UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
                                  WASHINGTON D.C. 20460
                                                              OFFICE OF THE ADMINISTRATOR
                                                               SCIENCE ADVISORY BOARD

                                      January 9, 2008

EPA-CASAC-08-006

Honorable Stephen L. Johnson
Administrator
U.S. Environmental Protection Agency
1200 Pennsylvania Avenue, NW
Washington, DC 20460

       Subject:      Clean Air Scientific Advisory Committee's (CASAC) Consultation on
                   EPA's Sulfur Dioxide Health Assessment Plan: Scope and Methods for
                   Exposure and Risk Assessment (November 2007 Draft)

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, met on
December 6, 2007 for a consultation on EPA's Sulfur Dioxide Health Assessment Plan: Scope
and Methods for Exposure and Risk Assessment (November 2007 Draft). The CASAC uses a
consultation as a mechanism for individual technical experts to provide comments to guide the
Agency on technical issues early in the development of a document, before the first draft is ready
for peer review. Panel members offered oral comments at the meeting as well as written
comments (attached to this letter). This CASAC consultation, like all CASAC consultations,
was conducted under the requirements of the Federal Advisory Committee Act, which include
advance notice of the public meeting in the Federal Register.

       There will be no formal report from the CASAC as a result of this consultation, nor do
we expect any formal response from the Agency. CASAC offers the attached individual
comments for the Agency to consider as it moves forward with his Health Assessment for Sulfur
Dioxide.

-------
       We look forward to conducting a peer review of the first draft for the Exposure and Risk
Assessment document as part of the CASAC's continuing role in the National Ambient Air
Quality Standard review.
                           Sincerely,

                                 /Signed/

                           Dr. Rogene Henderson, Chair
                           Clean Air Scientific Advisory Committee
Attachments

Attachment A: Roster of CASAC Sulfur Oxides Primary NAAQS Review Panel
Attachment B: Compilation of Individual Panel Member Comments on EPA's Sulfur Dioxide Health
Assessment Plan: Scope and Methods for Exposure and Risk Assessment (Draft, November 2007)

-------
                                                                         Attachment 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

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, Associate 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

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.

-------
                                                                   Attachment B
Comments from CAS AC Sulfur Oxides Primary NAAQS Review Panel on EPA's Draft Sulfur
Dioxide Health Assessment Plan: Scope and Methods for Exposure and Risk Analysis
(November 2007)
Comments from Dr. Cowling	2
Comments from Dr. Crawford-Brown	7
Comments from Dr. Schlesinger	11
Comments from Dr. Seigneur	12
Comments from Dr. Frank Speizer	13
Comments from Dr. Wyzga	17
Comments from Dr. Larson	19
Comments from Dr. Thurston	22
Comments from Dr. Kenski	23
Comments from Dr. Gordon	25
Comments from Dr. Hattis	26
Comments from Dr. Kinney	37
Comments from Dr. Russell	40
Comments from Dr. Samet	46
Comments from Dr. Ultmann	48
Comments from Dr. Pinkerton	49
Comments from Dr. Sheppard	51
Comments from Dr. Balmes	58

-------
Comments from Dr. Cowling
My comments are organized below in response to each of the several Charge Questions posed in
Karen Martin's November 2007 transmittal letter for Lydia Wegman to Holly Stallworth.

Air Quality Considerations:
1. Based on the low estimated contribution of policy-relevant background SChto overall
ambient SCh levels, staff is considering a proportional (i.e., linear) approach to adjusting
air quality to simulate just meeting potential alternative SCh standards that are below
recent air quality concentrations. Do the Panel members have comments on adopting a
proportional approach to simulate just meeting more stringent alternative air quality
standards?
      Such a proportional approach seems very sensible to me

2. Recognizing that current ambient air quality concentrations are lower than the current
standards, the draft Health Assessment Plan discusses two alternative approaches to
simulating ambient SCh levels associated with just meeting the current SCh standards: use
of historical air quality data (e.g., possibly pre-2000) when ambient levels were at or above
the current standards, or use of a proportional (i.e., linear) approach to adjust SCh levels
upward. Do the Panel members have advice or comments on these two alternative
approaches to simulating air quality just meeting the current SCh standards?
      Being a student of history I would favor using historical data to simulate air quality
      parameters that just meet the current standards.

Exposure Analysis:
1. In considering the exposure analysis broadly:
a. Do Panel members have any comments on the general structure and overall two-tier
approach that staff plans to use for the exposure analysis? Are the criteria that staff plans
to use for deciding whether to conduct a Tier II analysis clear and appropriate?
      The description of the two-tier approach is outstandingly clear as presented in this
      document. Unfortunately, however, I have no personal experience on which to base an
      informed judgment about the issue of appropriateness of the two-tier approach.

b. Have the most important factors influencing exposure to SCh been clearly accounted for
and described?
      Yes. It appears to me that the important factors influencing exposure have been
      accounted for and have been described clearly.

c. The draft plan describes the basis for and selection of population groups  of interest (i.e.,
children, asthmatics  (children and adults), and the elderly) for which SCh exposure
estimates  are to be developed. Do Panel members generally agree with the groups of
interest identified in  the draft plan?
      I certainly agree with the population groups (of interest or concern) that have been
      identified in the draft plan.

-------
2. In considering the Tier I exposure assessment:
a. Do Panel members agree that a statistical model using available ambient 5-minute
monitoring data is appropriate for estimating expected exceedances of very short-term (5-
minute) potential health effect benchmarks?
       I have no experience on which to base an informed judgment in response to this question.

b. Do Panel members agree with the approach of applying a statistical model to estimate 5-
minute concentration exceedances at monitoring locations where only 1-hour monitoring
was performed for evaluating the extent of 5-minute peaks associated with meeting
alternative standards with longer averaging times?
       I presume there is an adequate body of measurement data where both 5-minute and 1-
       hour measurements have been made in various locations across this country, and that the
       correlations between these parallel measurements can provide an adequate basis for
       developing a statistical model of reasonable reliability.  If such parallel data sets are not
       available, or if correlations between 5-minte and 1-hour data are highly variable,
       however, it seems risky to use a statistical approach of indeterminate reliability.  See
       Checklist question 3 in the "Guidelines for Formulation of Statements of Scientific
       Findings to be Used for Policy Purposes:"
       3) IS THE DEGREE OF CERTAINTY OR UNCERTAINTY OF THE STATEMENT
       INDICATED CLEARLY? Have appropriate statistical tests been applied to the data used in
       drawing the conclusion set forth in the statement? If the statement is based on a mathematical or
       novel conceptual model, has the model or concept been validated? Does the statement describe
       the model or concept on which it is based and the degree of validity of that model or concept?

3. In considering a potential Tier II exposure assessment:
a. Do Panel members agree with the combined emissions/dispersion modeling approach to
estimate short-term (hourly) SCh concentrations in close proximity to SCh emission
sources?
       I have no experience on which to base an informed judgment in response to this question.

b. Do Panel members have comments or advice regarding the described binning of sources
and development of prototype stacks/facilities?
       I have no experience on which to base an informed judgment in response to this question.

-------
c. Do Panel members agree with the approach using peak-to-mean ratio cumulative density
functions (PMR CDFs) to estimate very short-term peak concentrations from the 1-hour
modeled concentrations?
       I have no experience on which to base an informed judgment in response to this question.

d. Do Panel members generally agree that the approach described using APEX is
reasonable and appropriate to estimate the occurrence of very short-term (5 minute) SCh
peak exposures?
       I have no experience on which to base an informed judgment in response to this question.

4. Do Panel members have any comments or advice regarding the general approach to
addressing uncertainty and variability in each Tier of the exposure assessment as described
in the draft plan?
       I find the description of the general approach to addressing uncertainty and variability in
       each Tier of the exposure assessment very clear as presented in the draft plan.

Health Risk Assessment:
1. Do Panel members have any comments on the general structure and overall three-tier
approach that staff plans to use for the risk assessment? Are the criteria that staff plans to
use for deciding whether to conduct a Tier III risk assessment clear and appropriate?
       The description of the three-tier approach is outstandingly clear as presented in this
       document.  Unfortunately, however, I have no experience on which to base  an informed
      judgment about the issue of appropriateness of the three-tier approach.

2. In considering the Tier I risk assessment:
a. Do Panel members agree with the approach of having a qualitative assessment of health
endpoints to identify which are likely candidates for a more sophisticated and  quantitative
tier of assessment?
       Although it seems reasonable that a qualitative assessment of health endpoints might be
       used to identify likely candidates for a more sophisticated and quantitative tier of
       assessment. As indicated earlier, however, I have no personal experience on which to
       base an informed judgment about the use of qualitative assessments in making choices
       about quantitative tier assessments.

b. Do Panel members agree with our initial observation that controlled human exposure
studies demonstrate strong evidence for bronchoconstriction in exercising asthmatics
following 5-10 minutes SCh exposure?
       I have no experience on which to base an informed judgment in response to this question.
c. Do Panel members agree with staffs initial observation that the strongest epidemiologic
evidence is for respiratory symptoms in asthmatic children and  respiratory-related
hospital admissions and respiratory-related emergency department visits in asthmatics and
others with respiratory conditions?
      I have no experience on which to base an informed judgment in response to this question.

-------
3. In considering the Tier II risk assessment:
a. In general, are staff plans to use potential health effect benchmarks to address
respiratory effects demonstrated in exercising asthmatics in controlled human exposure
studies clear and appropriate?
      I have no experience on which to base an informed judgment in response to this question.

b. Do Panel members generally agree with the tentatively identified potential health effect
benchmark of 0.5 to 0.6 ppm for exercising asthmatics following 5-10 minutes SCh
exposure?
      I have no experience on which to base an informed judgment in response to this question.

c. Do Panel members generally agree with the staff's approach of focusing on areas around
major sources of SCh with respect to concerns about 5-10 minute peak exposures related to
the respiratory effects observed in controlled human exposure studies?
      Yes, this approach seems very reasonable to me.

d. Do Panel members generally agree with staffs approach of focusing on urban areas with
respect to concerns about 1- and 24-hr and annual SCh concentrations related to
respiratory effects observed in epidemiologic studies?
      I have some misgivings about focusing so strongly on SO2 concentrations in urban areas
      that people (including both susceptible and vulnerable populations in other regions with
      somewhat higher exposures (such as the Pacific Northwest, Hawaii and Alaska) may be
      short-changed in the planned assessment processes.

e. Do Panel members have any comments or advice with respect to staffs approach of
gathering additional information to characterize the SCh ambient air quality that existed at
the time various key U.S. and Canadian studies addressing respiratory effects were
conducted to see if the concentration-response relationships observed in these
epidemiologic studies are related to particular SCh levels and associated averaging times,
geographic location and/or  season, and the inclusion of various copollutants?
      I have no experience on which to base an informed judgment in response to this question.

-------
4. In considering a potential Tier III risk assessment:
a. Do Panel members generally agree that there is insufficient information to develop
credible exposure-response relationships for use in a quantitative risk assessment based on
the controlled human exposure evidence?
       I have no experience on which to base an informed judgment in response to this question.

b. Do Panel members have any comments or advice with respect to the general approach or
specific factors to be considered in deciding whether or not to proceed to a Tier III
quantitative risk assessment for the respiratory-related health endpoints based on
epidemiologic evidence discussed in the draft plan?
       I have no experience on which to base an informed judgment in response to this question.

5. Do Panel members have any comments or advice with respect to the general approach to
addressing uncertainty and variability in each Tier of the risk assessment as described in
the draft plan?
       I am very impressed with the clarity of presentation of the general approaches described
       in this draft plan for addressing uncertainty and variability in each Tier of the proposed
       risk assessment! At the same time, however, I can only trust that skill in development of
       written descriptions is not a cover for lack of skill within the assessment team for
       drawing of appropriate scientific inferences from analysis of available data and
       information!

-------
Comments from Dr. Crawford-Brown
My overall impression of the document is that it presents a reasonable path forward on assessing
risk in a topical area (sulfur dioxides) where the data are somewhat sparse. However, it took me
several readings to piece together the structure of the assessment process due to poor
organization of the document. The key point of confusion was that the authors describe a process
that is different for effects that have been studied through clinical trials and ones that have been
studied through epidemiological results. This produces in two different ways of characterizing
risk for these two categories of effects, and two different levels of detail in the characterization.
This separation - or at least the basis for it - is not made evident, however, until late in the
document, and so the reader is left partially confused in the first two thirds. I kept sensing that
there were two streams of thought and assessment at play, but never had a concrete statement of
that until late in the document. The writing overall needs to be improved.

This improvement is needed both to  clarify the issue that there are two kinds of assessments (for
the two kinds of data) and to explain how precisely the epidemiological data are to be assessed.
Despite several readings, I cannot understand what they  intend to do with the epidemiological
results. For the clinical results, they clearly intend a modified form of hazard quotient or margin
of exposure. That will result in a quantitative measure of at least hazard. But in the case of the
epidemiological results, the authors talk repeatedly of "looking for patterns" in the
concentration-response data, with no indication of what they mean by "patterns". I suppose  they
might mean looking for changes in the slope factor or some other risk summary as one moves
across studies of concentration-response in different geographical settings and populations, but
there is never a succinct statement as to what they mean by a "pattern" so I remain unsure. And it
is not at all clear what they intend to do with such patterns, or even what the measure of hazard
or risk will be at the end of the day. I see no reason why it cannot be a benchmark health effect
as in the clinical trials, with a  similar calculation of hazard quotient or margin of exposure. This
part of the document needs to be significantly improved.

I have a series of more specific comments:

1.  The first full  paragraph on page 2  appears to indicate that the  focus will be on short-term
exposures that take place in close proximity to local source emissions. The rest of the document,
however, does not appear to narrow the focus so tightly. This confusion needs  to be clarified.

2.  The tiering system is OK but still somewhat confusing to read. The first problem is that it is
not clear what specific results from a first tier would send the assessors to the second tier (or
what results would prevent them from going there).  Some VERY loose criteria are mentioned
(e.g. on Page 36), but these  are quite generic and the real question is what kinds of answers to
these issues would  constitute staying in a tier or advancing.

The second problem is that there are two tiers for exposure and three for risk. I suppose the
authors intend that the exposure assessment could proceed to a different tier than the risk
assessment (yielding 6 cells in a 2  x 3 table), but that seems to me an unwarranted approach.

-------
Better to have just two tiers that apply to the entire process. I see no merit to having, for
example, a tier 2 exposure assessment and then a tier 1 risk assessment. The tier 2 exposure
assessment would contain a level of detail that could not be met by the concentration-response
part of the assessment.

Finally, tier  1 of the risk assessment is not a risk assessment at all. It is a hazard identification. I
presume one would need to do a hazard identification as prelude to the assessment process.
Overall, then, I recommend just two tiers for the entire assessment rather than this system of
separate categories of tiers for separate parts of the assessment.

3. On Pages 5 and 6,1 am generally supportive of the approach mentioned for proportional "roll
down" of concentrations, so long as one can assume that control  strategies really would affect all
geographic areas equally (which I doubt, but the error introduced will not affect the fraction of
population at or near a benchmark health effect level). But I am not sure about the utility of a
"roll up" procedure based on the historical data, since it is not clear to me how one would
determine which particular past historical data are most representative of what conditions will be
like overall once a new regulation is in place. I'm not saying the  idea is intrinsically wrong, only
that I don't know how it would be executed.

4. On Page 7, the authors speak of the "relative degree of confidence". I have no idea what this
means. In the same paragraph, they refer to a criterion (for moving to a more detailed and
quantitative  uncertainty analysis) if such an analysis adds "value". No coherent explanation of
what "value" means in this instance is given,  either here or later in the document. It might mean
either that it better informs the decision or that it leads to different regulatory results. In the latter
case, however, I don't see any discussion of how uncertainty relates to any kinds of decisions
that might be made, and so it is not clear how one is to decide "value" in this utilitarian sense.

5. Beginning on Page  8,1 began to have a problem with understanding the role of the tiers of the
assessment.  At first, I  thought tier 1 might be a kind of screening assessment in which the
assessor is asking: If I make several simplifying assumptions that tend to all overstate the risk, do
I see any evidence of a significant risk? If yes, I will go to tier 2.  If no, there is no need for me to
proceed with any more detailed analysis.

But then the document describes the choice of moving to tier 2 as being related to the availability
of data, and  not to any specific results one sees from tier 1.  So tier 1 does not seem to be  getting
used as a screening tool. I can't understand why one would even do tier 1 if the data are available
for  tier 2.

6.1 am supportive of the use of PMR values to get at the short-term exposures in the geographic
areas where only longer-term averages are available. I presume the assessors will develop CDFs
for  PMRs under different conditions (near sources, away from  sources,  etc) and apply the
appropriate CDFs to non-monitored areas. The document at least hints at this, even if it is not
expressed well. An example is on Page  10, where the bulleted list evidently applies to this issue,
but the reader is not told why these four bulleted issues are being presented, or how their answers
would affect the development of and application of CDFs.

-------
7. On Page 13, the authors describe (at the bottom) an issue of 10 or 15 minute averages. It is not
clear if this is to be a rolling average from the 5 minute predictions, or whether new PMR CDFs
would be developed starting with the original monitoring data.

8. On Page 14, the authors appear (at the bottom) to be saying that measurement error is small
compared to other sources of uncertainty. I would in general agree, but there will need to be
some evidence of this before this source of uncertainty is ignored.

9. In several places, including on page 15, the authors mention a kind of sensitivity analysis to be
performed, and then state that they will determine whether a given parameter or term does or
does not contribute to uncertainty. All parameters and terms and models contribute to
uncertainty. I assume they mean something like "contribute significantly".

10.1 was not sure how results less than the MDL or MQL will be factored into the analysis of
PMR distributions. Perhaps only results above the MDL or MQL will be used?

11.1 am assuming that uncertainty factors will not be incorporated into any Health Benchmarks
used.  If they are, then this will need to be reflected in the uncertainty analyses.

12. There is a very general issue I want to raise concerning the incorporation of activity levels in
the assessment. To the extent the clinical data are used, this makes sense,  since the effects at a
given concentration are tied to activity level.  So it will be necessary to estimate the activity level
of an  individual in the exposed population to determine which clinical exposure-response curve
to use. But for the epidemiological results, variations in activity level are already hidden inside
the slope factors. In fact, the slope factors at low exposures probably are driven by the fraction of
people who are both sensitive and exercising in a population at the time the study was done.  So it
might not be appropriate to do a detailed exposure assessment, complete with  inter-subject
variability of exercise patterns, and then apply the slope factor or other risk summary from an
epidemiological study to all exposed individuals regardless of activity level. Having said this,
however, I am not sure the authors intend to do this anyway, since I cannot understand from  the
document HOW they intend to use the epidemiological results.

13. There are two ways to use the monitoring results for air measurements in conjunction with
dispersion models. One is to calibrate the models to the data. The other is to use the data in a
model-to-monitor comparison for purposes of uncertainty analyses. The authors appear to be
leaning towards the latter, but this isn't stated clearly.  In any event, I would prefer the former.

14. On Page 19,1 assume the modeling will allow for  overlap of plumes from  multiple sources in
a geographic area. This isn't stated.

15.1 am generally supportive of the use of APEX. The one caveat I would apply here is that  this
may be more detail than is justified by the concentration-response results. And it will be difficult
to defend the idea that any resulting PDFs of exposure reflect actual exposures on time periods as
short  as an hour or less. This is an area of assessment in which the uncertainties are very large
due to the extreme variation of an individual's activities during a day. It will be important to

-------
present the assessment as a scenario analysis of representative exposures in a hypothetical (but
reasonable) population, and not as an accurate representation of actual exposures to individuals.

16. On Page 24, the authors describe the use of a national average for asthma prevalence rates.
But if this were valid, it would imply that these rates don't depend on geographic location, which
would in turn imply that they don't depend on levels of exposure to air pollutants, which seems
to go in the face of the basis for many of the NAAQS standards. I am not saying this is a bad
approximation, or even the best that can be done, but it does lead to a logical inconsistency.

17. On page 24, the authors use a phrase that appears often in the document: ".. .assessment
would take into account...". I agree with the sentiment, but no guidance is given as to HOW or
IN WHAT SENSE something will be taken into account.

18. On Page 25, the authors mention sensitivity analysis. I support the performance of such an
analysis,  but a decision must be made as to whether it will be a local SA  (adjusting one
parameter at a time) or a global SA (adjusting multiple parameters and looking at contribution to
variance).

19. On Page 25, the issue is again raised of comparing model results to monitors. The problem
here (which also appeared in NATA) is that a model may get a peak value correct but have it
shifted slightly in  space. So if one simply compares model results at a point against monitor
results at the same point, an overstatement is obtained of the uncertainty  in exposures to a
population.

20. On Page 31, the authors use the phrases "source-oriented focus"  and "urban-area oriented
focus". I  am not sure what these mean or why they are needed, unless the decision is whether to
use a representative set of locations based on source type or a set based on general urban
characteristics.

21. Also  on Page 31, in the last paragraph, my lack of clarity as to what is being done with the
epidemiological results makes it impossible for me to understand this paragraph. Again, the
treatment of the clinical results is clear in the document (once one gets to Section 4, at least), but
not the treatment of the epidemiological results.

22. On Page 33 and at several other points, the authors point to the need for baseline incidence.
This is only true if a relative risk, rather than absolute risk, model is used. The epidemiological
papers certainly report relative risk summaries, but they also report the primary data from which
absolute risk values can be calculated. Of course, one does not want to calculate absolute  risk
values if the biological processes are truly more consistent with relative risk (where the excess
incidence is itself a function of the background incidence).
                                            10

-------
Comments from Dr. Schlesinger
p. 24. 3.3.4. First bullet should read "Healthy Children."

p. 27. 1st paragraph. Health endpoints are not causal to ambient SO2. Rather, ambient SO2 is
causal to health endpoints.

p. 28, 2nd paragraph. What is the criterion for judging whether or not a health effect that is
considered to be of public health concern will not be appropriate for inclusion in quantitative
assessment?

p. 30. Should elderly be included in bulleted list?

p. 35, 2nd paragraph. In the last sentence, it is noted that risk estimates may sometimes be
developed using two different models. What will be the criteria for determining which model
will provide some basis for evaluating the ultimate NAAQS?
                                           11

-------
Comments from Dr. Seigneur
The two-tier approach for exposure assessment and the three-tier approach for risk assessment
appear to be logical ways to proceed.  The various steps of each approach are described with
sufficient detail for the reader to understand the technical approach and the sources of the data to
be used. The use of AERMOD for the Tier 2 exposure assessment is appropriate.

Emissions vs. concentrations:

The discussion of the Tier 1 exposure assessment (Section 3.2.1, p. 8) focuses on the largest
emitters. Clearly, the analysis must address the largest SC>2 emitters, but one must keep in mind
that a smaller emitter with a short stack may have a greater impact in terms of SO2 ground-level
concentrations than a large emitter with a tall stack.  The atmospheric dispersion aspect of the
exposure assessment will be addressed explicitly in the Tier 2 exposure assessment but the
potential limitations of the Tier 1 assessment must be clearly  stated when the results are reported.

Figure 2 presents emissions by source categories. The year of this emission inventory should be
stated because some source categories (e.g., coal-fired power plants) are being controlled and the
emissions of those  source categories will decrease. Also, one must note that some source
categories, which may appear small in a nationwide inventory, may be quite relevant in an
exposure assessment because they are concentrated in a few geographical areas (e.g., ocean-
going ships in ports).

Areas of interest for exposure assessment:  In the first paragraph on page 10, it is stated that
cities in California  report the lowest mean concentrations and that cities in the Northeast report
the highest.  This result is consistent with SC>2 emissions from coal-fired power plants being
historically greater in the Northeast than in the West. However, this result depends strongly on
the locations of the 862 monitors.  One may assume that the monitoring network was designed to
track the impact of 862 emissions from coal-fired power plant emissions. As this source
category is being controlled, other source categories may become of concern and the existing
monitoring network may not characterize their impacts properly. For example, SO2 emissions
from ocean-going ships could lead to significant  862 concentrations in ports (they typically burn
1.5% sulfur content fuel during transit within sulfur emission control areas, SEC As) but monitors
may not be located strategically in those areas. This point should be kept in mind for the Tier 1
exposure assessment and the potential impacts of ship emissions in areas where those emissions
are concentrated (ports and channels) should be explicitly addressed in the Tier 2 assessment.
For example, some port areas could be selected for detailed concentration modeling (e.g.,
Houston, Los Angeles) under the Tier 2 assessment.
                                           12

-------
Comments from Dr. Frank Speizer
Air Quality Considerations
1.  Use of proportional approach to adjust simulated potential alternative SO2 standards.
       This seems reasonable but there is an issue (see next comment)
2. Use of historic air quality data pre 2000 vs. proportional approach. .
       It is not an unreasonable use of historical data. However, with sites quoted in which there
are extremes of Policy Relevant Background that can be less than 1% (Ohio Valley) vs. >70%
(Volcanic region) it is not clear that there is a simple alternative approach. Additional regional
consideration will need to be played out.

Exposure Analysis
1.  Tier I approach
       . End of 2nd paragraph on page 7: It is not clear what is meant by: ".a quantitative
assessment of uncertainty .. .for selected components of the assessment".
       Section 3.2.1 Approach seems fine but there ought to be a model approach that would
allow for some validation that what is being done to make the proportion  estimates of 5 minute
maximums of <0.5ppm (or <0.4 or any other max). Given the limited number of co-located 5
minute and 1  hr samplers some "jack-knife", or other multiple statistical estimate of degree of
concurrence ought to be possible.
       Page 8-9. Given that electric generation -75-85% of SO2 emissions, and virtually all of
this is point source generated, there must be enough modeling data to allow for some generally
hourly predictive modeling by distance from source (e.g. <20Km, 20-40Km, >40Km).  Given
such data and the possibility of co-location with some continuous monitors, I do not think Staff
should block these data into yearly assessments but should use the co-located data across all
years for the model development and then use trends over time to assess frequency of 5 minute
max's over given levels (as is and other alternatives). I would be most disappointed if Staff
concluded that they could not get past Tier I.
       Page 14, end of first full paragraph.  It appears that Staff has not yet decided it plans to
use 5 minute max's of O.Sppm or 0.6ppm or the criteria for selecting even lower levels. Factors
influencing exposure levels needs some more discussion on how the choice will be made. One
issue that should come up later in the health risk discussion relates to the potential sizes of
subgroups that will be either susceptible or vulnerable. Back on page 3 Staff estimates that 0.7-
1.8% of total  asthmatic population could be exposed to outdoor SO2 > O.Sppm for >5 minutes
while exercising.  That translates to a very big number given 10 million US asthmatics!
       Page 16, first bullet,  sentence beginning 5 lines from end:  Not clear what this is saying.
Aren't ambient and outdoor concentrations the same?
       Criteria for assessing the uncertainty seem appropriate. I would add an additional bullet
that states that some kind of validation will be implemented and that some criteria of validation
are used to accept or reject the modeling be included before moving to Tier II (I am assuming
that such modeling will be acceptable and that Tier II will be performed.
3.  Tier II exposure assessment
                                           13

-------
       a. Modeling SO2 by proximity to sources.  See above.  This is a very reasonable
approach given the vast majority of source emissions are concentrated in stationary sources.
       b. Binning. The bins seem appropriate but it would be worth obtaining population
exposure estimates downwind from these bins in perhaps 3 tiers of <20, 20-40, >40Km of
distance.
       c. Using peak-to-mean ratio cumulative density functions. This seems reasonable but
may not go far enough. Given that a single peak over O.Sppm in an hour in the past predicted at
least another peak in the same hour over 70% of the time and currently predicts about 35% of the
time, does more thought needs to be given to modeling multiple exposures? Alternatively, does
it suggest that additional modeling needs to be done to estimate how much the hourly average
needs to be lowered to have 1  or less peaks above a certain value?

Top of page 23, sentence beginning end of line 5.and discussion in next paragraph.   Surely time-
location activity patterns must be almost random within waking hours.  The estimates of overlap
of activities within any given hours must relate to: 1) being outdoors (otherwise getting half the
dose); 2)Exercise; 3)Frequency of 5 min averages over O.Sppm; 4) somewhere between 35 and
70% of the 12 5 minute averages being over O.Sppm; 5) other factors.
Section 3.3.4, page 24.  Population groups of interest.  Although this is an improvement over
what was offered for NO2, if possible I think it would be useful to consider children broken
down further. I think it would be better to consider birth- preschool (near home); 4 or 5  to 9
(local community); and 10-18 (active outdoor physical activity). I recognize that the data  may
not exist but at least the breakdowns for exposure might be considered. The other groupings
seem appropriate, except might want to consider those adults carrying a chronic respiratory
disease and or CVD diagnosis as a separate (potentially more susceptible) group.
4.  General approach to uncertainty.
Page 25, last paragraph before section 3.4.  This paragraph discusses estimating model
uncertainty and suggests relying on "informed judgment" It is not clear whose judgment is being
relied upon. Is this related to the desire mentioned elsewhere to conduct "expert opinion
assessments"?

Health Risk Assessment
1.  General Structure seems appropriate except for what I would consider one major omission.
Throughout this discussion little assessment is given to the idea that there are within almost all
population groups subsets of particularly sensitive people.  I am not talking about the identified
susceptible or vulnerable.  Going back decades exposure studies in otherwise normal people
always identified individuals who were more sensitive to SO2 than the group being studied.
(Generally this amounted to 10-20% of the population being examined). The same seems to be
true in population studies and as one moves to exercising adults, asthmatics and exercising
asthmatics the percent susceptible or vulnerable increases. Often in the general population
studies these particularly susceptible people are  overwhelmed by the larger groups and the
relationships for the group  are considered null. They obviously are not null and in conducting a
risk assessment that is supposed to take into account a  margin of safety for the particularly
susceptible I fear the effects in these groups are being downplayed.
                                           14

-------
Finally, once again Staff has included on page 28, first paragraph in there decision matrix of
whether to conduct a Tier III risk assessment an unacceptable criteria of time and resources to
complete the task. If a Tier III assessment is warranted this cannot be a criteria for not doing it!
2.  Tier I health risk assessment
       Questions a-c. Agree with planned assessment as interpreted from the initial draft of the
ISA that the primary focus should be on the  respiratory outcomes as described.  However, in the
ISA there  were rather convincing evidence that for short term exposures in adults with pre-
existing disease that cardiovascular short term effects were present. This will need to be
revisited after further discussion of the ISA.

3.  Tier II  risk assessment
       a. Agree with potential respiratory health risks for the controlled exposure studies.
However,  these are mostly designed to assess and understand potential mechanisms for the risk
observed in free-living populations. The plan as outlined for a two Tier effort, like for the
exposure assessment, seems somewhat arbitrary as to whether it is called a two tier effort or a
logical progression in gathering the data necessary (and I believe from the draft ISA) available to
do all that is proposed.  The short term exposure assessment is well documented to move
forward, particularly for the respiratory outcomes described. With regard to the long term
assessment particularly for hospitalizations and mortality by sub-regions, this may have to await
the assessment of the draft ISA.
       b.  Selection of benchmarks of 0.5-0.6ppm.  I am concerned that these may be too high,
particularly because as indicated at a minimum these values also predict a second or greater
number of times during the same hour at least 35% of the time. For asthmatics exercising
outdoors this is an unacceptable exposure. Also there are studies in asthmatics that show
responsiveness below O.Sppm.
       c. Focus on areas around major sources. Yes, but will need population estimates by
distance from point sources.
       d. Yes, however, Staff needs to quantify those  places in urban areas that are within some
defined downwind distances (<20Km) to significant point sources.
       e. Once again the ISA should have summary tables that provide data on exposure
parameters for the studies to be used in assessing averaging times, location and season in
addition to co-pollutants. Working with staff in producing these table will provide data needed
here. I do  not suggest that those tasked with  doing this risk assessment go off independently to
construct such tables, since this might become an excuse for not doing the assessment that needs
to be done (because of lack of time or resources).
4.  Tier III risk assessment
       a. I do not agree that there is insufficient information to develop a credible exposure
response relationship from controlled human exposure evidence.  SO2 is one of the most and
best studied  pollutants in terms of controlled human exposure.  The data base is quite rich and
with careful  assessment has been used to assess a variety of dose-response relationships and has
identified  susceptible and vulnerable subgroups.  I believe this will come out in our assessment
of the ISA. At least I would reserve judgment on this  point  until after review of the ISA.
       b.  Here again I believe the success of modeling of exposure from the 5 min max from the
1 hour will dictate whether it would be useful to consider using the epi data for a Tier III
assessment of short term effects.  For longer term effects, consideration of identifying
                                            15

-------
susceptible subgroups (e.g. adults with preexisting disease, as a result of long term residence in
highly polluted regions or with repeated 24 hour exposures) will need to be assessed further.
                                             16

-------
  Comments from Dr. Wyzga
  Overall comment: It is difficult to describe and evaluate the plan generically.  At times I
  personally cannot understand the methods as described.  A fuller explanation with examples
  would enable me to judge them better. Given the deadlines faced, however, it would be
  impractical to redescribe the methods with more detail. The best approach would be to apply
  them and alter them, if necessary, given various reviews of their implementation.

  Health Risk Assessment

  Question 1: It seems to me, from reading the ISA and from following the literature, the risk
  assessment targets the correct health responses, those of asthmatics with relatively short
  exposures. Given this, I'm not convinced that there will much value in placing much additional
  effort and resources in a Tier I Health Effects Evaluation.

  Question 2 a: See above; I would not place much additional effort on this qualitative, but
  proceed to Tiers II and III.
          2b: I agree; since these studies are performed with controlled exposures, there is less
  concern about confounders. In addition, these studies find responses after exposures as short as
  5 minutes; I am unaware of any epidemiological study that has or can adequately address  such
  exposures.
           2c: I agree - exercising asthmatics appear to be especially vulnerable.

  Question 3 a. I believe so.
           3b. this seems reasonable
           3c. yes
           3d. I worry about the short-term SO2 issue and whether urban monitoring data are
  really characteristic of exposures. The approach is clearly more reasonable for the 24-hour
  exposures. I have no easy solution to offer for the one-hour problem. Near-source exposures
  may be more important for one-hour exposures and should be considered.
           3e. I'd be happy to share any data collected by EPRI and its Contractors. I think the
  Agency has to be aggressive in seeking such data.  The States may have more detailed data than
  has been reported.

  Question 4a. Risk assessments can be undertaken for specific  sources. They could be undertaken
  as illustrative. I would note that I was a co-author of one such risk assessment several years ago.
  See: P.  C. Freudenthal, H. D. Roth, T. Hammerstrom, and C. Lichtenstein, "Health Risks  of
  Short-Term SC>2 Exposure to Exercising Asthmatics", JAPCA, Journal of the Air and Waste
  Management Association, 39(6), June 1989.
            4b.  I would urge the panelists to consider breaking up any study area into small
geographic units.  I worry that effects may be seen in an area because several individuals were
exposed to a plume with much higher concentration levels than  measured by the monitor. The
monitor values may reflect the existence of a plume, but at levels much lower than the plume;
hence health effects would be detected, but in reality these effects are caTused by exposures to the
                                             17

-------
higher plume values. This is not an easy issue, and there may be no easy solutions; hence
whatever approach is taken, it is important to state caveats and limitations of the analysis.

Question 5:  To the extent possible I would urge that the consideration of uncertainties be
embedded into the risk analysis rather than undertaking a baseline analysis and several sensitivity
analyses. The result should be some distribution of estimates.

            If I recall correctly, chamber study effects were seen in exercising asthmatics who
were medication-free. This factor could be considered in the overall risk assessment because
medication is taken to protect from responses to lots of environmental agents in addition to SO2
and air pollution.
                                              18

-------
Comments from Dr. Larson

Air Quality Considerations:
Based on the low estimated contribution of policy-relevant background SO 2 to overall
ambient SO 2 levels, staff is considering a proportional (i.e., linear) approach to
adjusting air quality to simulate just meeting potential alternative SO 2 standards that
are below recent air quality concentrations. Do the Panel members have comments
on adopting a proportional approach to simulate just meeting more stringent
alternative air quality standards?

This is a reasonable approach, given the very low policy-relevant background concentrations in
most areas of the U.S.  Although there are a few areas affected by natural sources (some
locations in the Northwestern U.S.), the current levels are very low.

Recognizing that current ambient air quality concentrations are lower than the current
standards,  the draft Health Assessment Plan discusses two alternative approaches to
simulating ambient SO2 levels associated with just meeting the current SO2 standards:
use of historical air quality data  (e.g., possibly pre-2000) when ambient levels were at
or above the current standards, or use of a proportional (i.e., linear) approach to adjust
SO2 levels upward. Do the Panel members have advice or comments on these two
alternative approaches to simulating air quality just meeting the current SO2
standards?

The historical data has the theoretical advantage of including any non-linearities in the model.  It
is not clear that this is an actual advantage (one major non-linear effect would be due to a large
policy-relevant background value, which does not exist).  Another drawback is that the historical
approach would capture a different mix of sources than currently exists. Therefore the relative
shape of the distribution of short term values could be different than it is now.

Exposure Analysis:
1. In considering the exposure analysis broadly:
a. Do Panel members have any comments on the general structure and overall two-tier
approach that staff plans to use for the exposure analysis? Are the criteria that staff
plans to use for deciding whether to conduct a Tier II analysis clear and appropriate?

The criteria seem reasonable and practical.

b. Have the most important factors influencing exposure to SO2 been clearly accounted
for and described?

Given the dynamic complexities  of plume dispersion, it is reasonable to capture some of this
with a deterministic model such as AERMOD.  The daily changes in mixing depth strongly
influence the downwind impacts from  a given elevated source.

c. The draft plan describes the basis for and selection of population groups of interest
                                            19

-------
(i.e., children, asthmatics (children and adults), and the elderly) for which SO2
exposure estimates are to be developed. Do Panel members generally agree with the
groups of interest identified in the draft plan?

This seems to be consistent with the ISA.

2. In considering the Tier I exposure assessment:
a. Do Panel members agree that a statistical model using available ambient 5-minute
monitoring data is appropriate for estimating expected exceedances of very short-term
(5-minute) potential health effect benchmarks?

There is relatively limited data on the 5 minute values. However, there is no reason to distrust
what little data there is.

b. Do Panel members agree with the approach of applying a statistical model to estimate
5-minute concentration exceedances at monitoring locations where only 1-hour
monitoring was performed for evaluating the extent of 5-minute peaks associated
with meeting alternative standards with longer averaging times?

If the model is based on historical data or on locations that are qualitatively different than the
locations of interest (e.g. point source impacted vs general urban), then care should be taken to
make sure that the potential source impacts are appropriately considered. Another issue is the
source geometry-tall stacks vs near ground level releases. These two geometries can, in
principle, lead to different peak to mean values, especially during convective daytime conditions
(c.f. Van Dop et al Boundary Layer Meteorology 116, 1-35, 2005; also Luhar et al. Atmos.
Environ. 34, 3599-3616, 2000; Franzese andBorgas J. Appl. Met., 41, 1101-1111, 2002).
3. In considering a potential Tier II exposure assessment:
a. Do Panel members agree with the combined emissions/dispersion modeling approach
to estimate short-term (hourly) SO2 concentrations in close proximity to SO2 emission
sources?

The approach described in equations 1 and 2 is an attempt to allocate the distribution of 5-minute
SO2 values within in a given hour.  Some thought might be given to the use of the standard
deviation of wind direction for each hour, if such data are available. This is an indirect measure
of the breadth of this distribution and is invoked in odor models.  The effects of narrow plumes
that have recently passed over water (relatively smooth surface) would narrow the 5-minute
distribution. These effects are not predicted by AERMOD (as far as I know),  but could be
crudely captured by geographical variables.
b. Do Panel members have comments or advice regarding the described binning of
sources and development of prototype stacks/facilities?
                                           20

-------
It might be useful to look at the measured peak to mean ratios as a function of wind direction.
This could be done using a conditional probability function, i.e., looking at the probabilities of
occurance from a given direction when the ratio exceeds a certain value.  If the high peak values
are due to major point sources, this approach could "point" to the source's location. In turn, this
might allow one to segregate point source-influenced measurements from those influenced by
more widely distributed sources. Also, as stated earlier, the release height is an important
determinant of peak to mean ratios.

c. Do Panel members agree with the approach using peak-to-mean ratio cumulative
density functions (PMR CDFs) to estimate very short-term peak concentrations from
the 1-hour modeled concentrations?

Seems reasonable, especially allowing these values to vary with 1-hr levels.

d. Do Panel members generally agree that the approach described using APEX is
reasonable and appropriate to estimate the occurrence of very short-term (5 minute)
SO2peak exposures?

In general, it is a reasonable approach. I cannot tell how sensitive it is to the assumptions of the
actual 5-minute distributions for any given hour. It would seem important. Perhaps this can be
included as part of the uncertainty analysis.

e. Do Panel members have any comments or advice regarding the general approach to
addressing uncertainty and variability in each Tier of the exposure assessment as
described in the draft plan?

Given all the uncertainties and lack of locations with 5-minute data, the approach of discussing
the potential uncertainties is reasonable.

General Comment on Risk Assessment:

If possible, it would be good to try to include the fact that in most industrial parts of town, the
incomes are lower and the access to managed care for asthma is poor. Yet these  are the same
locations that experience the highest peak concentrations from industrial emissions.
                                           21

-------
Comments from Dr. Thurston

       This presented plan appears to sufficiently address the needs for the scope and
approaches for, and highlights key issues in, the estimation of population exposures and health
risks posed by SOx under: 1) existing air quality levels; 2) upon just meeting the current SCh
primary NAAQS, and; 3) upon just meeting potential alternative standards under consideration
by the Administration.
       I do have several concerns, however.  On page 2, the report limits itself to the effects of
gaseous SC>2 alone, dismissing any role by particulate sulfates, or sulfur dioxide's interactions
with PM in general, in assessing the possible health impacts of sulfur oxides. This ad hoc
decision seems too yield too narrow a definition, and may cause an underestimation of the risks
of sulfur oxides, as well as in the benefits achievable via sulfur oxide emissions reductions.
       Another concern that I have is in regard to the reliance on multiple pollutant models for
risk assessment, as proposed in the middle  of page 35. This, despite the fact that the report
states in the same paragraph: "When collinearity exists, inclusion of multiple pollutants in
models often produces unstable and statistically insignificant effect estimates for both SC>2 and
the co-pollutants.".  Thus, a reliance  on published single pollutant model coefficients would
seem far preferable.
       With regard to the Health Risk Assessment Charge Question 3.b. (Do Panel members
generally agree with the tentatively identified potential health effect benchmark of 0.5 to 0.6 ppm
for exercising asthmatics  following 5-10 minutes  SCh exposure?), I would say that I feel that this
is too high, based upon my reading of the ISA draft.  This ignores evidence and biological
plausibility regarding the lowering of threshold by the co-presence of PM, which is always the
case in the environment, and would also provide no margin of safety vs. the clinical study
results.  I should think a benchmark closer  to 200 ppb would be more appropriate and of more
interest to CAS AC.
       Finally, with regard to Health Risk  Assessment Charge Question 1,1 must object to the
inclusion of time and resources as a criteria for determining whether to do a Tier II analysis (pg.
36, last two lines). This Tier III analysis needs to be done, and has this been known about by the
EPA for years, so this should not appear as a criteria. Drop this last bullet from the list.
                                            22

-------
Comments from Dr. Kenski

Air quality considerations:

Q. 1: With respect to charge question 1, the proportional approach for simulating concentrations
below recent data to examine scenarios that just meet alternative standards is acceptable.

Q2:  I have a slight preference for using historical data to examine scenarios with concentrations
above current standards, because it may be that the historical data have subtle differences in
distribution that would not be captured by the proportional 'roll-up' approach. Alternatively, an
analysis of distributional differences could be performed to demonstrate that distributions from
higher-concentration historical conditions do not differ appreciably from current lower
concentrations.

Exposure analysis:

Ql. The general  structure and approach are logical.  It's actually not clear what the criteria for
deciding to conduct a Tier II exposure analysis are, versus a Tier 1 assessment.  The document is
quite clear about how and what will be done, but it seems to implicitly assume that both will be
performed (and there doesn't seem to be any reason not to perform the exposure assessment
through Tier II).  The factors in Sec. 3.4 are a little vague. If the ambient air characterization
leads us to believe that no current ambient concentrations are above any potential alternative
standard, we're done? No further exposure or risk assessment is necessary? The important
factors influencing exposure and populations of interest have been accounted for.

Q2.  I liked the proposed model for estimating peaks at monitors with 1-hour data.  Until it's
actually tested with some of the sites where 5 min data are available, it's not possible to give it
an unqualified approval, but it seems eminently reasonable for generating the needed data. Of
course the exposure assessment will need  to document the performance of this model and
document its contribution to the overall uncertainty assessment.

Q3.  The Tier II approach made a lot of sense.  I wonder,  however, if the choice of most recent 3
years of meteorology is necessarily best?  Is there any evidence to indicate that years vary
significantly in their potential to be more or less conducive to high SO2 concentrations,
independent of changes in emissions? E.g., perhaps cooler summers have slower SO2->SO4
conversion and so SO2 concentrations are higher at near-source monitors? Maybe an
examination of yearly CDFs or quantile-quantile plots would show year-to-year differences.  If
so, then perhaps  an argument could be made for selecting years that are more likely to have
higher SO2, to ensure that modeled concentrations would be conservative.

The binning approach is a reasonable one; the  development of bins is an interesting problem in
itself. The inclusion  of terrain as a variable is  important;  it's not clear whether this is definitely
going to be incorporated or just examined as a possible option.

Q4:  The discussion of uncertainty was helpful.
                                           23

-------
Health Risk Assessment:

Ql:  The approach to the health risk assessment was clearly laid out and reasonable.  The criteria
for conducting a Tier III assessment were more clear than those for the exposure assessment
section (especially the 1st paragraph on p. 28). Also, the 2nd paragraph on p. 28 was a
particularly nice description of the goals of this process.

Q2:  Based on the data presented in the ISA, I agree with the staffs assessment of the health
endpoints and susceptible populations of most interest.

Q3:  I have no expertise in health risk assessment, so I can only answer these questions based on
what was presented in the ISA. That  said, I agree with the staffs choices with respect to health
benchmarks and the focus on exercising  asthmatics.  Certainly the decision to focus on areas
around major sources and on urban areas is appropriate.

Q4&Q5: No additional comments.

Sec.  3.2.1, 1st paragraph: This description of monitoring  could use some additional clarification.
Do the 94 monitors that report 5 minute maxes report one maximum 5 minute concentration per
hour (or day?), or 12 5-minute values per hour (is this what is meant by 'containing continuous
monitoring'?  Even if AQS only contains one 5-minute max per hour, the states or local
organizations that collected the 5 minute data may have archived measurements for the other  11
5-minute intervals. Please be sure to  check with them, since the number of monitors is limited,
to see what additional measurement data might be available.

Sec.  3.2.1, 2nd paragraph: Electric generating units are the largest source of SO2 nationally, but
on a local scale many other sources are significant - industrial coal use, refineries, coking, metal
processing, paper mills,  and shipping (bunker fuel use). The proximity of these types of sources
to the monitors will need to be considered in the analyses proposed in Sees.  3.2.1.1 and 3.2.1.2,
not just EGUs.  The last sentence of this paragraph makes it sound like proximity to these other
sources may or may not be accounted for in the data analyses,  (this seems to be addressed
adequately in later sections of the report, just not right here)

Sec.  3.2.1.4, p. 14, 2nd paragraph: The application for this analysis of population density isn't
clear. Will these estimates of susceptible populations be  generated just for the vicinity around
each ambient monitor or scaled up for the nation?
                                            24

-------
Comments from Dr. Gordon

       The Plan is well conceived and written and the tiered approach is appropriate for the task.
Because of a lack of expertise on exposure assessment and modeling, I will comment only on the
health portion of the risk assessment.  The conclusion that adverse respiratory effects are the
strongest health findings appears to be valid and clearly substantiated by the ISA. The
advancement of the respiratory hospital admissions and ER visits to a Tier III analysis is needed
and verified.  It is puzzling, however, why the Assessment Plan indicates that while there is clear
evidence of bronchoconstriction in asthmatics after short term exposure to 0.5 to 0.6 ppm sulfur
dioxide, it has been decided not to do a Tier III evaluation on this health effect.  The dose
response for acute bronchoconstriction has been know for nearly 2 decades and a Tier III
evaluation of this health effect is warranted. As stated for Tier II evaluations, the approach may
be different for epidemiology and controlled human exposure studies, but the quantitation of
acute data from controlled human studies is feasible. If EPA feels that this quantitative
assessment is not possible, then additional justification is  required.

       The Plan states that a Tier III risk assessment depends on a number of factors including
"whether or not there is adequate time and resources". Given the enormous effort and resources
(time and money) used to put together the ISA and the Assessment Plan (funded research,
scientific review of grants and publications, EPA scientists writing the ISA and the Plan,
CASAC panel members' review process, etc.), it is unclear why resources may  not be available
to accomplish this last and critical step in a timely fashion.
                                           25

-------
Comments from Dr. Hattis
1. Based on the low estimated contribution of policy-relevant background, and ambient SO2
levels, staff is considering a proportional (i.e., linear) approach to adjusting air quality to
simulate just meeting potential alternative SO2 standards that are below recent air quality
concentrations. Do the Panel members have comments on adopting a proportional approach to
simulate just meeting more stringent alternative air quality standards?

       This seems generally reasonable to me.

2. Recognizing that current ambient air quality concentrations are lower than the current
standards, the draft Health Assessment Plan discusses two alternative approaches to
simulating ambient SO2 levels associated with just meeting the current SO2 standards:
use of historical air quality data (e.g., possibly pre-2000) when ambient levels were at
or above the current standards, or use of a proportional (i.e., linear) approach to adjust
SO2 levels upward. Do the Panel members have advice or comments on these two
alternative approaches to simulating air quality just meeting the current SO2
standards?

       To the extent possible, the goal should be to represent a realistic future scenario—one
       that might actually occur. One such scenario would be  a generalized increase in present
       emissions resulting from increased SO2-emitting economic activities of all kinds.  It
       seems likely to me that this would approximately correspond to the proportional (linear)
       approach rather than the historical reconstruction.
Exposure Analysis:

1. In considering the exposure analysis broadly:
a. Do Panel members have any comments on the general structure and overall two-tier
approach that staff plans to use for the exposure analysis?  Are the criteria that staff
plans to use for deciding whether to conduct a Tier II analysis clear and appropriate?
b. Have the most important factors influencing exposure to SO2 been clearly accounted
for and described?
c. The draft plan describes the basis for and selection of population groups of interest
(i.e., children, asthmatics (children and adults), and the elderly) for which SO2
exposure estimates are to be developed. Do Panel members generally agree with the
groups of interest identified in the draft plan?

       Yes.

2. In considering the Tier I exposure assessment:

a. Do Panel members agree that a statistical model using available ambient 5-minute monitoring
data is appropriate for estimating expected exceedances of very short-term (5-minute) potential
health effect benchmarks?
                                           26

-------
b. Do Panel members agree with the approach of applying a statistical model to estimate 5-
minute concentration exceedances at monitoring locations where only 1-hour monitoring was
performed for evaluating the extent of 5-minute peaks associated with meeting alternative
standards with longer averaging times?

       Generally the idea of modeling the 5 minute peaks with the aid of empirical data and a
       statistical model is a good one. I have not grasped the exact statistical model to be  used
       sufficiently, however, to be sure that it fully realizes the opportunities presented by the
       available data and takes precautions to correct for the artifactual spreading of the data
       from measurement error. On the latter issue, it is somewhat troubling to see the
       discussion to the effect that only "valid" measurements will be used.  Fine, and
       impossible 5 minute/hourly PMR values less than 1 or greater than  12 will be excluded.
       But this does not mean that the effects of residual measurement error in spreading out
       both the 5 minute and 1 hour average observations have been excluded. Any set of
       empirical observations has measurement error.  In general the observed lognormal
       variance will be the sum of the real lognormal variance of real SO2 levels and  some
       lognormal variance attributable to measurement errors. However only real variation
       affects real people's exposures and risks. Thus to get an estimate of the true frequency  of
       high values of the exposure distributions (and the corresponding ratios of 5 minute/1 hour
       levels) it is important to estimate the measurement error variance (likely different for the
       shorter vs longer averaging times) and subtract that from the variance of the crude
       observations.

3. In considering a potential Tier II exposure assessment:
a. Do Panel members agree with the combined emissions/dispersion modeling approach
to estimate short-term (hourly) SO2 concentrations in close proximity to SO2 emission
sources?

       Yes.  I do, however, think that to the extent possible some effort should go into
       comparing observed and dispersion model predicted distributions of hourly SO2 levels  at
       monitors near specific sources. Based on the results of this comparison, the distribution
       of hourly SO2 levels for unmonitored sites may be adjusted for better accuracy.

b. Do Panel members have comments or advice regarding the described binning of
sources and development of prototype stacks/facilities?
c. Do Panel members agree with the approach using peak-to-mean ratio cumulative
density functions (PMR CDFs) to estimate very short-term peak concentrations from
the  1-hour modeled concentrations?
d. Do Panel members generally agree that the approach described using APEX is
reasonable and appropriate to estimate the occurrence of very short-term (5 minute)
SO2 peak exposures?

       Yes, generally to b, c, and d, subject to my earlier comments about the need to separately
       remove the effects of measurement errors from the 5 minute and hourly data that give rise
       to the PMR CDFs.

 4. Do Panel members have any comments or advice regarding the general  approach to
                                           27

-------
addressing uncertainty and variability in each Tier of the exposure assessment as
described in the draft plan?
       The second paragraph on page 7 says in part, "At each tier of the exposure assessment, an
       evaluation of the uncertainties will be performed and the relative degree of confidence in
       the exposure estimates will be determined." "Determined" is a bit stronger word than I
       would like to use in general for an uncertainty analysis.  Consider substituting the more
       modest terms, "estimated" for a quantitative analysis, or "assessed" for a more qualitative
       or semi-quantitative discussion.

Health Risk Assessment:
1. Do Panel members have any comments on the general structure and overall three-tier
approach that staff plans to use for the risk assessment? Are the criteria that staff plans to use for
deciding whether to conduct a Tier III risk assessment clear and appropriate?

       I think so.

2. In considering the Tier I risk assessment:

a.  Do Panel members agree with the approach of having a qualitative assessment of health
endpoints to identify which are likely candidates for a more sophisticated and quantitative tier of
assessment?

       Yes.

b.  Do Panel members agree with our initial observation that controlled human exposure studies
demonstrate strong evidence for bronchoconstriction in exercising asthmatics following 5-10
minutes SO2 exposure?

       Yes.

c.  Do Panel members agree with staffs initial observation that the strongest epidemiologic
evidence is for respiratory symptoms in asthmatic children  and respiratory-related hospital
admissions and respiratory-related emergency department visits in asthmatics and others with
respiratory conditions?
       Yes.
3. In considering the Tier II risk assessment:

a. In general, are staff plans to use potential health effect benchmarks to address respiratory
effects demonstrated in exercising asthmatics in controlled human exposure studies clear and
appropriate?
                                            28

-------
       The proposal is clear.  As with the NOx analysis I have reservations about the general
       plan to use "health effects benchmarks" and the incidence of exceedances as the main
       analytical approach. As I illustrated in my comments on the ISA (reproduced below), an
       approach that uses a crude log probit dose response function together with quantitative
       assessment of the full distribution of exposure concentrations is quite feasible.

       I think the ISA document could have gone a little farther in analyzing the data in Table
2.4.2 on SO2 concentration distributions observed by existing monitors in CSMA's for different
averaging times. Figure 1 shows lognormal plots of the data in this table.  From the
correspondence of the data points to the fitted straight lines, it can be seen that particularly for
the shorter averaging times, the data are well described by lognormal distributions. In the fitted
regression line the intercept is an estimate of the logarithm (base 10) of the geometric mean and
the slope is an estimate of the logarithm of the geometric standard deviation. For example, the
estimated geometric mean for the maximum 1  hour daily averages of the readings from CSMA
monitors is 100.806 = 6.4 ppb and the estimated geometric standard deviation is 100.524 —about
3.34.  These results allow us to make at least some quantitative estimates of the likely frequency
of ambient outdoor exposures at levels associated with various incidences of short term
responses to SO2 in populations that have been studied in clinical  settings (see below).

       I would have preferred a more quantitative treatment of the issue of human variability in
the undoubted causally related responses observed from clinical exposures to SO2. I am
particularly intrigued by the possibility of a more quantitative analysis of the individual subject
response data of Horstman et al. (1986) reproduced in Figure 3.1-6, and any other similar data
sets.

       For the analysis of human variability in Figure 2 below I have extracted the individual
Horstman et al. data as best I could from the figure provided in the ISA and the accessible
abstract (I could not easily obtain the original paper). Figure 2 is based on the conventional
assumption for probit  analysis that in the population of asthmatics studied there is a lognormal
distribution of individual thresholds for the response  (a doubling of airway resistance during
exercise). In this case the intercept is an estimate of the log of the SO2 level needed to elicit the
response in the median asthmatic (100.0189 = 1.044 ppm = 1044 ppb) and the slope is an
estimate of the log of the geometric standard deviation [the Log(GSD) in our terminology] of
individual response thresholds (100.374 = 2.37).
                                           29

-------
                           Figure 1
        Lognormal Plots of Data from Table 2.4.2—Distributions

        of S02 Concentrations (ppb) for Different Averaging Times
o.
o.
 I

o


1


I
u
a
o

U
S
o
-1
                         y = 0.806 + 0.524x RA2 = 0.990


                         y = 0.306 + 0.534x RA2 = 0.998


                         y = 0.442+ 0.418x RA2 = 0.971

                         y = 0.457+ 0.325x RA2 = 0.
1 hr max

1 hr ave

24 hr ave

1 yr ave
                                    Z-Score
                               30

-------
                         Figure 2
Lognormal Plot of the Distribution of Individual Sensititivities

(SO2 Concentrations Needed to Double Specific Airway
Resistance) For 27 Exercising Asthmatics (Horstman et al. 1986)
    S
    o.
    a.

    ^Ofi
    o
    -
               v = 0.0189+ 0.374x RA2 = 0.934
                            Z-Score
                           31

-------
       In previous efforts my colleagues and I have compiled a substantial database of
information on human interindividual variability for a variety of responses (see the website at
http://www2.clarku.edu/faculty/dhattis).   The log(GSD) of about 0.37 in this case is not at all
unusually large-it is actually toward the lower end of observations of variability in responses to
acute inhalation exposures compiled in our data base (Table 1) (however, it can be seen that in
many of these cases with larger variability the agents act via specific receptors or via allergic
processes that may well in general be subject to more variability than responses to nonspecific
irritants).

       Given the variability analysis in Figure 2, it is straightforward to make at least a tentative
projection of the likely incidence of responses for asthmatics similar to those studied by
Horstman et al. (1986) at any air level, assuming that the population distribution of response
thresholds is in fact perfectly lognormal:

                                       expected incidence of response (% of days
                                     expected to cause 100% increase in specific air
                                    way resistance for exercising asthmatics, ignoring
                                      the exposure duration difference between 10
                                     minute studied exposure and 1 hour duration for
                        ppb          the greatest 1 hour average in a 24 hour period)
                         10                          3.4E-08
                         20                          2.2E-06
                         30                          1.9E-05
                         40                          7.7E-05
                         50                          2.1E-04
                        100                         3.2E-03
                        150                           0.012
                        200                           0.028
                        400                           0.13
                        600                           0.26
                        800                           0.38
                        1000                          0.48
                                              32

-------
                                                                                       Table 1
                        Previous Observations of Human Interindividual Variability in Local Lung Function Responses to Inhaled Agents
log(GSD)
0.74
1.00
0.32
0.43
0.76
0.57
1.33
1.11
0.64
0.42
0.51
0.78
1.13
0.60
0.97
0.59
0.27
response studied
Air Cone. Needed to cause 10%, 15%, and 20% decrease
inFEVl
Air Cone. Needed to cause 10%, 15%, and 20% decrease
inFEVl
FEV1 change in relation to CXT of ozone exposure
(clinical)
FEV1 Increase by Antiasthmatic
PD20— concentration needed for 20% increase in
individual baseline value of FEV1
PD20— concentration needed for 20% increase in
individual baseline value of FEV1
Specific Airway Resistance PC50— concentration needed
for 50% increase in individual baseline value
Specific Airway Resistance PC50— concentration needed
for 50% increase in individual baseline value
Specific Airway Resistence— concentration needed for
20% increase in individual baseline value
Specific Airway Resistence-concentration needed for
100% increase in individual baseline value
Specific Airway Resistence— concentration needed for
15% increase in individual baseline value, mean of 2
trials with and without ozone
Specific Airway Resistence-concentration needed for
15% increase in individual baseline value, mean of 2
trials with and without ozone
Specific Airway Resistence-concentration needed for
20% increase in individual baseline value
Specific Airway Resistence— concentration needed for
20% increase in individual baseline value
Specific Airway Resistence-concentration needed for
20% increase in individual baseline value
Specific Airway Resistence— concentration needed for
20% increase in individual baseline value
Specific Airway Resistence-concentration needed for
20% increase in individual baseline value
population studied
Females— general population
Males— general population
N
748
810
Experimental subjects
Asthmatics
Atopic subjects
Atopic subjects
Bakers-occupationally
exposed
Bakers— occupationally
exposed
5733 smokers with mild to
moderate airflow obstruction
Healthy athletic adults, 18-50
Allergic asthmatic patients
Allergic asthmatic patients
9 year old New Zealand
Children
Allergic asthmatic patients
General adult population,
Norwegian community, Age
18-73
Nonsmoking adults with mild
asthma
Nonsmoking adults with mild
asthma
14
13
17
34
34
5733
66
9
6
813
15
490
17
18
agent
Methacholine
Methacholine
Ozone
Salbutamol
Ragweed allergen
Histamine
Wheat flour dust
Wheat flour
extract
Methacholine
Methacholine
Grass allergen
Ragweed allergen
Methacholine
Methacholine
Methacholine
Histamine
Metabisulphite
data source
Paolettietal., 1995
Paolettietal., 1995
McDonnell et al. 1995 analyzed in
Hattis 1998
Lipworth 1992, analyzed in Hattis
1998
Meerschaert, 1999
Meerschaert, 1999
Merget, 1997
Merget, 1997
Tashkinetal, 1996
Balmesetal., 1997
Hanania, 1998
Hanania, 1998
Sears etal., 1996
Hanania, 1998
Bakke, 1991
Evans, 1996
Evans, 1996
Source: Human interindividual variability database, updated as of 5/05, available "http://www2.clarku.edu/faculty/dhattis" discussed in
Hattis, D. "Distributional Analyses for Children's Inhalation Risk Assessments." Journal of Toxicology and Environmental Health, 71:1-9, 2008 in press.
Hattis, D. and Lynch, M. K. "Empirically Observed Distributions of Pharmacokinetic and Pharmacodynamic Variability in Humans-Implications for the Derivation of Single Point Component Uncertainty Factors
Providing Equivalent Protection as Existing RiDs." In Toxicokinetics in Risk Assessment, J. C. Lipscomb and E. V. Ohanian, eds., Informa Healthcare USA, Inc., 2007, pp. 69-93.
Hattis, D., Baird, S., and Goble, R. "A Straw Man Proposal for a Quantitative Definition of the RfD," Drug and Chemical Toxicology, Vol. 25, pp. 403-436, (2002).
                                                                                          33

-------
       Given the analyses presented earlier in my responses to charge questions 2, and 4-
6, and

       * Assuming that both the exposure distribution and the distribution of individual
       thresholds for response in asthmatics are perfectly lognormal,

       * Ignoring for now the exposure duration and intake difference between the one-
       hour exposures measured by the monitors and the 10 minute exposures used to
       measure effects in the exercising asthmatics studied by Horstman et al., and

       * Neglecting any systematic differences there are likely to be between individual
       personal exposures and air concentrations measured in the elevated outdoor
       compliance monitors

we can derive an estimate of the overall fraction of days that asthmatics similar to those
in the studied group.  We do this by cutting the assumed lognormal distribution of air
concentrations from 0 to 1000 ppb into intervals of 1 ppb, calculating the number of
asthmatic people who might be in each interval, and summing up the number likely to
respond during the maximum hour's exposure on each day (Table 2). Overall the fraction
of asthmatic-days expected to elicit a response of the severity recorded by Horstman et al.
(1986) is about 2.9 per  10,000. Interestingly, half of the total response incidence is
attributable to very rare high exposures (over about 230 ppb).  This results from the larger
estimate of variability in exposures, compared to the estimate of variability in human
response thresholds.

b. Do Panel members generally agree with the tentatively identified potential health
effect benchmark of 0.5 to 0.6 ppm for exercising asthmatics following 5-10 minutes
SO2 exposure?

       No. Effects are clearly observed in some people well below this level, and the
       effect incidence for almost any level can be estimated (see responses above to the
       questions on the ISA), if one is willing to postulate an overall lognormal
       distribution of individual thresholds—which seems reasonably compatible with
       available data and applicable theory.

c. Do Panel members generally agree with the staffs approach of focusing on areas
around major sources of SO2 with respect to concerns about 5-10 minute peak exposures
related to the respiratory effects observed in controlled human exposure studies?

       Yes.

d. Do Panel members generally agree with staffs approach of focusing on urban areas
with respect to concerns about 1- and 24-hr and annual SO2 concentrations related to
respiratory effects observed in epidemiologic studies?

       Yes.
                                        34

-------
                                        Table 2
  Illustrative Calculation of the Expected Fraction of Days on Which Exercising
Asthmatics Might Experience a Doubling of Specific Airway Resistance, Subject to
                           Extensive Assumptions (see text)
   Upper end of conctration      overall contribution to fraction of days with    cumulative total fraction of days
       interval (ppb)                    response in interval                     with response
            10                              1.6E-09                             1.6E-09
            20                              9.0E-08                             9.1E-08
            30                              5.4E-07                             6.3E-07
            40                              1.4E-06                             2.1E-06
            50                              2.6E-06                             4.7E-06
            60                              3.9E-06                             8.6E-06
            70                              5.1E-06                             1.4E-05
            80                              6.2E-06                             2.0E-05
            90                              7.0E-06                             2.7E-05
            100                              7.7E-06                             3.5E-05
            110                              8.2E-06                             4.3E-05
            120                              8.5E-06                             5.1E-05
            130                              8.7E-06                             6.0E-05
            140                              8.8E-06                             6.9E-05
            150                              8.8E-06                             7.8E-05
            160                              8.7E-06                             8.6E-05
            170                              8.6E-06                             9.5E-05
            180                              8.4E-06                             l.OE-04
            190                              8.2E-06                             1.1E-04
           200                              7.9E-06                             1.2E-04
           300                              6.4E-05                             1.8E-04
           400                              4.0E-05                             2.2E-04
           500                              2.4E-05                             2.5E-04
           600                              1.5E-05                             2.6E-04
           700                              9.7E-06                             2.7E-04
           800                              6.4E-06                             2.8E-04
           900                              4.3E-06                             2.8E-04
           1000                             3.0E-06                             2.9E-04
                                           35

-------
e. Do Panel members have any comments or advice with respect to staffs approach of
gathering additional information to characterize the SO2 ambient air quality that existed
at the time various key U.S. and Canadian studies addressing respiratory effects were
conducted to see if the concentration-response relationships observed in these
epidemiologic studies are related to particular SO2  levels and associated averaging times,
geographic location and/or season, and the inclusion of various co-pollutants?
       I think this is an ambitious undertaking, but worth trying.  The key issue of
       confounding might be addressed by trying to compare results of studies with more
       vs less vs different types of co-pollutant exposures, particularly organized by
       major sources of particulates in the areas studied by different authors.
4. In considering a potential Tier III risk assessment:

a. Do Panel members generally agree that there is insufficient information to develop
credible exposure-response relationships for use in a quantitative risk assessment based
on the controlled human exposure evidence?

       Not at all—the tables and figures I developed with only a couple of days of effort
       in the ISA response section above do exactly that for at least one type of response.
       A better job can be done with more efforts and a more sophisticated analysis, but
       surely some quantitative analysis of likely effect incidence is feasible.

b. Do Panel members have any comments or advice with respect to the general approach
or specific  factors to be considered in deciding whether or not to proceed to a Tier III
quantitative risk assessment for the respiratory-related health endpoints based on
epidemiologic evidence discussed in the draft plan?

       I think EPA should plan on doing a Tier III assessment for at least the simplest
       short term endpoints.

5. Do Panel members have any comments or advice with respect to the general approach
to addressing uncertainty and variability in each Tier of the risk  assessment as described
in the draft plan?
       Just that it is important to treat at least variability in susceptibility quantitatively
       based on existing data in available clinical observation papers. Uncertainty
       analysis methods also deserve some quantitative attention.
                                         36

-------
Comments from Dr. Kinney

Air Quality Considerations:
1.     I think the proportional approach for adjusting air quality is fine. However, I
question the value and purpose of rolling up concentrations from ambient to the level of
alternative standards. The only obvious reason to do that would be as part of a "benefits
analysis" to demonstrate the health benefits of having ambient concentrations below the
level of the standard. That's not the purpose of this exercise obviously, so why do it?
2.     Note concern expressed above. However, if you must do this, I prefer the
proportional adjustment method.

Specific Comments:
p. 5, para 3, last line: controlled exposure studies can provide useful exposure/response
functions for use in risk assessment; this should be noted here.
p. 8, para 2, lines 1-2 and elsewhere: the term "surrogate exposures" is mentioned here
and several other places, before any definition is provided.  Need to add a couple of
explanatory sentences early on to explain what is meant.  It becomes  clear later, but needs
to do so earlier.

Exposure Analysis:
1 .a.   The general structure and process for the two-tiered approach is well justified and
appropriate.
1 .b.   The most important factors  influencing exposure to SO2 have been clearly
accounted for and described.
I.e.   The population groups of interest are appropriately chosen.

2.a,b.  I think the statistical approach seems reasonable, although the description is
somewhat unclear, as noted on in my comments on the draft document.

3.a-d. Modeling approach is reasonable. The binning of exposures sounds ok, but the
devil will be in the details, and we'll need to see how well it works in practice. The PMR
CDF approach is reasonable.  I like the APEX modeling approach for getting at actual
personal exposure distributions.

Specific Comments:
p. 10, para 1, 7th to 5th line from bottom: lack of correlation also likely reflects
the  high proportionate uncertainty for concentrations  at or below
the  instrument  LODs

p. 13,  equations  and  last  para:   this  material is  a bit
confusing.   What  is meant  by "the appropriate function will be
applied"?   What is  being estimated?   Give  an example  calculation.

p. 15,  4th  para, last sentence:  This  is hard to  understand.   Edit
to clarify  meaning.   I had to read it  several times.
                                      37

-------
4.      Uncertainty approach makes sense in general.

Health Risk Assessment:

General comments:

Why would controlled exposure results not be useful for risk assessment?
Is risk assessment even warranted given the fact that concentrations are all below the
standard?

2.a-c.  Qualitative assessment is a good starting place. Agree with bronchoconstriction
findings. With respect to epi, I don't find any of the epi data compelling and robust for
SO2, although there are suggestions. The problem is that SO2 is too confounded by co-
pollutants, and the levels of SO2 are far far below levels that have ever been observed to
have relevant adverse effects in controlled studies.

3.a-e.  All very well justified approaches.

4.a.    No I do not agree with this. Unless I am mistaken, I don't think the document
includes a rationale for this decision.
4b.     More thought needs to go into deciding whether a tier III analysis would ever
make sense based on the epi  evidence alone.

Specific Comments:

p. 28, para 2, 4th numbered point: this one is a bit unclear; edit for clarity.

p. 29, para 3: although the ISA states that the SO2 effects were "generally" found to be
robust, this contrasts with my interpretation of the results presented in the ISA.
"sometimes" is a more accurate term to use  regarding SO2 robustness. Also, I take issue
with the ISA biological plausibility conclusion given the 2-3 order of magnitude higher
concentrations at which the lab-based findings are seen.

p. 33, section 4.4, 1st para: justification for the statement that controlled exposure studies
do not provide information to develop "credible exposure-response relationships" is
nowhere to be found in the supporting materials up to this point, including the ISA.  It is
particularly surprising given the extensive attention devoted to 5 minute concentrations in
the exposure work presented earlier. Why would one devote so much focus and effort on
short term SO2 if there were insufficient information to develop credible exposure-
response relationships ?  This argument needs to be laid out carefully and convincingly.
There may be a good argument, but it's not  here.

p. 34, 2nd para from end: Given all the other uncertainties in this assessment, it is
unreasonable to set the bar so high as to require same-location epi data before risk
assessment can be  conducted. It WOULD be preferable to have C/R data representative
of the region (e.g., NE US), but I don't think it is essential to require even this in order to
                                        38

-------
do an assessment. The list of uncertainties presented here are all real, but no more
problematic than those that appear in the exposure modeling for example.

p. 35, 2nd para:  it should also be recognized that multi-city results may not be optimal for
assessing effects in any one particular city and that uncertainties will be encountered if
this is done.

p. 35, 3rd para, at the end: Another option would be to rely only on C/R functions from
studies and models in which SO2 was included with co-pollutants AND where the SO2
effect was robust (i.e., the so2 effect did not change in going from single to multiple
pollutant models).

p. 36, section on criteria for determining approach (numbering of section seems off), first
bullet: need to be more clear about what is meant by "health effect benchmark levels
associated with current ambient conditions." Are you suggesting that you'll use epi
results to find thresholds?? You need to explain someplacehow such benchmarks would
be determined from the epi data, since apparently it is only the epi data that would inform
a tier III analysis.

p. 37, last line: Also should the proportion of the US population that is asthmatic,
outdoors, and exercising while ambient concentrations reach 5 minute peaks of concern.
                                        39

-------
Comments from Dr. Russell

First, a number of issues that arose from my reading the ISA as to how the information
from that document would be used down the road (e.g., in the Exposure and Risk
assessments) were answered to a reasonable degree. Indeed, there seemed to have been
additional forethought in the development of the Scope and Methods.  Thus, at the first
level, I am generally pleased.

In regards to the  air quality and exposure assessment, the greatest concern I have
continues  on from the ISA, that being that our current monitoring system likely provides
a relatively poor  characterization of the levels of SO2 typically found in an urban area.
The ranges of concentrations measured at any one monitor are very dependent on the
geometry, particularly distance and direction to the monitor. If a monitor is close to, and
from the direction of the dominant prevailing winds, of a major point source, it will have
higher concentrations, and likely higher PMRs.  On the other hand, a monitor further
away will  likely have lower PMRs. I wonder, would it not be easier to fit the data to a
log-normal distribution and/or develop correlations between the 5 min. max (and possibly
2" highest and third highest), along with the regression statistics, and use those to
develop the relationships?

In regards to Tier I exposure assessment, you note that you will be looking at estimating
peak 5 minute  levels at monitors... why not populations?

This document seems to suggest that there are 5 minute monitors and separate 1-hr
monitors.  This strikes me as strange.

One of the questions that staff will have to examine is the probability of capturing the
highest 5-minute average event in an area.  One of the problems with this is that the
observed PMRs are going to be highly dependent upon how close one is to the source,
and the coexistence of many sources in a region.  Thus, the PMR measured at a site may
not be very representative of the area under consideration. This gets back to the issue  of
exposure error: one (or a few) monitors may be relatively poor measures of both average
ambient air quality (much less personal exposure) and variability when the highest
concentrations are due to plumes with a relatively small spatial footprint.  It is interesting
that a peak SO2 level of 600 ppb was measured at a monitor... this would indicate that
the plume impacting that site had not dispersed much.  Doing a mass balance would
suggest that the plume cold not be very wide at that point.

The two statistical models (eq. 1 and eq. 2)  need to  be better explained, and the
ramifications of the choice of model form spelled out.  In eq (2), what are the individual
Pi's, and how are they calculated?  What do you mean that "m" is the number of peak
concentrations?  There can only be one.  At this point, one wonders why not just impose
a log-normal distribution? Indeed, I have a  suspicion that the effort being described on
top of page 20  can be done analytically if one assumes some sort of analytical form for
the cdf of  the PMRs and for the 1-hr concentrations. (I don't see any problem with such
                                        40

-------
an approximation, unless the data do not support such.)  One could see the possibility of
determining an empirical function where:
where c(t;) is the 5-minute average concentration for period i (i=l,12), and f is a function
of the average concentration and which period is being modeled. One would apply mass
balance constraints on the function and that it provides the observed distribution of
concentrations. f(. . .) would likely give something approaching a log-normal distribution,
where the geometric  standard deviation is a function of the observed concentration (or
maybe not, for simplicity: I worry that we might be trying to make this more complex
than need be).  I would have little  problem (others might) that if the limited data you have
can be reasonably fit with a log-normal distribution and that the gsd does not change that
much between locations, use the resulting analytical form for a national assessment. This
might provide a more efficient approach, and given that it would provide a more
statistically founded concentration structure, could arguably be better.

Page 11, last line: I think you mean reporting, not measuring.

In replying to the given questions:

    1 .  Is a proportional approach to just meeting more stringent standards reasonable?
          a.  Answer: Yes, with a condition.  The PRB is quite low, but it is likely that
              the application of controls to meet a more stringent standard will
              concentrate on large point sources, and very possibly ones near to the
              urban  area. Thus, PRB may not be the appropriate point to which to
              extrapolate.  Instead,  a low value resulting from a distributed set of small
              ubiquitous sources  may be better. Probably makes little difference, so I
              am not sure it is worth the effort to figure out what alternative vaule to
              use, but you might just mention this issue, (p.s.,  sorry about dickering
              about  the difference between linear and proportional.)
    2.  Should one extrapolate up  to just meeting the current standard or use historic air
       quality?
          a.  Answer: I would choose using historical data, though testing the
              difference at one or two locations. For one, this will  help lend confidence
              to how you extrapolate to meeting more stringent standards using the
              linear  rollback to the ubiquitous background concentration discussed
              above.

Exposure Analysis:

1 .      In considering the exposure analysis broadly:
       a.      Do Panel members have any comments on the general structure and
overall three-tier approach that staff plans to use for the exposure analysis? Are the
criteria that staff plans to use for deciding whether to conduct a  Tier II or Tier III analysis
clear and appropriate?
                                        41

-------
       Answer: Yes.  (The approach is fine.) EPA should compare and contrast their
approach to that used for other pollutants, and document why different methods are used.
Again, use each review to make the exposure and risk assessment a more systematic,
documented and turn-key.  One could see that in about three years (a couple more
pollutants) that a system much like that used for air quality modeling is used such that
with relatively little effort exposures, risks, variabilities, sensitivities and uncertainties
can be calculated, and the system as a whole has been intensely reviewed such that staff
need not spend such effort, and the community is more comfortable with the results.

       b.      Have the most important factors influencing exposure to SCh been clearly
accounted for and described?

Answer: I would tend to say yes, they have been discussed, but not accounted for in the
analyses.  As noted above, the potential for mischaracterizing SO2 levels, and area-wide
PMRs as they relate to exposures, is much higher for SO2 than other pollutants.

       c.      The draft plan describes the basis for and selection of population groups of
interest (i.e., children, asthmatics (children and adults), and the elderly) for which SCh
exposure estimates are to be developed. Do Panel members generally agree with the
groups of interest identified in the draft plan?

Answer: Yes.

    2.  In considering the Tier I exposure assessment:
       a.      Do Panel members agree a statistical model using available ambient 5-
minute data is appropriate... ?

Answer: It is a fine start.  One could conceive of a more comprehensive approach, but I
am not sure it is worth the effort (see above).

       b.      Do Panel members agree with applying a statistical model to estimate 5-
minute exceedences exist where only 1-hr monitoring is performed?

Answer: Mostly yes, but the uncertainty and sensitivity analyses should consider this
source of uncertainty and potential bias. Further, the model to be used needs to be better
explained and tested.

    3.  In considering a potential Tier II exposure assessment:

       a.      Do Panel members agree with the combined emissions/dispersion
modeling approach to estimate short-term (hourly) SO2 concentrations?

Answer: Yes,  as long as the model is evaluated and performance documented.

       b.
                                        42

-------
Answer:.

       c.     Do you agree with using PMRs to develop short term peak concentrations

Answer: See discussion above.

       4.  Is it reasonable to use APEX?

Answer: Yes, as long as the model is evaluated and performance documented.

   5.  Do Panel members have any comments or advice regarding the general approach
       to addressing uncertainty and variability in each Tier of the exposure assessment
       as described in the draft plan?

Answer: Provide, early on, results of some sensitivity analyses.  Do not overestimate
uncertainties going in.
                                       43

-------
Comments from Mr. Avol

SECTION 1
Pg 2, lines 5 and 6 - Won't the exposure and health risk assessments also
consider longer-term exposure (such as one-hour or 24hr), and not just five
minute exposures?

(Typo) P3, para2, line 4 - should be "...within a given day..."

(Typo) P3, para4, line 7 - should be "down", not "sown"

SECTION 2
(no comments)

SECTION 3
P8, last para - Given the importance of port-related SOx emissions (primarily
from bunker fuel or distillates in ship main and auxiliary engines), should identify
ships and ports as area worthy of closer scrutiny.

P15, bullet 2, first sentence - assertion that current ambient monitoring siting
captures anticipated occurrences of five-minute S02 peaks assumes emissions
are coming from electrical utilities and conventional historic sources... In terms of
5-minute averages, drifting plume touchdown points can become important; what
evidence is there that current monitors are properly sited for this?  What
evidence is there that current monitors are properly sited to capture 5-minutes
peaks downwind of ports and shipping lanes in coastal areas (both urban and
rural)?

P24, set of bullets - will these population sub-groups of concern be directly
drawn from the revised ISA, or will these four groups be the only ones chosen (in
other words,  what about other susceptible subgroups)?

SECTION 4
P28, paral, (4) - How does having time and resources to complete under the
schedule enter into this - isn't there a court-agreed-upon schedule to do it?

P29, paraS, Iine4 and Iine9 - The assertion that the ISA found S02 to be robust,
generally robust, after adjustment for PM and other co-pollutants should be re-
visited and clarified, based on the recent ISA discussions.

(Typo) p29, paraS, line 11 -"...in both the and human..." has a word missing or
an unnecessary "and"

P29, paraS, last sentence - Based  on John Balmes' comments regarding the
ISA, this conclusion about S02-induced altered lung host defense is incorrect.
                                    44

-------
P29, para 3 and para 4 - Given the discussion regarding clearer objective and
consistent definitions of the causal chain criteria, these comments about
"suggestive",  "plausible", "inconclusive", etc should be re-visited following
revision of the ISA.

P30, paraS, last sentence - is this comment about not focusing on long-term
exposure-related health effects consistent with the Agency's historical
perspective on S02? Didn't the previous review decline to establish a standard
for short-term exposures, in lieu of long-term exposure?  Does this create a
logical problem?

P31, paral, last sentence - doesn't the validity of this statement depend on the
correlation between 5 minute readings and 1hr or 24hr values...which we
apparently do not have thus far?

P36, last bullet - How can scheduling and resources be a viable rationale for not
going forward, if the framework for the decision to go to Tier III is in place and the
eligibility criteria are met?  Isn't this being unresponsive?

P37, last sentence - This presumes that living close to fossil-fueled emission
sources are the greatest concern, but these sources have been generally well-
controlled.  How about living close to a port or next to a rail line (burning fuel with
several thousand ppm S, if it is outside of California), or living along the coast by
a ship transit  corridor?
                                     45

-------
Comments from Dr. Samet

General Comments:
       In general, this document satisfactorily describes the approach that will be taken
to carry out the exposure assessment and risk assessment for SC>2. Initial links have been
made to the draft ISA. The document states explicitly that it will focus on the health
effects of SO2 and not consider secondarily formed particulate species. This makes the
clinical studies directly relevant to the risk assessment while implying that the risk
estimates from the epidemiological studies should be adjusted for PM; consideration
needs to be given to potential "double counting" of outcome events across PM and SC>2
risk assessments.
Exposure Assessment:
       The exposure assessment is quite detailed in describing the planned approach to
addressing the characterization of the frequency of short-term peaks and of human
exposures, including relevant subpopulations to these peaks. This aspect of the analytic
plan is well developed. I have the following specific comments:
   •   Given the possibility that the risk assessment may consider effects identified in
       epidemiological studies, should the plan be expanded to include longer-time
       frame concentrations and exposures?

   •   The uncertainty classification (given on page 7) needs development. What are the
       criteria for "minimal, moderate, and major" levels of uncertainty?

   •   Only a limited set of monitors offer information on 5-minute concentrations.
       Analyses directed at representativeness are needed.

   •   Can the exposure model incorporate the possibility that susceptible persons take
       behavioral measures to reduce exposures?

Risk Assessment:
       The document describes a three-tiered approach to the risk assessment.  General
criteria are offered for determining whether to undertake a Tier III assessment; I
recommend bringing greater specifity to the description of these criteria on page 28.
Specific comments are given below:
   •   The third paragraph on page 29 is confusing.

   •   Page 31 describes the approach to be followed for the acute responses, based on
       the clinical studies, and incorporating the "benchmark concentration". The text
       should be made more explicit. Is the scaling of risk linear in concentration?

   •   Page 35 states that"... staff judges that a C-R function estimated in the assessment
       location is preferable to a function estimated in some other location,..." This
       judgment warrants reassessment, particularly given the statistical instability of
       single city estimates. For those health outcomes with risk estimates available
                                        46

-------
from multiple locations, the aggregate estimate seems preferable, unless there are
particular characteristics of the location that would indicate that a locally derived
estimate is preferable.
                                  47

-------
Comments from Dr. Ultmann
Air Quality Considerations

    1.  Proportional approach seems reasonable to roll back current standard.
    2.  Proportional approach is more reasonable than historical approach, because of
       changes in conditions such as source distributions that could affect 1-hour mean
       and peak exposures in different ways.

Exposure Analysis
    1.  Considering the exposure analysis broadly, the two-tier approach is prudent and
       adequately explained in the document.
    2.  To the extent that quantity (and quantity) of the data is sufficient to provide
       reasonable power to the analysis, the application of a statistical model is a logical
       plan.
    3.  Considering a potential tier II exposure assessment, I am in general agreement
       with the approach, but because of the numerous uncertainties in dispersion and
       APEX models, I encourage that an attempt be made to validate the model
       simulations. The use of the PMR CDFS is reasonable given the low ambient 862
       levels and negligible background levels that would tend to linearize the values of
       the peak and the mean value.
    4.  I have no comments regarding the general approach to uncertainty.
                                       48

-------
Comments from Dr. Pinkerton

The introduction, overview and background on SC>2 NAAQS are extremely well written
and excellent in providing a framework for this document. These materials provide a
clear and necessary foundation to understand the basis for the scope and methods for
implementation of exposure and risk assessment for sulfur dioxide.

The organization and flow of the chapters and  content for the Scope and Methods for
Exposure Assessment and Risk Assessment Plan are extremely lucid and well written
with excellent figures and tables. The methods used for estimating 5-minute peak
concentrations are  clearly explained.  The APEX approach to simulate exposures that
occur in indoor, outdoor and in-vehicle microenvironments seems highly reasonable.
The selection of populations modeled to include  1) children (birth to 18 years), 2)
asthmatic children  (birth to 18 years), 3) asthmatic adults and 4) elderly (greater than 65
years) is also highly appropriate.

Air quality considerations for this document are particularly challenging in view of the
need to consider spatial and temporal levels of SC>2. A two-tiered approach as outlined
should be rigorously followed and fully implemented, with an adequate degree of liberal
application.  Such an approach should provide the necessary tools to provide for the
proper interpretation and analysis of exposure  scenarios that could be associated with
possible adverse health effects to better protect the public in the eventual decision to be
made for the next SO2 NAAQS.

The susceptible/sensitive populations have been adequately identified. However, due to
the type of exposure to  SC>2 involves in large measure in proximity to power plants,
consideration should also be applied to exposure by occupation associated with this
gaseous pollutant.

Annual and 24-averages from 1990 to 2006 clearly show a downward trend for SC>2.
Those exposed to the highest concentrations of 862 are limited select areas such as the
Ohio River Valley, the US Northwest and Hawaii.  Therefore, these areas require the
greatest consideration for health protection, rather than an average over multiple other
areas  experiencing relatively lower levels of SC>2.

Air quality considerations: In view of the anticipated low exposure levels, the use of a
proportional approach to adjusting air quality to simulate just meeting more stringent
alternative air quality seems to be a highly logical approach.

The discussion on variability and uncertainty is helpful to place into perspective variables
across time and space for individual exposure as well as assessment to estimate the
number of exceedances of alternative health effect benchmarks.

Minor comment:
                                        49

-------
Page 12: Figure 4 is referred to in the text as Figure 3.
                                         50

-------
Comments from Dr. Sheppard

Here is my summary understanding of chapters 3 and 4:
   •   Tier 1 exposure assessment is an exposure characterization aimed at predicting
       ambient concentration at the 5-minute level for use in a risk assessment based on
       evidence from controlled clinical studies.
   •   Tier II exposure assessment is an exposure characterization aimed at predicting
       personal exposure at the 5-minute level for use in a Tier III risk assessment.  It
       includes a dispersion model of concentration, but this is aimed at the 5-minute
       predictions needed for clinical studies rather than usual population exposure
       needed for epidemiological studies.
   •   Tier I risk assessment is a qualitative evaluation of health risks that provides
       preliminary input into the Tiers II and III risk assessments.
   •   Tier II risk assessment for controlled clinical study outcomes is a quantification of
       exposure aimed at determining the number of times the population would be
       exposed to benchmark doses.
   •   Tier II risk assessment for epidemiological study outcomes is an evaluation of
       exposure patterns to help with determination of population-level health effects of
       SO2.
   •   Tier III risk assessment for controlled clinical study outcomes were not discussed
       in chapter 4, but if done, based on information in chapter 3 they will use APEX
       predictions of 5-minute personal exposures.
   •   If done, Tier III risk assessment for epidemiological study outcomes will use
       concentration data from specific locations, incidence data estimated from the
       same locations, and concentration-response functions from single- and multi-
       pollutant models.

Discussion:
   •   The goals of the exposure and risk assessment aren't well aligned.  The exposure
       assessment focuses on exposures needed for controlled clinical studies while the
       bulk of the health assessment focuses on the epidemiological studies.
   •   Tier I exposure:
          o   Clarify the definition and discuss the prior analyses of PMR to clarify how
              these were done. Define the term "peak" -1 assume this is the maximum
              in the hour. Define whether a single random 5-minute measurement, the
              maximum 5-minute measurement, or all 12 5-minute measurements in
              each hour are used in the analyses presented and will be used for modeling
              planned.  The data in Figure 3 shows many PMR values below 1
              suggesting the 5-minute data are not actually peak measurements. Figure
              4 suggests a different set of data were used.
          o   Using different PMR distributions for different influential features is
              valuable. Both mean  1-hour concentration and proximity to source may
              contribute to the distribution of the maximum  5-minute concentration in
              an hour.
          o   Clarify the notation and wording in 3.2.1.4 as it is confusing.  (Is the
              model for an expected number (wording) or a probability(notation)?
                                        51

-------
       Incorporate the conditioning on influential features in the notation. Is r
       continuous? Why is P=l-p a cumulative probability and conditional on m
       but p isn't?
   o   I find some of the statements in 3.2.3 hard to believe. (Perhaps it is just
       semantics?) Quality of the monitoring data doesn't only depend upon
       quality of data collection, but also on the alignment of the data with the
       intended use.  In this case monitor siting features will affect the usefulness
       of the data. Temporal variability will also depend importantly on monitor
       siting.
   o   Looking ahead to the need for predictions of multiple measurements in an
       hour for use in Tier II, I question whether the PMR approach is
       appropriate.  Wouldn't it be more appropriate to develop the distribution
       of 5-minute data conditional on the hourly mean? This would then allow
       all the values of interest to be simulated directly from the distribution.
       (refer to section 3.3.2)
   o   P 14: Monitors are so sparsely sited that results of the analysis will need
       to be applied to unmonitored areas.  This is in contrast to the described
       approach which only  addresses populations that live in the vicinity of
       existing monitors.
   o   p 15 monitor siting:  Analyses recommended for the ISA and needed for
       modeling 5-minute data should inform this question and bound the
       uncertainty.  The key issue is how population exposure is represented by
       reliance on existing monitors.  It may be possible to make different
       modeling assumptions that directly take into account monitor siting.
   o   p 15 temporal representativeness: Clarify what the temporal profiles will
       be assumed to be representative of. Must this be assumed, or can an
       analysis be done to verify this assumption?
   o   p 15-16 spatial representativeness:  Once again this is a feature that should
       be able to be characterized rather than assumed.
   o   p 16 monitor to exposure representativeness: This bullet is confusing as
       written. This topic is not very clear in the ISA; improved ISA analysis
       and reporting may help with this HAP topic as  well.
Tier II exposure:
   o   Section 3.3.2 would not be so convoluted if the distribution of the 5-
       minute data were modeled given the 1-hour mean instead of modeling the
       PMR. Given the purposes of the modeling, it may not be important to take
       into account correlation of 5-minute measurements within the hour. This
       will certainly simplify the modeling.
   o   The description of the APEX model is good and EPA has had good
       success using this model for ozone health assessment. Refinements
       needed for SO2 modeling are new challenges.
   o   The combined uncertainty and variability analysis for APEX is
       commendable.
   o   In assessing APEX, direct comparison of predictions to measurements will
       be difficult because measurements are only for ambient concentrations
       while predictions will be for personal exposures.
                                 52

-------
    •   Tier I risk:
           o   I appreciate the explicit inclusion of qualitative risk assessment as its own
              tier.  This makes the entire process more open and transparent.
    •   Tier II risk:
           o   Separate 4.3.1 into separate subsections for controlled and epi studies.
           o   I don't understand the distinction being made between the national-scale
              analyses and urban-oriented analyses, (para 2 4.3.1)
           o   The evaluation of patterns of air quality data approach is ambiguous. This
              does not appear to be a risk assessment.
           o   Section  4.3.3  introduces a second level of tier II risk assessment in a
              section discussing variability and uncertainty.
    •   Tier III risk:
           o   Based on the  Tier II exposure discussion, the conclusion that there is
              insufficient information to develop the Tier III risk assessment for human
              clinical  studies does not  appear appropriate.
           o   Use the  section heading to clarify the description is for the
              epidemiological study results.
           o   I don't find differences in location of health studies versus risk assessment
              a compelling  enough reason to not conduct the risk assessment.
           o   Section  4.4.3: I suggest moving towards the unified uncertainty analysis
              approach described for exposure modeling in section 3.3.6.
    •   Determination of approach:  If at all possible, decision-making should be driven
       by scientific considerations as opposed to time and resources.

Other comments:
           •   p. 7: Regarding the choice of how many tiers will be done in the exposure
              and health assessments, decisions need to be open and transparent. I
              suggest  all tiers be addressed in the final document even if no analysis is
              done.
           •   P. 14: This approach will be used to simulate the probability of a short-
              term peak, not a realized value of a peak. Correct?
           •   P 19 before 3.3.2: Make sure  to clarify that output are predictions. Insert
              the word "predicted" before "SO2 concentrations".

Response to charge questions:

Air quality considerations:
2.  Recognizing that current  ambient air quality concentrations are lower than the
current standards,  the draft Health Assessment Plan discusses two alternative
approaches to simulating ambient SO2 levels associated with just meeting the current SO2
standards:  use of historical air quality data (e.g., possibly pre-2000) when ambient levels
were at or above the current standards, or use of a proportional (i.e., linear) approach to
adjust SO2 levels upward. Do the Panel members have advice or comments on these two
alternative approaches to simulating air quality just meeting the current SO2 standards?
I would use historical data.  I suggest limiting how far back to go in time, even if this
means using data that are below the current standard.
                                        53

-------
Exposure Analysis:
1. In considering the exposure analysis broadly:
a. Do Panel members have any comments on the general structure and overall two-tier
approach that staff plans to use for the exposure analysis? Are the criteria that staff
plans to use for deciding whether to conduct a Tier II analysis clear and appropriate?
Is the exposure assessment an analysis that should be done in isolation of the risk
assessment?  The planned uses of the analysis should be added to the criteria to conduct
the analysis.  Depending upon intended use, this could suggest a different  conclusion than
the other criteria, either less exposure analysis (e.g. if no risk assessment will rely on the
data) or more exposure analysis (e.g. if a population summary of the numbers of people
expected to experience certain exposures under certain activity conditions is needed).
b. Have the most important factors influencing exposure to SO2 been clearly accounted
for and described?
Should wind direction be incorporated?
c. The draft plan describes the basis for and selection of population groups of interest
(i.e., children, asthmatics (children and adults), and the elderly) for which SO2 exposure
estimates are to be developed. Do Panel members generally agree with the groups of
interest identified in the draft plan?
Yes. If possible the subgroup of responders should be represented. However it is not
currently known how to identify this subgroup.
2. In considering the Tier I exposure assessment:
a. Do Panel members agree that a statistical model using available ambient 5-minute
monitoring data is appropriate for estimating expected exceedances of very short-term
(5-minute) potential health effect benchmarks?
Analysis needs to be done to determine the comparability of the monitors with all 12 5-
minute averages in an hour vs. monitors with only the 5-minute maximum within an hour
vs. the monitors with no 5-minute data within the hour. If for the purposes of this
specific analysis, the data are comparable, or comparable subsets can be selected then I
agree it is appropriate to use the available 5-minute data for developing the model. As
noted above, I have questions about the notation in section 3.2.1.4., the basis for Figure 3,
and the plan to rely on the peak to mean ratio as the quantity of interest. I suggest instead
modeling the distribution of 5-minute averages  given the hourly mean  as an  alternative to
modeling the PMR.
b. Do Panel members agree with the approach of applying a statistical model to  estimate
5-minute concentration exceedances at monitoring locations where only 1-hour
monitor ing was performed for evaluating the extent of 5-minute peaks  associated with
meeting alternative standards with  longer averaging times?
No. See my  comments above questioning the need to model the PMR.
3. In considering a potential Tier II exposure assessment:
a. Do Panel members agree with the combined emissions/dispersion modeling approach
to estimate short-term (hourly) SO2 concentrations in close proximity to SO2 emission
sources?
Seems reasonable
b. Do Panel members have comments or advice regarding the described binning of
sources and development of prototype stacks/facilities?
                                        54

-------
 Seems reasonable
 c. Do Panel members agree with the approach using peak-to-mean ratio cumulative
 density functions (PMR CDFs) to estimate very short-term peak concentrations from the
 1-hour modeled concentrations?
 I question the decision to model the PMR, particularly when there is interest in multiple
 "peaks" within an hour. If peak means maximum, this concept doesn't even make sense.
 d. Do Panel members generally agree that the approach described using APEX is
 reasonable and appropriate to estimate the occurrence of very short-term (5 minute) SO2
peak exposures?
 I think 5-minute concentrations given the hourly average should be estimated/predicted
 directly from the distribution of 5-minute data given the mean, rather than focusing on
 the peaks. Is it reasonable to assume target ventilation/activity levels would not vary
 within a person over time?
 4. Do Panel members have any comments or advice regarding the general approach to
 addressing uncertainty and variability in each Tier of the exposure assessment as
 described in the draft plan?
 The unified uncertainty analysis for Tier II is good and I agree with the plan to
 incorporate it into Tier II.

 Health Risk Assessment:
 1. Do Panel members have any comments on the general structure and overall three-tier
 approach that staff plans to use for the risk assessment? Are the criteria that staff plans
 to use for deciding whether to conduct a Tier III risk assessment clear and appropriate ?
 I like the explicit addition of the Tier I qualitative risk assessment.  Criteria  for judging
 information suitable for a Tier III risk assessment shouldn't be so strict that  it limits the
 possibility of conducting a quantitative risk assessment. For instance, I think a criterion
 that the time series study estimates come from the same cities where they will be  applied
 is too strict.
 2. In considering the Tier I risk assessment:
 a. Do Panel members agree with the approach of having a qualitative assessment of
 health endpoints to identify which are likely candidates for a more sophisticated and
 quantitative tier of assessment?
 Yes. This section should also discuss endpoints that would inform a qualitative
 assessment but which would not progress to a quantitative analysis.
 b. Do Panel members agree with our initial observation that controlled human exposure
 studies demonstrate strong evidence for bronchoconstriction in exercising asthmatics
following 5-10 minutes SO2 exposure?
 Yes
 c. Do Panel members agree with staff's initial observation that the strongest
 epidemiologic evidence is for respiratory symptoms in asthmatic children and
 respiratory-related hospital admissions and respiratory-related emergency  department
 visits in asthmatics and others with respiratory conditions?
 Yes
 3. In considering the Tier II risk assessment:
                                        55

-------
a. In general, are staff plans to use potential health effect benchmarks to address
respiratory effects demonstrated in exercising asthmatics in controlled human exposure
studies clear and appropriate?
I don't understand why short-term peaks in the vicinity of SO2 sources won't be
considered since this is expected to be the major source of high 5-minute exposures, (sec
4.3.1 end of paragraph 2)
b. Do Panel members generally agree with the tentatively identified potential health
effect benchmark of 0.5 to 0.6ppmfor exercising asthmatics following 5-10 minutes SO2
exposure?
The ISA mentioned there were responders down to 0.25 ppm.
c. Do Panel members generally agree with the staff's approach of focusing on areas
around major sources of SO 2 with respect to concerns about 5-10 minute peak exposures
related to the respiratory effects observed in controlled human exposure studies?
The wording in section 4.3.1 of the document suggests the opposite.
d. Do Panel members generally agree with staff's approach of focusing on urban areas
with respect to concerns about 1- and 24-hr and annual SO2 concentrations related to
respiratory effects observed in epidemiologic studies?
Focus on urban areas for time series studies is appropriate, but I don't understand what
analysis will be done or reported.
e. Do Panel members have any comments or advice with respect to staff's approach of
gathering additional information to characterize the SO2 ambient air quality that existed
at the time various key U.S. and Canadian studies addressing respiratory effects were
conducted to see if the concentration-response relationships observed in these
epidemiologic studies are related to particular SO2 levels and associated averaging
times, geographic location and/or season, and the inclusion of various co-pollutants?
This section wasn't clear.
4. In considering a potential Tier III risk assessment:
a. Do Panel members generally agree that there is insufficient information to develop
credible exposure-response relationships for use in a quantitative risk assessment based
on the controlled human exposure evidence?
No
b. Do Panel members have any comments or advice with respect to the general approach
or specific factors to be considered in deciding whether or not to proceed to a Tier III
quantitative risk assessment for the respiratory-related health endpoints based on
epidemiologic evidence discussed in the draft plan?
Decision to proceed with a Tier III risk assessment for epidemiological studies of SO2
hinges  upon making the assumption that the concentration-response estimates from these
studies reflect a direct effect of SO2 on respiratory outcomes. This assumption can be
questioned based on one or more of: the ability to attribute the effect to SO2 given the
complex nature of the  pollution mixture, the often secondary role of SO2 in the study
analyses, and the ever-present challenge of making causal inference from observational
studies. Taking a precautionary approach, I suggest proceeding with the quantitative
estimates using both multi-  and single-pollutant SO2 models as is planned, and with
explicit recognition of the assumptions needed to even conduct the assessment.
                                        56

-------
5. Do Panel members have any comments or advice with respect to the general approach
to addressing uncertainty and variability in each Tier of the risk assessment as described
in the draft plan?
To the degree possible, move towards conducting a unified uncertainty analysis similar to
the one described in the exposure assessment section.
                                        57

-------
Comments from Dr. Balmes

In general, I think the approaches described in the document are appropriate to provide
important information that will be useful in the process of review of the NAAQS for
SOx.  I applaud the agency for the estimation of 5-min peak exposure data. This is a
relevant exposure metric given the consistent results from controlled human exposure
studies showing bronchoconstriction with such short duration exposures.  I also am
pleased that the agency is considering the use of emergency department (ED) asthma visit
data in the risk assessment. This is especially important given the relative weight of the
epidemiologic evidence of an association between SO2 exposure and asthma outcomes,
especially in children.  Asthma hospitalization data only capture the tip of the iceberg of
asthma morbidity and the addition of ED data allow a  greater proportion of the burden of
asthma to be assessed. I made this same point during the  ozone NAAQS  review and was
told that for the cities in which the exposure assessment was done, there were inadequate
asthma ED data to use in the risk assessment. For this review, more ED data should be
available - for example, California now has ED as well as hospital discharge data
available state-wide.
                                       58

-------