposure to
concentrations less than the lowest 400 ppb benchmark analyzed. At the very
least a series of lower benchmark values should be included in parallel
characterizations.
3. Do panel members have comments on the range of potential health effects
benchmark values chosen to characterize risks associated with 5-minute SCh
exposures?
As mentioned in my response above to question #1 of the air quality analysis, the
distribution of individual sensitivities among asthmatics mean that there are likely
to be appreciable numbers of asthmatics who respond to 5 minute exposure to
concentrations less than the lowest 400 ppb benchmark analyzed. At the very
least a series of lower benchmark values should be included in parallel
characterizations.
4. To what extent is the assessment, interpretation, and presentation of initial risk
characterization results technically sound, clearly communicated, and
appropriately characterized?
20
-------
5. The epidemiology literature will be used to qualitatively characterize SCh-
related health risks for health outcomes such as respiratory symptoms and
emergency department visits and hospital admissions for respiratory-related
causes. However, staff has judged that it is not appropriate to use the available
SCh epidemiological studies as the basis for a quantitative risk assessment in this
review. Do panel members have comments on this judgment and/or on the
rationale presented to support it?
In a late presentation that did not make it in to the written REA we are reviewing,
it now appears that the REA authors will derive a concentration-response function
from the clinical data for asthmatics and therefore be in a position to quantify the
extent of likely asthmatic responses to the full distribution of concentrations to
which people are exposed under different SO2 NAAQS options. This is a giant
step forward. However the current model under consideration is based on a 3-
parameter logistic function with a firm upper limit of population response that is
less than 100%. At least to provide sensitivity analysis for this model form, EPA
should include calculations based on a probit model with one or two
subpopulation modes. The probit model is based on an assumption that individual
thresholds for response have a lognormal distribution in the population. The two-
population model would hypothesize a mixture of two lognormals to describe the
population distribution20of thresholds among asthmatics. I prefer lognormal
distributions because they have at least some quasi-mechanistic justification: the
likely possibility that there are many factors that contribute to differences among
people in their individual thresholds, and the influences of these factors tend to act
multiplicatively. The normal distribution of log(threshold) values then follows
(approximately) from the central limit theorem. In my view it is better to choose
model(s) with some mechanistic justification because projections to effects at
lower concentrations than covered by the data depend crucially on the consistency
of the model form chosen to represent the real causal relationship.
21
-------
Dr. Ted Russell
Comments on SOx REA - 1st Draft
This document provides an analysis of air quality data and lays out the modeling
approach EPA plans to use to calculate the number of individuals exposed to SO2
levels of concern in association with varying potential standards. The Draft
begins with a history of the standard and overviews of SO2 exposure, at risk
populations and health effects. The two main components of the current draft are
the ambient air quality characterization for 5-minute exposures and the lay out of
the exposure analysis. The Health Risk Assessment and Risk Characterization
chapters are not yet fully developed. The current draft shows a significant amount
of effort.
A starting comment is that the Introduction should lay out a road map for the
document discussing what is being done and why. A second comment is that the
while some of the document is relatively easy to read, other aspects are more
difficult, and one is asking why are they doing this? How will they use this
analysis? Was this necessary? The Overview of the Assessment (Section 1.2.1)
is insufficient in this regard.
Chapter 2 on Human Exposure is quite brief, and is more properly titles an
overview. Given the consideration of 5-minute levels in the risk characterization,
it is curious that this concentrations at that averaging time are not even mentioned
in Section 2.3. Further, is there really much concern about the instrument being
used in regards to attainment demonstration? What is the reason for concern?
The longest of the three paragraphs in Section 2.3 is on the PRB, which is not
even used. This section should be more balanced and address the concerns
addressed in the rest of the document.
Chapter 6, Ambient Air Quality and Benchmark Health Risk Characterization, is
very dense at this point, at it is not even clear what is being gained from all of the
analyses. As a first comment, this chapter needs to be cleaned up and streamlined,
written with specific objectives in mind. Indeed, it appears that more analyses
might have been presented/done than needed.
A primary objective of Chapter 6 is to provide an appropriate characterization of
five-minute peak SO2 concentrations for use in exposure assessment. While
some monitors do provide 5-minute average data, most only provide one-hour
data. However, there are enough locations that provide both to develop
relationships between 5-minute and one-hour average peak concentrations. The
approach taken has many aspects of what I would deem appropriate, but could be
improved, I think. In particular, the 5-minute and 1-hour average concentration
data are derived from the same population of observations of SO2 concentrations
at a single location, so a solid understanding of the distribution of pollutant
concentrations, and correlations between 5-minute and 1-hour levels should
22
-------
provide an avenue for deriving 5-minute peaks from 1-hour peaks. This is the
approach they take in deriving a peak-to-mean ratio. Where I might differ in their
analysis is that I would start from the assumption that the concentration data
follow a log-normal distribution (this should be tested for individual sites as well
as the population as a whole), and for each site, derive the geometric mean and
standard deviation (GM and GSD). A COV uses the traditional standard
deviation, which is based on the underlying population being normally distributed,
which it is not. They could then analyze the relationship between the 5-minute
and 1-hour GSD's. Assuming that they find as good of a relationship between the
GSD's as they did between the COVs (and it would be difficult to think they
would not, given the results for the COVs), they can then readily identify
expected percentile values based upon the observed geometric means (care must
be taken in how to treat below detection limit values). This would negate the
need to do much of the Monte Carlo analyses they currently do. I would think
that a very reasonable functional dependence of the 5-minute peak on the 1-hour
maximum and 1-hour GSD can be found, and that this relationship can be used to
estimate the maximum (or second, third, etc.) 5-minute peak level at each monitor.
In essence, I am suggesting that they use that the concentration data likely are log-
normally distributed to simplify and strengthen much of their current analyses.
As part of this, they should develop the temporal correlation structure of the 5-
minute data, as well as the correlation between 5-minute and 1-hour average data.
From Section 6.2.3.6 and on, the document gets dense, and it appears as though
much of the analyses are not central to the objectives of the chapter. The
motivation behind the analyses need to be better brought out with respect to how
they will be used in the ensuing exposure and risk characterizations.
The APEX modeling chapter (Chapter 7) is much more readily understood than
Chapter 6. The one issue that I am a bit uncomfortable with in this chapter,
however, is the simulation of 5-minute SO2 peaks indoors. Given the inertia
effect of the indoor air diluting the peak levels, the temporal correlation between
outdoor 5-minute levels may be critical to correctly calculating the distribution of
5-minute average levels indoors. Currently, they assume that all of the other 5
minute periods ha d the same concentration. If one were to assume that there
were more structure, e.g., that half of the concentrations were zero, and the other
half at the peak, and further, that all of the peaks occurred together, one could get
a higher peak level indoors (and that level would be very sensitive to the
infiltration rate used).
Page 43 lines 27-28 "although though" should be corrected.
Page 45, line 17: "having an estimated" not "containing estimated".
Figure 6-7: Use a log scale for this type of graph.
Table 6-7: add "(ppb)" in the table
23
-------
Page 68, line 14: This does not really seem to match what is in Fig. 6-16 (and the
upper and lower rows of Fig. 6-16 are nearly the same, and no additional
information is transmitted by the upper row).
Table 7-8: "... the Missouri" What???
Page 143, lines 12 and 13: Replace NO2 with SO2.
Table 7-12: Add units.
Section 7.9.2 and Tables 7-14 through 7-17: Add complete units, e.g., per year,
etc.!
Page 7 llines 2-3 "400 ppb at any one monitor was between 20 to 60 times a year
... less than 1%..."
Figures 6-21 and 6-22: It looks as though there are fractional numbers here.
A first quibble is that the Introduction could be expanded to provide more of a
picture of what was to come. A few paragraphs laying out the approach would be
good, providing a flow of effort and information. Here they can define what
models are to be used and why, as well as the specific outcomes of interest, and
why. A second general comment is that the document is a bit uneven, with some
sections being thorough and readily understood, while others lacked motivation
and it was a bit difficult to see exactly what was done and why.
Page 86, line 12: The reference to the content of Fig. 6-28 is confusing.
Page 103, line 26 "... samplers for short term averages"
Page 103, Line 28 "...days, and 5-minute averages are never available."
Page 104, lines 3-7: Unclear what is being said (and why)
In response to the specific Charge Questions:
Air Quality Information and Analyses (Chapter 6):
1. We have evaluated SO2 air quality throughout the United States, using all
available 5-minute and 1-hour ambient monitoring data for years 1997 through
2007. To what extent are the air quality characterizations and analyses
technically sound, clearly communicated, appropriately characterized, and
relevant to the review of the primary SO2NAAQS?
24
-------
As discussed above, I have a few concerns about Chapter 6. First, that Chapter
does not clearly communicate what has been done, and why, and a large fraction
of what is there ends up appearing to be of secondary relevance to this review. In
many places, the figure and table captions need to be expanded to better convey
what is being plotted/tabulated. Technically, the use of a traditional COV is
questioned since it uses the traditional standard deviation, which is appropriate to
characterize populations that are normally distributed. Primary air pollutant
concentrations typically are log normally distributed, and thus one should log-
transform the data first. That said, I can support the spirit of how they are finding
5-minute peak values given 1-hour data, just that I would look to start with using
geometric means and standard deviations (taking care of how below detection
limit data are treated). I realize it is late in the process, but this is where there is a
mismatch between the ISA and REA in that the ISA has little on 5-minute average
SO2 levels, but it is central to the REA analyses.
In addition to considering characterizing the distributions assuming they follow a
log-normal distribution and developing the appropriate relationships and
correlations between the 5-minute and 1-hour concentrations, I would look to
streamline this chapter with the ultimate goal in mind: to characterize the
distribution of peak 862 levels, particularly those above 400-600 ppb (at least for
now). With that in mind, I would look to see what analyses are central to such.
For example, consider Figure 6-8. Why is one concerned with having 3 different
monitors in the county? What exactly is plotted (the figure caption is insufficient
as to what each dot represents)? The discussion on page 49-50 does not help
answer this question. I think the real question is independent of having three or
more monitors. The discussion related to Table 4 is a bit opaque: how is the COV
defined? How is the COV used? Finally, on page 54, one sees how multiple
monitors are used (but it is still not apparent why this is a requirement), but this
could have been rather simplified.
Section 6.4 starts off well, but then gets bogged down in analyses. (On the other
hand, Table 6-5 should also have slopes from the regression, and I assume that in
Figure 6-9, the RH Column is annual 1-hour average Max to be consistent with
the left hand column). For example, the upper rows in Figures 6-10 and 6-16
provide little extra insight. Fig's 6-13 and 14 are informative.
In regards to Table 6-7, not surprisingly, the distribution of the modeled 5-minute
maximums is not normal, so, again, a standard deviation is not an appropriate
measure. Again, I would providing the geometric mean and standard deviations.
2. To what extent are the properties of ambient SO2 appropriately characterized,
including ambient levels, spatial and temporal patterns, relationships between
various averaging times, and the relationship between ambient SO2 and human
exposure?
25
-------
As noted above, most of the analysis uses statistics and characterizations for
populations that are normally distributed, which they are not. There really is no
solid analysis of the relationship between ambient SO2 and human exposure,
except in the uncertainty section. Given the lack of analysis of this relationship,
the uncertainty discussion seems out of place.
3. Twenty locations were selected for detailed analyses, using ambient SO 2
monitoring data for years 2002-2006. What are the views of the panel regarding
the appropriateness of these locations, the time period of analysis, and the
approach used to select them?
The twenty locations are fine. As noted above, it is not apparent that there was a
need for having three or more monitors in a county was a necessary criteria.
4. In order to simulate just meeting either the current 24-hour or annual
standards, staff adjusted SO 2 air quality levels for the years 2002-2006 upwards
in all but one location. Ambient monitoring data in North Hampton County PA
were abovethe 24-hour standard in the year 2006 and were therefore adjusted
downward. To what extent is the approach taken technically sound, clearly
communicated, andappropriately characterized?
The approach used was fine, though not overly well communicated. This latter
paragraph (i.e., the charge question) actually brings clarity to what was done and
why.
5. What are the views of the Panel regarding the adequacy of the assessment of
uncertainty and variability?
At present, the analysis is qualitative. It should be a bit more quantitative in
regards to the probabilities of having concentrations exceeding specified values,
and the numbers of exceedences. It does reasonably well on the numbers, but
could do a bit better on probabilities of a certain number of exceedences.
Exposure Analysis (Chapters 2, 7):
1. To what extent is the assessment, interpretation, and presentation of the initial
results of the exposure analysis technically sound, clearly communicated, and
appropriately characterized?
2. The draft risk and exposure assessment evaluates exposures in selected
locations encompassing a variety ofSO2 emission source types in the state of
Missouri; these areas were chosen as an initial case study since 1) air quality
measurements indicated numerous exceedances of 5-minute benchmark values, 2)
there are multiple stationary source emissions above 1,000 tons per year, and 3)
there are numerous ambient monitors measuring 5-minute and 1-hour SO2
26
-------
concentrations. The second draft may also evaluate exposures in the remainder
of Missouri and also include areas of Pennsylvania, West Virginia, and other
locations with large SO2 emission sources. What are the views of the panel
regarding the appropriateness of these proposed additional locations and on the
approach used to select them?
While these locations are appropriate, I (and I think the panel) will always be
most interested in a national perspective.
3. Do Panel members have comments on the appropriateness and/or relevance of
depopulations evaluated in the exposure assessment?
They are fine to me.
4. To what extent are the approaches taken to model SO2 emission sources
technically sound and clearly communicated?
AERMOD is the appropriate tool if emissions-based modeling is decided to be the
best route, though I am not sure that one needs to go that way. Might one rely on
just the analysis of the observations? Does using AERMOD add an extra
complication?
5. Human exposures were modeled using APEX to simulate the movement of
individuals through different microenvironments. Do Panel members have
comments on the microenvironments modeled?
While APEX is an appropriate tool to be used in this case, the continued reliance
on APEX should push EPA to further evaluate the model across a range of
conditions and pollutants. The lack of evaluation in this application is not
comforting, though understandable given the limitations in measurements
available. Also, given the task at hand, i.e., simulating 5-minute maximums, how
infiltration is done is important. Also, it would be of interest to show where and
when the exposures to > 400, 500 and 600 ppb occur. Do they happen in the
home, at night, etc. This is an uncommon, short term, affect.
In Tables 7-14, 15: The number exposed above 0 should be all of the individuals,
independent of number of exposures.
In regards to the uncertainty discussion, the treatment of the AER's and air
conditioning prevalence could be very important if the 5-minute peak exposures
above the thresholds are happening indoors. These sections may need to be
bolstered if that is the case.
27
-------
Dr. Donna Kenski
Comments on SOx REA - 1st Draft
General comments and responses to charge questions:
1 We have evaluated SO2 air quality throughout the United States, using all
available 5-minute and 1-hour ambient monitoring data for years 1997 through
2007. To what extent are the air quality characterizations and analyses technically
sound, clearly communicated, appropriately characterized, and relevant to the
review of the primary SO2 NAAQS?
The 5-minute and 1-hour data has been exhaustively analyzed in Section 6, but it
was not always easy to see the path being followed or the logic of the method
pursued. I sometimes felt lost in the minutia, and had a hard time keeping all the
pieces of this analysis in perspective. So it could use some additional clarification
of the overall structure. Or maybe just some judicious editing with less detail an
more summarizing— some suggestions for items that could be sent to an
appendix are below. Other items were not explored as thoroughly as needed,
however. Two of the first sections brought up issues that were never returned to;
the duplicate dataset (6.2.1) and the distance from monitor to sources (6.2.2). For
instance, where did the duplicate data enter into the QA process? I couldn't find it
referred to again, after the first description, until Appendix A. It seems like this
dataset should have been used to test the PMR model, but I couldn't see
any indication of that. The analysis of the duplicates in the Annex was okay, but
this particular part of the dataset could have been used more effectively in model
validation. Likewise, the characterization of monitors by their distance from
sources seems like information that could have been used to improve or inform
the predictive model. The choice of COV as a predictive categorical variable is
reasonable, but the REA could have benefited from a more comprehensive
discussion of possible models and the rationale for that particular choice. Sec.
6.2.3.1 (Background) explains that peak concentrations are likely to be influenced
by distance from sources and source characteristics, but the subsequent
justification for the COV model was weak. Since data on source types,
emissions, and distance from monitors were available, it is not clear why they
were not explored at least briefly. More importantly, there is some discussion in
Sec. 6.2.3.6 (Evaluation of Estimation Procedure) of model fit and some outliers.
The poor model fit at 2 specific monitors is discussed as being perhaps a function
of the proximity of the monitors to the nearby sources, or some unspecified
characteristic of the sources that causes them to be poorly described by the
statistical model. These two cases are then excluded to demonstrate improved
agreement. But these two cases are among those that should have the closest
scrutiny, since they are generating values at the extremes of the distribution. They
should be examined in detail rather than discarded for the sake of showing better
model performance.
28
-------
2 To what extent are the properties of ambient SO2 appropriately characterized,
including ambient levels, spatial and temporal patterns, relationships between
various averaging times, and the relationship between ambient SO2 and human
exposure?
Most of the comments above pertain to this question as well. In addition, the trend
information in section 6.4.2 seems like it is of limited use in this analysis. Perhaps
it belongs in the appendix? It is well documented that SO2 concentrations have
been declining as a result of several regulatory programs. I'm not sure how those
trends are helpful in interpreting the risk and exposure assessments that are made,
or will be made, in this document. Some additional justification of this particular
analysis would be helpful. Excluding the results for the Caribou ID monitor (p.
70) is another instance where an outlier is discarded that might be more useful if
analyzed separately or in more detail to look at the reasons for its behavior. The
exclusion of the Hawaii County data, on the other hand, is perfectly valid.
3. Twenty locations were selected for detailed analyses, using ambient SO2
monitoring data for years 2002-2006. What are the views of the panel regarding
the appropriateness of these locations, the time period of analysis, and the
approach used to select them?
The first 3 of the 20 locations are certainly good choices. It is not clear exactly
why the remaining 17 were selected - i.e., why was it necessary to have 3
monitors in a county? Surely it is more important to have the highest-
concentration monitors represented? I can't tell what impact this choice of
monitors might have on the ultimate results of this analysis, but it seems as
though it might be significant in terms of the number of potential exceedances.
Consequently the selection rationale needs to be more completely justified,
and/or some of the higher concentration monitors should replace the 3-monitor
counties.
3 In order to simulate just meeting either the current 24-hour or annual standards,
staff adjusted SO2 air quality levels for the years 2002-2006 upwards in all but
one location. Ambient monitoring data in North Hampton County PA were above
the 24-hour standard in the year 2006 and were therefore adjusted downward. To
what extent is the approach taken technically sound, clearly communicated, and
appropriately characterized?
I thought this approach was fine and clearly communicated.
4 What are the views of the Panel regarding the adequacy of the assessment of
uncertainty and variability?
Frequently unclear. Section 6.2.1 starts off with a description of a dataset of
duplicated measures that were used for quality assurance, but the rest of the
document doesn't refer back to this particular dataset so it is hard to assess the
level of QA with these data. Then later, in Sec. 6.2.3.6 (~ p. 45) it sounds like the
29
-------
model estimates are compared with measured values at the same sites that were
used to develop the model. Maybe I'm misreading this? Obviously an
independent dataset should be used to evaluate the model. You could do this by
reserving some fraction of the data for this purpose if the collocated duplicates are
not suitable. In either case it is not clear whether the model has been evaluated
with the appropriate set of data and that should be clarified. Overall, Sec. 6.5 did a
nice job summarizing in a qualitative way the various sources of uncertainty. It
would be nice to have a tabular, graphical or bullet summary of the various
uncertainties described in section 6.5.
Specific comments:
p. 10, line 15 delete the 'is' after scheduled
p. 11 line 14 missing a period
p. 14 line 15 there -> their
p. 19 the bullet list of key conclusions is nice. In fact this whole introduction
section was well done.
p. 20 line 5 as low as 0.4 ppm
p. 22 line 15 visits
p. 25 1st paragraph add the n for this study.
p. 25 line 28 really per 40 ppb? Or 10 ppb?
p. 28 lines 19-23 This sentence is unbearably long and should be broken up and/or
reworded for clarity
p. 32 line 3 data were assembled
p. 35, figure 6-1 The labels in this figure don't seem right. How could hydro
power contribute 30% of SO2 emissions? And in part B, electric power
generation is allocated 2%, vs. fossil fuel power generation at 45%. Needs
clarification.
p. 40, footnote reference to Fig. 4 should be Fig. 6-4
p. 42, line 17 1-hour measurements should be 5-minute?
p. 42, footnote Why was a uniform distribution used? It's not clear whether that's
the appropriate choice here; provide some justification. Also, should be ...was
based on selection of a value...
p. 43, line 28 delete 'though'; change resultant to resulting
p. 50, line 8 delete 'both'
p. 57 Fig. 6-9 Add units to coefficient of variatiability
p. 59 Caption of this figure is a bit confusing, as it uses the words exceedance and
benchmark interchangeably. Since exceedance has a specific regulatory
implication, and that's not what's being discussed here, it would be better to stick
with benchmark (this situation comes up in numerous places in the text and
figures)
p. 64, line 27 series of figures
p. 65, Fig. 6-15 Add units to coefficient of variatiability
p. 68, line 4 re -> Fig.
pps 69 and 73: In both of these figures, the top row and bottom row are so similar
that they can't be meaningfully distinguished from each other. Replot on a log
scale to show the differences, if they are important (and here's another
exceedance vs. benchmark confusion)
30
-------
pps. 75-78, Figs. 6.20-6.23 It might be helpful to color-code the points by year
and graphically make the point that the concentrations above the benchmark
occurred only in early years.
Pps 84-86, Figs 6.24-6.26 This series of figures is not very effective as laid out.
They would be better if the bars were side by side so we could see the change in
magnitude between the as-is and adjusted values.
p. 100, line 18 ...; however, it incorporates...
p. 100, line 29 delete data
p. 101, line 16 impact on
p. 102, line 1 delete those
p. 104, lines 1-3 fix incomplete sentence
p. 108-110, Figs 6.34-6.36 the symbols aren't really legible on these plots
-------
Dr. Patrick Kinney
Comments on SOx REA - 1st Draft
Exposure Analysis Charge Question 5: Human exposures were modeled using
APEX to simulate the movement of individuals through different
microenvironments. Do Panel members have comments on the
microenvironments modeled?
Overall, the approach taken by EPA in applying APEX to the SO2 exposure and
risk assessment represents best available practice using currently-available
modeling tools. The microenvironments chosen for inclusion, and the parameters
assigned to each, are reasonable.
It is worth noting that the human activity data base upon which the modeling
work depends represents a compilation of results from human activity surveys
conducted between 1982 and 1998, and thus are 10 or more years old. EPA
should consider updating these data periodically, both by summarizing results
from more recent time/activity survey studies, and if necessary, by sponsoring
new population-based surveys.
p. 114, line 23 through p. 115, line 2: this discussion is unclear.
p. 143, lines 12-14: this text is for NO2. Please edit for SO2.
32
-------
Dr. Steven Kleeburger
Comments on SOx REA - 1st Draft
Characterization of Health Risks (Chapters 3, 4, 5, 6, 7, 8, 9):
3. Do panel members have comments on the range of potential health effects
benchmark values chosen to characterize risks associated with 5-minute SCh
exposures?
The potential health effects chosen for consideration that are consistent or
in common between the ISA and REA documents include respiratory symptoms
(e.g. wheeze, chest tightness, cough, substernal irritation), lung function (e.g.
change in FEV1, sRaw,, decrements in lung function in the presence of
respiratory symptoms, and cardiovascular parameters. Given the existing
epidemiological, clinical, and animal model investigations of health effects
related to 5-10 minute SO2 exposures, the selection of these health effects was
reasonable. Moreover, the potential "affected individual" or
susceptible/vulnerable subpopulation(s) were appropriate. The presented
summaries suggested appropriately that individuals with asthma and potentially
other preexisting lung diseases (e.g. COPD) are more likely to have an adverse
outcome in response to short-term peak exposure to SO2 than individuals without
preexisting disease.
Genetic background and age as susceptibility factors were also presented.
While the REA appropriately indicated that limited data exist to reach a
conclusion regarding the importance of genetic background as a susceptibility
factor, the REA should include a statement indicating genetic susceptibility needs
to be better characterized. Only one polymorphism has been evaluated for
increased risk of susceptibility to SO2 effects (-308 TNF promoter SNP) and thus
represents only a beginning. The revision of the draft REA provides an excellent
opportunity to propose that a more thorough examination of genetic contribution
is needed. The current evidence for genetic component of host responsivity to
other criteria pollutants is strong (e.g. ozone), and it is likely that genetic variants
will also be important in response to SO2.
Comments similar to the above can be made for differential responsivity
attributable to age, although more studies currently exist that suggest age is an
important susceptibility factor. Nonetheless, recommendations could be made for
additional investigations to understand the relationship between age and response
to 5-10 minute exposures to SO2, especially in the very young and elderly.
33
-------
Dr. Timothy Larson
Comments on SOx REA - 1st Draft
1. We have evaluated SO2 air quality throughout the United States, using all
available 5-minute and 1-hour ambient monitoring data for years 1997 through
2007. To what extent are the air quality characterizations and analyses
technically sound, clearly communicated, appropriately characterized, and
relevant to the review of the primary SO2NAAQS?
The staff are to be commended for compiling and distilling this short term data.
These analyses are relevant to the review of the primary NAAQS, given that there
is strong evidence for effects from these short-term exposures above certain
thresholds. These data are limited in geographical scope, but inclusion of the 5-
minute maximum data as well as the continuous 5-minute data provides a
reasonable data base.
2. To what extent are the properties of ambient SO2 appropriately characterized,
including ambient levels, spatial and temporal patterns, relationships between
various averaging times, and the relationship between ambient SO2 and human
exposure?
The spatial variation of 24-hour and annual averages across the country on a large
scale is well characterized. However, there are relatively few urban areas with
multiple monitors and so it is difficult to assess intraurban spatial patterns based
upon measurements. Therefore the reliance on plume models to infer the smaller
scale variations is the only reasonable approach that is available. Those areas
with multiple monitors been identified and given appropriate priority for inclusion
in the larger modeling exercise.
The use of a pdf for the peak to mean ratios rather than applying a single value is
appropriate. Defining a few different pdfs categorized according to SC>2
concentration and to proximity to major sources is a creative and useful approach
that appears to converge to stable predictions in the final simulation.
The REA should clarify the major cause of the observed short-term peak
concentrations, specifically whether these concentrations are correlated with peak
emissions or unfavorable meteorology. As it stands, the implication is that
meteorology is driving these peaks. The contribution to short-term ambient peaks
from "upset" emission conditions at nearby major point sources as documented by
continuous emission monitoring information should be clarified. If there is a
correlation between the peaks and "upset" emissions, this would provide another
way to assess the appropriate locations for analysis as discussed in question 3
below.
34
-------
3. Twenty locations were selected for detailed analyses, using ambient SO 2
monitoring data for years 2002-2006. What are the views of the panel regarding
the appropriateness of these locations, the time period of analysis, and the
approach used to select them?
The three locations with relatively high 5-minute peaks are an obvious choice.
The other 17 locations could have been chosen by any number of criteria.
Choosing to limit the analysis to locations with multiple monitors in a given
county is one reasonable approach aimed at capturing more spatial variation
relative to a single monitor. Ranking the sites using the minimum adjustment
factor (typically the one based on the 2nd highest 24 hour maximum) is reasonable.
However, one could also argue that some of these 17 additional locations could
have been chosen based on potential for high downwind concentrations at
locations other than the monitoring site. For example, using the emissions
information in Table A-4 one can identify sources in Georgia, Kentucky,
Minnesota, New York and Montana that have high emissions but whose locations
are not included in the final list of 17. Some of these sources are somewhat
isolated, but not all of them. Given that the exposure assessment in the REA
predicts very few encounters with high 5-minute peak values at ground level,
including some of these locations could alter the results. One approach is to do a
simple screening level analysis based on plume impacts at all sites (e.g.
Aerscreen) and then rank the locations.
4. In order to simulate just meeting either the current 24-hour or annual
standards, staff adjusted SO 2 air quality levels for the years 2002-2006 upwards
in all but one location. Ambient monitoring data in North Hampton County PA
were above the 24-hour standard in the year 2006 and were therefore adjusted
downward. To what extent is the approach taken technically sound, clearly
communicated, and appropriately characterized?
The approach seems reasonable, given the lack of spatial information needed in
order to include a space/time interaction (rather than the pure temporal adjustment
based on one site applied equally to all sites). The approach is clearly
communicated.
5. What are the views of the Panel regarding the adequacy of the assessment of
uncertainty and variability?
Additional limitations include the fact that: 1) instances of building downwash of
the plume is not being considered in the model (especially for older coal plants
with relatively short stacks) and 2) that the effects of complex terrain are not
being incorporated because the modeling locations chosen are not in such terrain.
Monitors sited to capture the effects of building downwash or plume impaction on
nearby, elevated terrain would measure higher peak hourly SO2 levels than if they
35
-------
were located in flat terrain with unobstructed flow between the monitor and the
stack, even if the emission rates are moderate. Those sited in the wake of
buildings might also display different peak to mean ratios due to the different
turbulence structure in this microenvironment.
Specific Comments:
p. 35 The pie chart in Figure 6-1 lists Hydroelectric Power Generation as a source
of SO2. Are these emissions from facilities that combine both hydroelectric and
coal-fired power plants? Hydroelectric plants by themselves do not emit SO2.
p. 42, line 17 should read "did not contain 5-minute measurements.."
pp 46-47 The monitor with the maximum number of 5-minute peaks in Figure 6-
7 is actually located at the base of a ridge that runs between the Glover smelter
stack and the receptor site. The site that is further away from this source is in
open, flat terrain where the model presumably performs much better.
p 59 The bottom row of Figure 6-10 is presented as if it is a subset of the top row,
yet the y-axis scales indicate the opposite is true.
p 125 line 20. Were any receptors located above stack height? Ditto for the
results shown in Figure 7-3.
PI43 Table 7-11 Other microenvironments could include the recirculating cavity
induced by building downwash that is located next to a stack with less than GEP
stack height and the elevated receptor on an isolated hill that is directly downwind
at plume centerline height (the plume wrapping case under stable conditions aloft).
36
-------
Dr. Kent E. Pinkerton
Comments on SOx REA - 1st Draft
Characterization of Health Risks (Chapters 3, 4, 5, 6, 7, 8, 9):
General Comments:
The first draft of the REA provides an excellent overview and extensive
documentation that will be critical for the risk and exposure assessment plans in
the review of the SO2 national ambient air quality standard. The identification of
sources for human exposure is important to clearly establish in order to better
characterize personal exposure to ambient concentrations. County selection based
on known SC>2 sources and archived SC>2 monitored data is excellent in providing
substantive characterization for benchmark health risks for 5-minute peak SO2
exposure. I fully agree with the designated at risk populations to 862 exposure
and feel the human clinical studies are highly appropriate to form the basis for
establishing the potential health effect benchmark values. The characterization of
air quality and exposure analysis is impressive and presented in great detail.
Charge Question 4. To what extent is the assessment, interpretation, and
presentation of initial risk characterization results technically sound, clearly
communicated, and
appropriately characterized?
Response: It is my impression the assessment, interpretation, and presentation of
initial risk characterization results are technically sound, clearly communicated,
and highly reasonable in the manner it has been outlined and reported in this first
draft. The approach for assessing exposure and risk associated with 5-minute
peak SC>2 exposure is extremely reasonable and based on the findings of the
controlled human exposure studies. Although 0.4 to 0.6 ppm 862 is being
selected from these human clinical studies as the appropriate range to use in
benchmark analyses associated with 5-minute peak SC>2 concentrations, it
continues to be critical that 0.2 to 0.3 ppm peak SC>2 exposure also shows effects.
Therefore, it is important to further justify the higher concentration of 862
exposure selected to use in this process. Also county selection for basing
substantive characterization for benchmark health risks for 5-minute peak SC>2
exposure should be clarified to insure how representative each location is and how
the sum of the findings will be applied across the country for risk and exposure in
establishing a national standard.
Specific Comments:
37
-------
Section 4.2.4 Decrement in lung function in the presence of respiratory symptoms
(pages 22-23). For the study by Schwartz et al (1994), once co-pollutants were
adjusted, was the 862 effect still significant? If so, please indicate. As stated, the
effect is substantially reduced.
Although the staff has decided that it is not appropriate to use the epidemiological
studies as the basis for a quantitative risk assessment, these studies continue to
provide further validation of SC>2 exposure effects and should be given some
consideration. It is good to see qualitative assessments of the epidemiology will
be considered, but it would be good to specifically define how this qualitative
assessment will be used.
38
-------
Dr. Jonathan Samet
Comments on SOx REA - 1st Draft
General Comments:
This first draft REA for SC>2 provides extensive documentation for the plan on
developing exposure profiles for the susceptible population. It still does not
provide full details of the approach for assessing health risks, although the general
framework is set out. The document would benefit if more overall structure were
provided initially for the general approach that will be followed. In fact, it is not
until chapter 8, which discusses the health risk assessment, that a general
framework is offered for the risk and exposure assessment in Figure 8-1. It would
be useful for readers if this figure were provided much earlier in the document. In
fact, readers of the extensive chapters on assessment of concentration data and of
exposure estimates would benefit from a better presentation of the overall
structure of the risk assessment.
With regard to the characterization of health risks, my specific comments follow:
Charge Question 1
This question refers to the overall characterization of the health evidence for SC>2.
The draft REA draws on the ISA in selecting the outcomes and exposure-response
relationships to be used. The reliance on the clinical studies of persons with
asthma is appropriate. There is a clear documentation of a causal association and
the exposure-response relationship has been characterized with reasonable
certainty. I am less certain as to the nature of the "qualitative" assessment that
will be carried out using the epidemiological data (Charge Question 5). The
positive risk estimates from the epidemiological studies selected will, of course,
indicate an adverse effect. I did not find sufficient specificity on the approach and
how the resulting information would be useful for assessing policy options.
The discussion of uncertainty and variability remains completely generic. At this
point, while there is extensive discussion of these matters with regard to exposure,
and a probabilistic approach is described for addressing uncertainty in health
estimates, the overall approach in the risk characterization remains to be specified.
Specific Comments:
Chapter
#-Page #
2
Line
#
13-14
Comment
The concern with regard to misclassification
arises in the
39
-------
Chapter
#-Page #
2-11
2-12
2-13
2-13
2-13
2-14
3-16
3-17
4-22
4-22
4-23
4-23
Line
#
17
28-29
16
22-23
23-24
3-5
20
11-14
16-18
20
6-9
26-28
Comment
context of hypothesis testing, and not necessarily with
exposure assessment.
What is the distinction between "instantaneous" and "peak"
exposure?
Is this uncertainty with regard to limited detection relevant to
the discussion of peaks?
Replace "reliable" with "accurate"
This sentence is far too general and needs specificity.
While SC>2 levels may be difficult to measure at lower
concentrations, they have little relevance to health.
The finding of low site-to-site correlations implies higher
spatial variability.
Does this section on "age" refer to children and elderly
persons with asthma?
The definition of "vulnerability" seems to have slipped from
that in the ISA. Scenarios reviewed here refer to greater dose
and not necessarily to a greater potential for exposure, the
definition of vulnerability previously used.
In what way was the evidence found to be "most robust"?
What was the criterion?
This comment concerning the epidemiological studies seems
inconsistent with the view given that they do not address SC>2
alone.
The lag structure identified in this study seems quite
inconsistent with the findings of the clinical studies. A
comment is needed.
A change in the estimate with inclusion of additional
variables in the model does not necessarily imply
confounding.
40
-------
Dr. Richard Schlesinger
Comments on SOx REA - 1st Draft
Overall, the document could use some general streamlining and editing to make it
read "smoother." For example, in Chapter 2 it is not clear why information from
earlier documents than the current ISA is repeated in some detail here (e.g., page
20 lines 1-13; p. 21 lines 12-18.) rather than presenting a summary of past studies.
In addition, this document should use the details presented in the ISA to support
the assessment approach used and not repeat details even of key studies (e.g., p.
20,1. 14-27). Chapter 7 is very di
p. 11,1. 20-21. This sentence is not necessarily totally true. While exposures are
clearly likely in vicinity of source, SO2 is a regional pollutant as well and
exposures may be in areas away from specific sources.
p. 17,1. 11-14. Here there is a mixing of susceptibility and vulnerability.
Asthmatics are susceptible and people who work outside in general may be more
vulnerable.
p. 19,1. 4 and 1. 9.1 think there is an error here in that the same terms are used in
two places.
p. 116, Section 7.3.1. There needs to be better justification for use of data from
Missouri when it was indicated that it was one of a few states that apparently had
data that would allow for assessment of the modeling approach used.
41
-------
Dr. Christian Seigneur
Comments on SOx REA - 1st Draft
Air Quality Information and Analyses (Chapter 6)
Overall, the air quality analysis is technically sound and appropriate for an SOX
risk and exposure assessment. My major comment pertains to Question 4: Some
discussion on how an annual average air quality standards can be compared to 5-
minute average values (see Section 5.2) is warranted. Alternatively, could EPA
simply state that an analysis of the current annual NAAQS is inappropriate based
on Table 5-3 of the ISA (see Section 4.1) since the presence or absence of any
causal relationship cannot be inferred for any long-term exposure related effects
(morbidity and mortality)? For example, Figure 6-30 shows that there is a fair
amount of scatter between the number of exceedances of the 5-minute health
benchmarks and the annual average 862 concentration. Then, only the short-term
(24-hour average) standard would be analyzed.
Exposure Analysis
The exposure analysis chapters are clearly written and the overall technical
approach is sound. The use of AERMOD for atmospheric dispersion modeling
and APEX for population exposure estimates is appropriate. My major comment
pertains to Question 2: The areas selected tend to focus on inland areas impacted
by large stationary sources (coal-fired power plants, cement plants, chemical
manufacturing plants, smelters). Thus, the potential impact of mobile sources is
not directly addressed. As stationary sources undergo emission controls, the
relative importance of some uncontrolled mobile sources (e.g., diesel-powered
ships) may increase. Therefore, it would be worthwhile to model an area (in
addition to Missouri) where ship emissions could have a significant impact on the
population (e.g., Houston, TX or Los Angeles, CA).
Editorial comments:
p. 11, line 21: "principal" instead of "principles".
p. 12, line 15: add "coal-fired" before "electric generating units".
p. 14, line 15: "their" instead of "there".
p. 15, line 24: "attributable" instead of "attributible".
p. 19, the bullet on line 9 ("Short-term respiratory morbidity") under "inadequate
to infer the presence or absence of a causal relationship" should be deleted since it
is listed on line 5 as "sufficient to infer a causal relationship" (see Table 5-3 of
ISA 2nd draft).
42
-------
p. 29, line 21: "Canadian".
p. 37, line 7: add "minute" after "continuous-5".
43
-------
Dr. Lianne Sheppard
Comments on SOx REA- 1st Draft
Air quality information and analyses (chapter 6):
This chapter relies exclusively on the monitoring data. The introductory section
should be expanded to describe the purpose of the analyses.
While the analysis that relates the full 5-minute dataset to the 5-minute maximum
dataset appears generally appropriate (there is an important exception noted
below), the question of spatial representativeness is not considered outside of the
universe of available monitors. What is the SO2 monitoring network supposed to
represent? Is an unweighted summary of this network the best way to
characterize 5-minute maxima?
Concern about the 5-minute dataset comparisons: The poor model fit at 2
monitors (see figure 6-7 p 47) needs much more careful investigation. Note that
at the monitor with the highest number of measured exceedances, the number
missed by the prediction exceeds the number of measured exceedances at any of
the other monitors in the dataset.
There is something strange about 2004 in Figure 6-12 (p 61) that suggests some
undocumented feature of the dataset that produces such a low normalized number
of exceedances. The discussion on p. 58 mentions an Iron County Missouri
monitor that ceased operation in 2003, but more needs to be done to determine if
conclusions about trends reflect real phenomena or are merely features of the
dataset that should not be generalized. This is one example of an aspect of the
analysis that comes up several times in the chapter: it is important to be able to
distinguish temporal trends in number of monitors in the network from downward
trends in the concentration of SO2. Analyses need to be done to ensure that
reductions in SO2 over time are real and not just an artifact of the change in the
monitoring network. (For another example see the discussion on lines 8-12 p.
71.)
Many tables and figures need added clarification of titles, headings, or axis labels
to ensure the reader doesn't interpret modeled or adjusted concentrations as
though they are measured concentrations. The information may be in the caption,
but it is easy to miss there. Examples include Table 6-12 (cone summary), Table
6-14 (cone summary), Figures 6-15, 6-23, 6-25, 6-30, 6-31, 6-32.
Add to Table 6-9 the number of monitors in each county and the number of
neighborhood scale monitors.
The comparison of Figure 6-13 with 6-21 and 6-14 with 6-22 suggests much
stronger correlation in the modeled than measured data, and monitors with much
44
-------
higher number of exceedances at low values of annual average in the measured
data than in the modeled data. These figures should be put on the same page and
direct comparisons made. These comparisons suggest peaks may be
underestimated, particularly for low annual average concentrations.
The uncertainty and variability discussion needs work. It is unclear what the
analyzed monitors are supposed to represent spatially. This is relevant to a 6.51
conclusion (pi00 1 29-30) since which monitors are included has significant
impact on the generated results (even when the measurements themselves are of
good quality). 6.5.4: There is almost certainly lack of spatial representation and
uncertainty due to exclusion of monitors near local sources (102 1 28-30).
Undefined spatial representation of monitors is one of the biggest challenges in
making sense of the results presented in this chapter, and I would even discourage
the assumption that data are representative of the locations analyzed (102 1 21-22)
since the spatial scale is not defined in that comment and there is huge spatial
variability in SO2 even within areas. 6.5.7: The section on the statistical model
discusses the data, not the model. The reasons for exclusions need to be
documented. The poor fit of the model in two locations is of concern and should
be discussed.
Exposure analysis (chapters 2, 7):
Chapter 2 should look ahead to the use of estimates of exposure developed in
chapter 7 for health risk analyses. Here are some questions:
• Is it surprising that there is poor site-to-site correlation of SO2 among
monitors when these monitors are sited to capture local sources?
• If the number of 5-minute peak exposures to asthmatic individuals is as
low as is estimated in chapter 7, is it worth continuing to the health
analysis? This turns out to not be the right question, since chapter 8 uses
the full range of the predicted SO2 exposure distribution and does not rely
on exceedances alone. Can chapter 2 (or chapter 5) discuss the
groundwork for the understanding of the exposure and risk assessment
chapters?
Chapter 7 seems overall reasonable, with the exception of a few details discussed
below. Assuming no changes, the key conclusion of the exposure analysis in this
chapter is that in the modeled area, the number of potentially harmful exposures
to at risk individuals for short 5-minute periods is low. Now with insight into the
use of these predictions in chapter 8, the focus of chapter 7 needs to be expanded
to also assess the entire distribution of predicted SO2 since that is used in the risk
assessment. Finally, analysis should be done to align the estimates in chapter 6
with those produced in chapter 7 so the reader can understand why and how the
two sets of estimates of peak exposures are different.
Concerns with the exposure model:
45
-------
• p 142: The estimation of the additional 5-minute concentrations forces all
the other values near the mean. This reduction in variability of the
modeled data should effectively reduce the predicted number of 5-minute
exceedances in any given hour. The analysis presented in Table 6-15
suggests this variance reduction will be too strong for the intended use of
the modeled data.
• p 132: The comparison of measured data to the extremes of the
distribution of modeled data appears to be a very weak test of the
predictive capacity of the AERMOD model. Even so, Figure 7-4 suggests
the predicted data don't capture most of the distribution of the measured
data at that monitor, even if the upper tail is within bounds.
Response to charge questions:
o To what extent is the assessment, interpretation, and presentation of the initial
results of the exposure analysis technically sound, clearly communicated, and
appropriately characterized?
See comments above.
o The draft risk and exposure assessment evaluates exposures in selected
locations encompassing a variety ofSO2 emission source types in the state of
Missouri; these areas were chosen as an initial case study since 1) air quality
measurements indicated numerous exceedances of 5-minute benchmark values,
2) there are multiple stationary source emissions above 1,000 tons per year,
and 3) there are numerous ambient monitors measuring 5-minute and 1-hour
SO2 concentrations.
The second draft may also evaluate exposures in the remainder of Missouri
and also include areas of Pennsylvania, West Virginia, and other locations
with large SO2 emission sources. What are the views of the panel regarding
the appropriateness of these proposed additional locations and on the
approach used to select them?
The initial case study location selection is reasonable. I suggest only adding
additional locations if the value-added can be defined. The results are a scenario
based on a set of assumptions and local conditions. Are the local conditions as
influential as the assumptions? It may be of more value to spend the modeling
effort evaluating the sensitivity to the assumptions in the current location.
Regardless of the choice, it needs to be stated several times in the document that
the estimates apply to a limited population.
o Do Panel members have comments on the appropriateness and/or relevance
of the populations evaluated in the exposure assessment?
These appear appropriate.
46
-------
o To what extent are the approaches taken to model SO2 emission sources
technically sound and clearly communicated?
I defer some aspects of this response to my colleagues who know about
implementation of AERMOD. With respect to evaluating predictions, the
approach is better than the first draft NOx REA, but there is still room for
improvement. The key consideration for quality of predictions for the purpose of
this analysis is whether the predictions capture the variation observed in the real
world. This includes both the peaks, which are the focus of the exposure
estimation exercise, as well as the full exposure distribution, which is needed as
input for the health risk assessment exercise. Figure 7-3 is a weak test of quality
of predictions, but it is reassuring that the predicted maxima at all locations within
4 km are greater than the observed. Figure 7-4 is less reassuring, even for this
weak test. It suggests most of the predicted distribution underestimates measured
data, even though the peaks are reasonably comparable. I suggest also adding an
evaluation of the AERMOD predictions at the receptor locations where there are
monitors.
o Human exposures were modeled using APEX to simulate the movement of
individuals through different microenvironments. Do Panel members have
comments on the microenvironments modeled?
The modeling of movement of individuals appears appropriate. However the
approach to estimating 5-minute peak concentrations will bias the results
downwards. The distribution of number of peaks in an hour (Table 6-15) should
be used instead of a procedure that sets all other observations in the hour at the
mean. Peaks are likely to be correlated in time and this should also be factored
into a revised algorithm.
Characterization of health risks (Chapters 3,4,5,7,8,9):
The summary of the health evidence from the ISA seems reasonable (chapters 3,
4). Chapter 5 is an introduction to the analyses in the rest of the document and
could be used to lay out criteria for proceeding to later chapters and integrating
interpretation across chapters (comment is particularly relevant to chapters 6-8).
The new chapter 8 results presented at the meeting look promising. It is notable
that while chapter 7 focuses on benchmark values, chapter 8 uses a modeled dose-
response function for the entire range of SO2 concentrations, so it targets different
information in the exposure data. This suggests revisiting the modeling in chapter
7 to assess prediction of exposures below the peaks and adding the focus on the
entire distribution to the chapter discussion. As a second point, the risk
assessment will be sensitive to the assumed functional form of the exposure-
response function. Since the data used to fit the function are so limited, the shape
of this function and its behavior at the low end of the exposure distribution are
highly uncertain. The 3-parameter logistic is particularly problematic. It fits a
47
-------
limiting value parameter (a) which allows for a threshold in the population
response for an exposure concentration that is below 100% (when a
-------
forms a physical barrier that would affect measurements at the monitor
location.) This topic comes up again (105 11-12). Further analysis of the site,
such as discussed here, suggests why this site may be underpredicted. It also
suggests features to assess in other sites in evaluation of predictions.
46 1 17: Is 1.02 the regression coefficient?
55 1 14: Table 6-5
58: Here is an example where the inference about SO2 trends over time is
confused by the trends in monitoring. There needs to be an analysis that
distinguishes these two features. (The current analysis shown in Figure 6-12
is a good attempt, but it doesn't tell the story clearly.) It may be better located
in the ISA.
59 Figure 6-10: Add the number of monitors with no exceedances to the
bottom of this figure. It may be necessary to add the total number of monitors
to this figure as well, unless it can be shown elsewhere (e.g. in Figure 6-9).
Same comment p 69 figure 6-16.
6814: A COV of 0% does not seem reasonable. Some comment or
appropriate data exclusion is warranted.
68 1 6: It does not appear useful to use data from a monitor with only 2 1-hour
SO2 measurements. Also, clarify what "no below detection limit
substitutions" means, particularly with respect to the mention that the dataset
has 0 values.
71 1 4: The discussion regarding "this frequency would only apply" is unclear.
7118-10: Is there evidence that the monitors with exceedances are being
dropped? An analysis of existing data should be able to address this question
so it will become unnecessary to speculate on this point.
11115: Insert at the end of the sentence "when there are exceedances".
Ill Table 6-15: Consider adding a row with 0 exceedances in the table for
perspective (if it doesn't detract from the intent of the table).
1111 20: Somewhere in this chapter there should be a discussion of the
meaning of "peak" since in this chapter it is just high values, not maxima
within a time frame (although the approach used effectively constrains the
number of peaks to not exceed one per hour).
117 1 19: It will be worthwhile mentioning the non-point sources emissions
used for SO2 in an appropriate place in the chapter.
1181 5-6: Explain why the pairing of point sources with local meteorological
stations was done.
1181 13: Figure 7-1.
119 figure: why the random assignment to domains of receptors within
multiple met station domains? Also better clarification of the symbols would
help.
120 1 14: Sentence unclear.
124 114-16: Comment on how much the stack locations moved and if impact
is known, include that as well.
49
-------
128 Table 7-5: Include an overall row with summations or average
percentage, add "predicted" to the emissions for modeling, and merge cells so
the total domain emissions column isn't confusing.
132 1 14-22: Since the exposure modeling in this chapter also needs to do a
good job estimating the entire distribution for the risk assessment in Chapter 8,
the underprediction of 95% of the distribution for AERMOD is troublesome.
More should be done to refine the AERMOD prediction to better capture the
entire distribution.
1431 12,13: SO2
147 7.9: The "health risk characterization" focuses entirely on peaks which
misleads the reader into believing this is the only purpose of the exposure
modeling. This section should be revised to also address the input needed for
chapter 8.
153 1 9-15: There is also the problem of constraining all remaining 5-minute
concentrations to be equal to the mean.
168 19-20: The inference from this planned analysis should be discussed.
Appendix A: I suggest adding the site classification(s).
50
-------
Dr. Frank Speizer
Comments on first draft of Risk and Exposure Assessment SO2.
July 29, 2008
Discussion of Clinical studies: Page 15, line 28. This is taken from the ISA but
there is an inconsistency in the ISA in that the actual study quoted showed 5-13%
of subjects exposed to 0 .2 ppm for 5-10 minutes had significant changes in sRaw
and FEV1 respectively (see figure 4.1 in ISA). Thus to indicate that the effect
level was "...as low as 0 .4-0.6 ppm ..." is misleading. I certainly would not like
to be in the group that dropped my FEV by 15%! This unfortunate statement is
repeated throughout the next section and seems to set a quasi threshold for
consideration of short term effects. This needs to be rethought with the idea of
moving the minimal documented effect down from ">0.4ppm" to 0.2 ppm.
Discussion of the Epidemiological Short term studies. Although the studies are
reasonably accurately reported they tend to ignore the phenomena indicated above.
There are subgroups of individuals that as a class are more sensitive than others to
SO2 and in most of the epi studies these subgroups are not considered. For
example even among asthmatics, which as a group are believed to be more
sensitive, there are individuals not sensitive and those that are. See above only
60% of asthmatics responded to Ippm. Thus in the multicity studies or asthma
ED studies there must be individuals who are not sensitive. As well as those that
are extremely sensitive. So in reporting results as generally positive but not
significant what is really being reported is positive results with wide confidence
intervals generated by the misclassification of the "phenotype" of asthmatics that
lumps together sensitive and non-sensitive subgroups. This needs to be discussed
and if possible factored into the risk assessment calculations.
Section 5.2 From my first comment above it is clear that I believe that staff has
chosen the wrong range for the benchmark analysis. They can go ahead and do
0.4-0.6ppm but they should also do the same analysis for 0.2-0.4ppm, since 13%
of asthmatics are a big number.
Section 5.3: I agree with the plan to obtain more detailed SO2 air data from US
and Canadian authors but isn't that totally impractical based upon the court
ordered deadlines? I suspect that Staff will come back to us claiming they made
the request maybe even got the data but it was too late to incorporate in analyses.
Why not use existing network data to get the distributions out; without tying it
specifically to data used by authors in studies that are now some years old?
Chapter 6 ambient air quality and benchmark health risks for 5 minute peak
exposures.
This is an excellent start. One gets a reasonable "feel" for the available 5 minute
average data. However, I would like to see similar plots for the exceedences of
51
-------
>200ppb as well as these data at >400ppb. Specifically repeat paragraph bottom
of page 56 along with figure 6-9 to 6-14 for >200ppb. I feel rather insistent that
these calculations be done for lower levels and the justification is spelled out on
page 112 section 6.5.9. Staff indicates as is the case that the studies reported for
ethical as well as practical reasons were done on mild-moderate asthmatics. More
severe asthmatics would be more susceptible that these mild asthmatics. Thus,
with 13 %of such asthmatics having a 15% or greater drop in FEV1, at 200ppb it
does not seem justified to start the risk assessment at 400 ppb where over 20% of
mild asthmatics are responsive. This simply is the level of responsiveness that
was measured in the clinical studies and to ignore it would be irresponsible.
Chapter 7: I note that in comparing table 7-2 to table 7-7 that although sites are
designated by name and location in the former in the later they are all designated
as Rural with the largest urban fraction being 17 and 19% and all others 5% or
less. In addition all are air port locations. If this is the case these sights certainly
do not represent population exposures. This could be a serious concern if there
are regional sources located at these airport sites that impact the monitors. Some
discussion, unless I missed it, should be presented on this issue.
The analysis suggested further in Chapter 7 that focuses on the Missouri sites
does a good job of considering the model specification. This seems to work for
these sites, perhaps because as stated on page 131 "all sources in Missouri are
considered rural..." If this becomes the basis for the entire modeling of exposure
something has to be done with longer range transport and more urban sites.
There is a discrepancy between tables 7-7 and 7-8 and table 7-9. Two things in
these tables don't make sense. Perhaps it is a decimal point placement. From the
table ages 1 through 10 gives a total of about 1-2% asthmatics. It also indicates at
the youngest ages Females outnumber males. Most studies I believe would say
the opposite. Secondly in Table 9 and the text above suggest for these same ages
about 10% of the children would be asthmatics (a more reasonable number).
Need to adjust something.
Page 148, Section 7.9.2
This is were the selection of 0.4ppm vs 0.2ppm becomes important as the
following tables show a very substantial differences in number of persons with
exposure above a certain level.
In Chapter 8 again sets the stage with 0.4-0.6 ppm as the risk level. The
discussion of uncertainty needs to include a section on what if the effect level is
lower. (I clearly have indicated that I believe it is) Therefore the discussion
might be turned around and after using the 0.2-0.4 numbers discussion the
variability of response rather than uncertainty of findings.
* / This issue was discussed and resolved in the July 30-31, 2008 meeting. There was no
discrepancy between the tables.
52
-------
Chapter 9: I think it still might be worth considering the fact that many of the
existing epi studies do show positive effects and some way of incorporating the
fact that 15- 60% of asthmatics are responders to SO2 means that these overall
effects that are not significant does not mean they are not positive. Therefore
some risk assessment of the estimated responder populations might be worth
calculating.
Look forward to seeing next draft.
53
-------
Dr. George Thurston
Comments on SOx REA - 1st Draft
Air Quality Information and Analyses (Chapter 6):
1. We have evaluated SO2 air quality throughout the United States, using
all available 5-minute and 1-hour ambient monitoring data for years 1997
through 2007. To what extent are the air quality characterizations and analyses
technically sound, clearly communicated, appropriately characterized, and
relevant to the review of the primary SO2NAAQS?
RESPONSE: The data appear to be the best available for the analyses attempted,
but they need to be subdivided by monitor type, especially source-oriented vs. not
near a major SC*2 source. In addition, the peak-to-mean ratio model (Equation 6-
1) seems overly simplistic, in that it does not implicitly address the variability in
the COV. Instead, the sites are placed in "bins" according to their COV, which
means that a range of COV s are handled similarly. Instead, it would seem that
fitting another model term (dependent on the COV) would be a more appropriate
approach, and might avoid the outliers found when testing the bin model (pages
45-46). In addition, it is not clear to me that the test of goodness of fit is
independent of the original fit.. .is it? Or is the EPA just testing the model on the
fit derived from the same data? The best situation is to develop the model on one
set of data, and test it on another separate set of data. Please clarify which data
were used to fit the model, and which were used to test the fitted model.
2. To what extent are the properties of ambient SO2 appropriately
characterized, including ambient levels, spatial and temporal patterns,
relationships between various averaging times, and the relationship between
ambient SO2 and human exposure?
RESPONSE: This seems to have been accomplished the best that can be done
with the data available for the purpose. However, it would be helpful to sub-
characterize these data as a function of site type (i.e., source oriented vs. other
categories), so as to better understand how the populations most at risk (i.e., near
major SC*2 sources) differ from people located elsewhere.
3. Twenty locations were selected for detailed analyses, using ambient SO 2
monitoring data for years 2002-2006. What are the views of the panel regarding
the appropriateness of these locations, the time period of analysis, and the
approach used to select them?
RESPONSE: Acknowledging that data limitations do exist, it still seems to me
that this analysis should focus on areas where violations are most likely, i.e., in
counties where major point sources exist, such as in and around Jefferson County,
Ohio. Therefore, I think an additional source-oriented criteria should be added to
focus the analyses more on areas where the problem of concern here (i.e., high
54
-------
peak impacts) is most relevant. Moreover, the fact that the Source-oriented
Caribou, ID site is poorly fit (pg. 70) may in fact indicate that the model is not
performing well in situations of greatest interest in this analysis. Finally, the
benchmarks employed are too high, and should be lowered, as even chamber
studies of pure PM have exhibited effects down to 200ppb (e.g., see Figure 4-1 og
the REA), and the animal toxicology indicates that effects are seen at much lower
levels when particles are co-present with the SO2, in agreement with the
epidemiology showing associations at ambient-level short-term SC>2 (e.g., Peel et
al, 2005).
Thus, the last sentence on page 112 should instead read something more
like: "Therefore, the potential health effect benchmarks based on these clinical
studies likely underestimate risks in the general population because people in the
general population with greater susceptibility are considered, and the exacerbating
effects of co-present ambient paniculate matter are also not considered in such a
limited analysis of risk.
4. In order to simulate just meeting either the current 24-hour or annual
standards, staff adjusted SO 2 air quality levels for the years 2002-2006 upwards
in all but one location. Ambient monitoring data in North Hampton County PA
were above the 24-hour standard in the year 2006 and were therefore adjusted
downward. To what extent is the approach taken technically sound, clearly
communicated, and appropriately characterized?
RESPONSE: Yes, this seems a reasonable approach to estimation.
5. What are the views of the Panel regarding the adequacy of the assessment of
uncertainty and variability?
RESPONSE: Again, it is not clear to me that the uncertainty analysis considers a
dataset distinct from the data used to develop the model in the first place (e.g., in
the accuracy estimation on page 105). Please clarify this in the text. Moreover, I
feel that a quantitative risk assessment based upon the SO2 epidemiological
studies (e.g. of respiratory ED visits) could provide a useful input to the
uncertainty analysis. Since the number of people ending up at the Emergency
Department should be a small subset of the number of people experiencing
bronco-constriction, a quantitative estimate of the SC>2 associated ED visits would
provide a check on whether the number of people affected are being
underestimated. Such an analysis would be a useful addition to the assessment of
uncertainty.
Exposure Analysis (Chapters 2, 7):
1. To what extent is the assessment, interpretation, and presentation of the initial
results of the exposure analysis technically sound, clearly communicated, and
appropriately characterized?
55
-------
RESPONSE: None on this aspect.
2. The draft risk and exposure assessment evaluates exposures in selected
locations encompassing a variety ofSO2 emission source types in the state of
Missouri; these areas were chosen as an initial case study since 1) air quality
measurements indicated numerous exceedances of 5-minute benchmark values, 2)
there are multiple stationary source emissions above 1,000 tons per year, and 3)
there are numerous ambient monitors measuring 5-minute and 1-hour SO2
concentrations. The second draft may also evaluate exposures in the remainder
of Missouri and also include areas of Pennsylvania, West Virginia, and other
locations with large SO2 emission sources. What are the views of the panel
regarding the appropriateness of these proposed additional locations and on the
approach used to select them?
RESPONSE: I'd like to see more analyses of locations near major SO2 sources,
like power plants, in counties such as Jefferson County, OH, and surrounding
counties.
3. Do Panel members have comments on the appropriateness and/or relevance of
the populations evaluated in the exposure assessment?
RESPONSE: The populations considered (i.e., asthmatics) may be too narrow for
the standard-setting process, which would lead to small estimated numbers of
people affected. Rather than clinical studies of subset populations, and only to
pure SC>2, the application of epidemiology-based risk factors would provide a
greater relevance to the general population. At a minimum, risk estimates (e.g. of
ED visits) should be conducted based on epidemiology in order to determime if
the population considered is plausible. The number of SC>2 induced respiratory
ED visits should be a subset of the population at risk, so if the epidemiology gives
larger numbers, that would provide a test as to whether the full population at risk
has been considered, or not.
As to the asthmatic populations to be considered, it may be important to
consider that the people in the general public who have regular anti-inlammatory
medications prescribed by their physician may represent an especially affected
subpopulation, especially on a day when they have not taken their prescribed
medication.
4. To what extent are the approaches taken to model SO2 emission sources
technically sound and clearly communicated?
RESPONSE: Appears to be state-of-the-art and well explained. The dependence
on likely incomplete source emissions inventories is a potential weakness.
5. Human exposures were modeled using APEX to simulate the movement of
individuals through different microenvironments. Do Panel members have
comments on the microenvironments modeled?
56
-------
RESPONSE: No.
Characterization of Health Risks (Chapters 3, 4, 5, 6, 7, 8, 9):
1. What are the views of the Panel on the overall characterization of the health
evidence for SO 2? Is this presentation clear and appropriately balanced?
RESPONSE: Section 8.2.3 needs to clearly point out that these studies are for
pure SC>2 only, and do not fully represent conditions in the real world, as the SC>2
interactions with PM (that is always present in the real world) are not considered.
This may well lower the levels at which the symptoms noted can be experienced.
The mechanism for this is likely that the particles provide a vector for sulfur
oxides to be transported deeper into the lung in solution and as reactant products.
Indeed, Chen et al. (1992) have revealed that approximately 10 times as
much pure sulfuric acid (H^SO^ is required to give the same lung airway hyper-
sensitivity effects in guinea pigs as when the acid aerosol is present as a surface
coating on a particle (200 wg/m3 H2SO4 mist vs. 20 z/g/m3 H2SO4 when surface
coated on a particle). (Chen LC, Miller PD, Amdur MO, Gordon T. (1992).
Airway hyperresponsiveness in guinea pigs exposed to acid-coated ultrafme
particles. J Toxicol Environ Health. Mar;35(3): 165-74.) Sulfuric acid is one
potential surface reactant product of SC>2 and particle-surface reactions. Thus, it
might well be possible that, in the real world where particles are always co-
present, the acute effects noted with pure SC>2 at 200 ppb may well be
experienced at much lower SC>2 exposure concentrations, and this should be
considered here and throughout this document. Moreover, it might be argued by
some that this effect is covered by the PM standard, but the very acute effects
considered here are associated with SC>2, and, also, there is no one-hour or 5-
minute PM standard, so even though PM co-presence is apparently involved in
exacerbating the impact of SC>2, the effects under consideration here are
something very distinct from the longer averaging time PM effects controlled by
that standard, and must be considered in this document and SC>2 standard-setting
process.
2. The characterization of health risks focuses on potential health benchmark
values identified from the experimental SO2 human exposure literature on lung
function with accompanying respiratory symptoms. What are the views of the
Panel on using potential health benchmarks from this literature to characterize
health risks?
RESPONSE: Such controlled clinical studies of pure compounds are very
important for proof of concept and for evaluating biological plausibility, but not
for risk assessment as proposed here. Epidemiological studies should be applied
for that process, as they consider real people in real world situations.
3. Do panel members have comments on the range of potential health effects
57
-------
benchmark values chosen to characterize risks associated with 5-minute SO2
exposures?
RESPONSE: The benchmarks selected are too high. First, there are effects
documented in the pure SC>2 clinical exposure studies at levels at least down to
200 ppb. Second, the exacerbating effects of co-exposure to PM on the health
impacts of SC>2 exposure in the real world is ignored: it is plausible that co-
exposure to PM will cause these effects at much lower levels than indicated by the
clinical exposures to pure SC>2 alone. Finally, the epidemiology concur with this
point in that they show associations that the ISA finds sufficient to infer a causal
relationship (see page 19 of the REA).
4. To what extent is the assessment, interpretation, and presentation of initial risk
characterization results technically sound, clearly communicated, and
appropriately characterized?
RESPONSE: I find the assessments based on the clinical studies to be
insufficient alone, as the full extent of effects in the real world on a wider
distribution of the population cannot be fully incorporated using this approach.
Epidemiology should also be applied in this process, if only as a quality control
check on the plausibility of the size of populations affected. As noted above, a
comparison of clinical study-based estimates vs. epidemiology-based estimates
would provide at least a check on the plausibility of the population estimated as
affected based on the clinical studies that are based on pure SC>2 alone.
5. The epidemiology literature will be used to qualitatively characterize SO2-
related health risks for health outcomes such as respiratory symptoms and
emergency department visits and hospital admissions for respiratory-related
causes. However, staff has judged that it is not appropriate to use the available
SO 2 epidemiological studies as the basis for a quantitative risk assessment in this
review. Do panel members have comments on this judgment and/or on the
rationale presented to support it?
RESPONSE: While this is a worthwhile analysis to conduct, I feel strongly that
the epidemiological studies of SC>2 can and should also be used to conduct a
quantitative risk assessment, if only as a check on the clinically-based estimates.
Furthermore, I don't think that looking at the correlation of percentile SC>2
concentrations vs. statistical significance is a worthwhile, or very meaningful,
exercise. Too many other variables (such as power) enter into the determination
of statistical significance for this to be a meaningful exercise. The EPA should
move forward with a quantitative risk assessment based on the epidemiological
studies available, albeit noting the uncertainties and limitations, in order to
provide a fuller and more relevant risk assessment than allowed via relying only
upon the clinical studies-benchmark approach for quantitation, as they propose in
this draft document.
58
-------
Dr. James Ultman
SOx REA Comments
General Comments on the Document
The first draft of the REA clearly puts forth the susceptible population (i.e.
asthmatics) and health effects (i.e., clinically-observed symptoms and lung
function decrements) that will be the focus of the health risk assessment. It is also
is evident that this REA will extend previous assessments by a detailed analysis of
the consequences of short-term and peak exposures under alternative forms and
levels of the NAAQS.
Chapter 6.
General
Because of the limited number of monitoring data on peak exposures, staff has
developed an imaginative but previously-unvalidated stoichastic method to
extrapolate from short-term hourly exposure data to peak exposure concentrations.
The rationale for the method is that "..the temporal and spatial pattern in SO2
source emissions is influenced by the type(s) of sources and its operating
conditions and that this emission pattern(s) will be reflected in the ambient SO2
concentration distribution measured at the monitor." Based on this rationale, the
coefficient of variation (COV) of 1-hour exposure measurements is used as a
predictor of the peak-to-mean ratio (PMR) of the hourly measurements.
The selection of COV as a predictor variable is justified by analyses of the data
from 98 monitors where co-localized peak and hourly averaged SO2
concentrations. These analyses include: the linear correlation of the COV's of 5-
minute samples with the COV's of 1-hour samples (figure 6-2); the convergence
of the predicted PMR values to the measured PMR values (figure 6-5); and a
comparison of the mean predicted PMR value to the measured PMR at each
monitor.
The latter analysis is presented as a test of the "accuracy" of the PMR estimation
method. Since the measured values used to evaluate the method is the same data
set used to obtain the cumulative distribution functions (CDF) used in the
simulations, this analysis does not validate the method. It would have been better,
in theory, to divide the 98 monitors into two subsets—one subset for determining
the CDF and another for validating the method.
Even after reading appendix A, I find the details for the many algebraic
computations performed in this chapter hard to follow (e.g., see lines 6-12). Such
computations would be more transparent if they were presented as equations, or
even better, supported by idealized graphs that showed how a (hypothetical)
59
-------
concentration-time trace from a monitor was averaged over 5-minute and 1-hour
intervals that were then averaged together, etc.
I also find it hard to follow the progression of analyses in sections 6.4 and 6.5. It
appears that the "as is" analysis of exceedances above the health effects
benchmarks is obtained from the full 98-monitor data set, whereas the "just
meeting the current standard" analysis of exceedances is obtained from the 20-
county data set. If this is indeed the case, then is would be inconsistent to
compare the two analyses. To avoid such confusion, the chapter would benefit
from a more informative introduction, either at the beginning of the chapter or at
the beginning of the major sections.
Specific Comments
Page; lines
33; 11 Spatial siting of monitors should, in principle, impact both horizontal as
well as vertical distances from point sources. Are the distributions of vertical
distances of the 98 monitors upon which the PMR method is based similar to the
vertical distances at which all 1-hour monitors are placed?
36; 12 This "model" equation gives the impression that PMR is a parameter. In
fact, PMR depends on Ci_hOUr- It might be better to write the equation as a
definition of PMR.
39; 8 Does the Thompson reference provide validatation the stoichastic approach
used in the current document?
42; 3-5 I don't see why these results are "consistent" with each other. Perhaps,
more explanation is needed.
46; 11 I wouldn't say that the table entries exhibit "good agreement."
Responses to the Charge Questions - Air Quality Information and Analysis
1. The clarity and flow of the many analyses in this chapter could be significantly
improved. The mechanics of a particular analysis are not always clear. Moreover,
the relationship among the many analyses is hard to follow. With respect to the
technical aspects of the chapter, I feel that additional thought needs to be given to
validating the PMR estimation procedure.
2. There is an abundance of basic numerical information in the chapter, but at
some point, it needs to be distilled into a set of more easily appreciated
observations and conclusions.
3. No comment.
60
-------
4. It was not clear to me, from the contents of this chapter, how the roll-up factors
determined in 20 selected counties will be applied to the exposure and health risk
assessment on a national level.
5. The primary source of uncertainty is the lack of validation of the PMR
methodology.
61
-------
Dr. Ronald Wyzga
Comments on SOx REA - 1st Draft
Overall Comments: It would have been helpful to have had more time to review
this document. It is a very lengthy and complex document. Given the available
time to review it, my review is at best cursory.
By and large, I find the approach taken in this document to be reasonable. The
assessment focuses upon short exposures to asthmatics, which I believe to be the
key issue for SO2.
Charge Questions for Exposure Analysis:
1. To what extent is the assessment, interpretation, and presentation of the
initial results of the exposure analysis technically sound, clearly
communicated, and appropriately characterized?
My review is cursory, but at first glance it appears to be technically sound and
appropriately characterized.
2. The draft risk and exposure assessment evaluates exposures in selected
locations encompassing a variety of SO2 emission source types in the state
of Missouri: these areas were chosen as an initial case study since 1) air
quality measurements indicated numerous exceedances of 5-minute
benchmark values, 2) there are numerous ambient monitors measuring 5-
minute and 1-hour SO2 concentrations. The second draft may also
evaluate exposures in the remainder of Missouri and also include areas of
Pennsylvania, West Virginia, and other locations with large SO2 emission
sources. What are the views of the panel regarding the appropriateness of
these proposed additional locations and on the approach used to select
them?
I agree that attention should be given to those areas where there are exceedances
presently and where there are major SO2 sources. Given its size, particular
attention should be given to Allegheny County (Pittsburgh), Pa. If there is a
tradeoff in resources between extent of detail in estimating exposures and the
number of areas studied, I would favor emphasis on the former. I think Missouri,
Pennsylvania, West Virginia, and possibly Ohio would provide a good
understanding of the risks in states where exposures are above average. If there
are any remaining instances of high exposures associated with smelter operations,
these might be considered as well.
62
-------
3. Do Panel members have comments on the appropriateness and/or
relevance of the populations evaluated in the exposure assessment?
I would agree with the focus on asthmatics. I note the apparent discontinuities in
some asthma prevalence rates in Table 7-7; can these be verified? It would be
useful to obtain data on the relative number of asthmatics who are routinely
medicated as this group does not appear to respond to peak SO2 exposures.
4. To what extent are the approaches taken to model SO2 emission sources
technically sound and clearly communicated?
Given my limited expertise in the use of air quality models, I leave it to my
colleagues to judge this issue.
5. Human exposures were modeled using APEX to simulate the movement
of individuals through difference microenvironments. Do Panel members
have comments on the microenvironments modeled?
The APEX model is well-suited for exposure analyses to be undertaken here. My
only question about microenvironments is whether roadside exposures should be
considered. I have been involved in some studies which suggest that meaningful
exposures to SO2 can occur from sulfur-containing diesel fuels (which are being
phased out); if this is correct, near-roadway exposures, could be higher. On the
other hand since this source of SO2 is being curtailed significantly, it could be
fruitless to consider this source in future regulatory scenarios.
Specific comments:
p. 13,11. 26-28: Should special note be made about the amount of time spent
indoors as indoor exposures are negligible except in the rare cases where there are
indoor sources.
p. 15,1. 22: insert "exercising" before "asthmatics".
11. 28 and follows: should a comment be made that exercising asthmatics
who are medicated do not appear to respond to SO2 in human clinical studies.
p. 34,1. 8: "hydroelectric"???
1. 19: replace "is" with "are".
p. 36,1. 12: Is this equation too simple? Do we need to consider wind direction?
p. 37,1. 16: Which 6 states?
p. 42,11. 18-25:1 wonder if there are better ways to do this, by considering the
proximity of monitors to sources and/or considering such factors as wind speed. I
would be interested in more details about the distribution of COVs as well.
63
-------
p. 47,11-17-18: In general I worry about the communication of this scenario; it
can be very misleading when so few areas exceed the current standard. I hope
this scenario is well-caveated.
p. 54:1 wonder if it useful to consider adjustments based upon the annual average
concentration given the uncertainty associated with the relationship between
short-term concentrations and the annual average.
p. 55,11. 12-13: Is there also a statistically significant trend?
p. 71,1. 3: "than"
p. 135, Table 7-7: There are some curious discontinuities in the prevalence rates
by age, especially for males; see the differences between age 3 and 4, 4 and 5, and
16 and 17. Are these numbers correct?
p. 145,11. 14-17: Are there any data to suggest some consideration of near
roadway exposures? Sulfur in diesel fuel may have influenced such exposures in
the past.
p. 158,1.4: "Introduction"
p. 165: Medication use could be another category of uncertainty.
64
------- |