United States EPA Science Advisory EPA-SAB-EC-WKSHP-02-001
Environmental Board (1400A) January 2002
Protection Agency Washington, D.C. wwiv.epa.gov/sab
EPA Workshop on the Benefits of
Reductions in Exposure to
Hazardous Air Pollutants:
Developing Best Estimates of
Dose-Response Functions
An SAB Workshop Report of an
EPA/SAB Workshop
-------
NOTICE
This report has been written as part of the activities of the EPA Science Advisory Board,
a public advisory group providing extramural scientific information and advice to the
Administrator and other officials of the Environmental Protection Agency. This workshop
report, however, does not document an advisory committee meeting, organized with the purpose
of providing advice to the Agency. Instead, it documents a workshop open to the public and
announced in the FEDERAL REGISTER. This report has not been reviewed for approval by the
Agency and, hence, the contents of this report do not necessarily represent the views and policies
of the Environmental Protection Agency, nor of other agencies in the Executive Branch of the
Federal government, nor does mention of trade names or commercial products constitute a
recommendation for use.
Distribution and Availability: This EPA Science Advisory Board report is provided to the
EPA Administrator, senior Agency management, appropriate program staff, interested members
of the public, and is posted on the SAB website (www.epa.gov/sab). Information on its
availability is also provided in the SAB's monthly newsletter (Happenings at the Science
Advisory Board). Additional copies and further information are available from the SAB Staff at
the following address: US EPA Science Advisory Board (1400A), 1200 Pennsylvania Avenue,
NW, US Environmental Protection Agency, Washington, DC 20460.
-------
TABLE OF CONTENTS
1. BACKGROUND 1
2. MAJOR THEMES 4
APPENDIX A: LIST OF WORKSHOP PARTICIPANTS A-l
APPENDIX B: AGENDA B-l
APPENDIX C: WORKSHOP PROSPECTUS DISTRIBUTED BEFORE THE WORKSHOP
C-l
APPENDIX D: STRATEGIES FOR BRIDGING THE GAPS BETWEEN ECONOMISTS
AND HEALTH SCIENTISTS D-l
APPENDIX E: BACKGROUND PRESENTATIONS [Available on the SAB website
(www.epa.gov/sab) as part of this Workshop Report]
E-l. Presentation, "Current Approaches to Cancer and Noncancer Risk Assessment:
Implications for Developing Best Estimates of Dose-Response Functions,"
presented by Dr. William H. Farland
E-2. Presentation, "HAP Benefits Analysis in Section 812 Reports to Congress;
Briefing for SAB/EPA Workshop, June 22, 2000," presented by Mr. James
DeMocker
APPENDIX F: WHITE PAPERS [Available on the SAB website (www.epa.gov/sab) as part of
this Workshop Report]
F-l Dr. Lester Lave, Graduate School of Industrial Administration, Carnegie-Mellon
University, Pittsburgh, PA. Estimating the Benefits of Abating Toxic Air
Pollutants: What do Benefits Assessors Need from Risk Analysis - and Why are
they Unlikely to Get it?
F-2 Dr. Bernard Goldstein, Environmental and Occupational Health Sciences
Institute, Robert Wood Johnson School of Medicine, Rutgers, Piscataway, NJ.
Benzene White Paper.
F-3 Dr. Lorenz Rhomberg, Gradient Corporation, Cambridge, MA. Challenges in
Projecting Human Health Impacts from Exposures to Perchloroethlyene.
11
-------
F-4 Dr. Bernard Weiss, University of Rochester, Rochester, NY. Calculating the
Economic Benefits of Reductions in Manganese Air Concentrations.
APPENDIX G: WRITTEN SUBMISSIONS FROM KEY DISCUSSANTS [Available on the
SAB website (www.epa.gov/sab) as part of this Workshop Report]
Dr. Roy Alpert, Division of Environmental Health, University of Cincinnati
Dr. John C. Bailar III, Department of Health Studies, University of Chicago,
Chicago, IL.
Dr. Trudy Cameron, Department of Economics, University of California, Los
Angeles, CA.
Ms. Laurie Chestnut, Stratus Consulting, Boulder CO.
Dr. A. Myrick Freeman, Department of Economics, Bowdoin College,
Brunswick, ME.
Dr. Dennis Paustenbach, Exponent, Menlo Park, CA.
Dr. V. Kerry Smith, Center for Environmental and Resource Economics Policy,
Department of Agricultural and Resource Economics, North Carolina
State University, Raleigh, NC.
Dr. Lauren Zeise , California Environmental Protection Agency, Oakland, CA.
in
-------
1. BACKGROUND
Hazardous Air Pollutants (HAPs) include pollutants identified in the Clean Air Act;
pollutants known to cause or suspected of causing cancer or other serious human health effects,
such as birth defects, neurological damage, and respiratory disease.1 While it is clear that
reducing emissions of HAPs to the atmosphere will reduce exposure levels to chemical agents
that can cause serious health problems, there is no scientific consensus on the best way to
quantify such risk reductions for the purpose of a benefits analysis. Nor is there consensus on
how to value the reductions in risk, or what risk measures will be needed for valuation. The goal
of this workshop was to consider improvements to methods for estimating changes in health
risks resulting from regulations of HAPs that can be combined with valuation functions to
estimate monetized benefits of HAP reductions. Improved methods for assessing HAPs benefits
will assist the Agency in analyzing the economic value of its programs and in preparing reports
to Congress, such as analyses of the benefits and costs of the Clean Air Act (CAA), as required
by Section 812 of the CAA.
The workshop responded to a recommendation from the Health and Ecological Effects
Subcommittee of the Advisory Council on Clean Air Compliance Analysis in 1999. Members
and consultants from the Science Advisory Board (SAB) served individually on a committee
with Agency staff to plan the workshop, where Agency staff, SAB members and consultants,
experts outside the Board, and the public explored possible new methods for monetizing HAPs
benefits. EPA explicitly sought a broad spectrum of views at the workshop and did not seek a
consensus recommendation from workshop participants.
The workshop brought together expert discussants in the fields of economics, health
science, and risk assessment as related to managing HAPs, with the help of the workshop chair
and moderator, Dr. Michael Kleinman, College of Medicine, University of California, Irvine,
California. The workshop took a "case study" approach to address two main issues:
a) Whether it is possible to produce best estimates of the central tendencies and
distributions of the dose-response functions for a set of well-defined health
endpoints for each of the case-study HAPs for use in the future activities on air
quality and exposure modeling and how that might best be done.
b) How best to identify limitations and uncertainties in both risk assessment methods
and economic models with regard to changes in health risks from reductions in air
toxic emissions.
Section 112(b) of the Clean Air Act describes hazardous air pollutants as those pollutants "which present, or may present, through
inhalation or other routes of exposure, a threat of adverse human health effects (including, but not limited to, substances which are known to be,
or may reasonably be anticipated to be, carcinogenic, mutagenic, teratogenic, neurotoxic, which cause reproductive dysfunction, or which are
acutely or chronically toxic) or adverse environmental effects"
1
-------
Details about the workshop process, a list of participants and agenda can be found in Appendices
A-B.
The centerpiece of the workshop was a dialogue between economists and risk assessors.
Dr. Lester Lave, an economist from Carnegie-Mellon University, set the stage with a white
paper that documented his view of what benefits assessors need to know; why current
toxicological information is not suited to provide this information; and a research agenda that
would improve this estimation.
In response to his challenge, three distinguished experts prepared white paper case
studies (see Appendix F). Examining chemicals for which there are substantial databases on
health effects and exposure, the case studies were designed to illustrate the diversity of situations
among various HAPs:
a) Benzene: a case with substantial data on cancer and non-cancer effects and
strong hazard identification.2 Case presented by Dr. Bernard Goldstein,
Environmental and Occupational Health Sciences Institute, Robert Wood
Johnson School of Medicine.
b) Perchloroethylene: a case with substantial, but not as consistent, data on cancer
and data on noncancer hazards.3 Case presented by Dr. Lorenz Rhomberg,
Gradient Corporation.
c) Manganese: a case of a neurotoxin that causes many structural and functional
effects, including a condition similar to Parkinson's Disease, and whose effects
may accelerate the process of aging or onset of disease, albeit perhaps
significantly after exposure.4 Case presented by Dr. Bernard Weiss, University
of Rochester.
The White Papers can be found on the SAB website (www.epa.gov/sab) as Appendix F of this
Workshop Report.
These presentations were followed by a discussion where expert panelists (three
Although the first case (Benzene), exemplified a HAP for which there is substantial epidemiological and toxicological evidence, it
also highlighted some common areas of uncertainty in HAP risk assessments: extrapolation from health effect observations at higher exposures to
assessment of lower exposure scenarios (e.g., uncertainties regarding carcinogenesis and the shape of the cancer dose response curve at low
exposures); issue of "background" levels; cumulative burden; and other exposure-related issues.
The second white paper (Perchloroethylene) highlighted additional issues and raised the possibility that the conclusions drawn from
toxicological databases may lead us to conflicting or inconclusive results. Specifically, uncertainties in extrapolating health effect observations
from laboratory animals to humans were highlighted.
The third paper (Manganese) explored a very different terrain. In this case, the author created a conceptual model highlighting the
idea that specific studied endpoints may be markers for a broader set of conditions and that the challenge for economists may be valuing a shift in
the onset of risk or in the risk profile/trajectory of a population over their lifetime (after perhaps a long latency period).
-------
economists and three health scientists) each were asked to address the two central workshop
questions and reflect on the approaches proposed in the three chemical-specific white papers.
Most of the expert panelists, and the Workshop Co-Moderator, Dr. Roy Albert, expanded on
their brief presentations in written comments submitted after the workshop. These written
submissions can be found on the SAB website (www.epa.gov/sab) as Appendix G of this
Workshop Report.
-------
2. MAJOR THEMES
The three white papers and remarks by Dr. Lester Lave generated lively discussion but
no consensus on a proposed methodology. Many specific options emerged in the discussion and
in panelists' written comments. Appendix F lists many of the different strategies suggested for
bridging the gaps. There was, however, general agreement that this was a fertile area for further
study and that there was a need for further cross-disciplinary work between risk assessors and
economi sts/analy sts.
During the workshop discussion, it was noted that the different disciplines are pursuing
fundamentally different questions. In general, regulatory toxicologists are asking, "What level is
likely to be safe?" Economists, on the other hand, are asking, "What are the number of cases
reduced per small unit change in pollution concentration from individual regulations?"
Importantly, economists noted that they would like to have central tendencies and distributions,
rather than upper-bound estimates.
Although the discussants were in general agreement that there were benefits associated
with reductions in air toxics emissions, they were split in their opinion about whether using the
criteria pollutant model would work for air toxics. The case studies stimulated discussion about
the advisability of pursuing a pollutant-by-pollutant, endpoint-by-endpoint evaluation of
economic benefits.
The difficulties raised by the first two case studies were discussed. Namely, for the 188
listed HAPs, if we followed the criteria pollutant model, some difficulties would include:
limited health effect data; difficulties in modeling multiple assaults on a target organ;
contradictory or inconclusive evidence with respect to how many cases might develop or what
organ would be targeted in humans; difficulties in extrapolation from animal models;
uncertainties in extrapolating to lower doses; difficulties in evaluating background exposures
(especially for national policy applications); and lack of resources to evaluate fully each
chemical or class of chemicals listed in the Clean Air Act. The inability of economists to value
relatively small risk reductions from endpoints whose biological significance is difficult to
understand (e.g., change in platelet count) would hamper this approach even if the biological
data base were complete and yielded best estimates and a characterization of variability. It was
also discussed that the purpose of the workshop was to discuss one area (dose-response
estimates) but that there are uncertainties in all of the other analytical steps (e.g., emissions,
dispersion, exposure, and valuation). See Dr. Cameron's written comments (Appendix G) for a
fuller explanation.
Possible alternatives to the criteria pollutant approach were also suggested. One
suggestion was to evaluate "cleaning up the dirty stuff or thinking of the HAP regulation as an
insurance policy. Thus, economists would study the utility or value of peace-of-mind derived
from knowing that the public was protected from the bundle of effects, either known or
unknown, related to the listed HAPs. Another suggestion was to evaluate whether the public
-------
held a different value for full elimination of an involuntary risk versus small incremental changes
in that risk. A third promising possibility was to study the value of avoiding entry into a risk
pool or shifts in the curve of population's onset of disease (e.g., as suggested in Dr. Weiss's
paper). The Agency is considering introducing approaches similar to these proposals as
alternatives to a damage-function approach for assessing benefits for control of criteria
pollutants.
Although there was no consensus among the discussants, the research directions
suggested by this workshop suggest a dual approach. The first avenue would involve continuing
to address the HAPs using a damage function approach of estimating health effects avoided by a
given policy and determining the economic value of avoided effects. The first two case studies
(benzene and percholoroethlyene) lay out these issues very clearly. In addition, the comments
by Dr. Cameron and Dr. Zeise provide insights and research suggestions.
The second research approach might involve some of the other ideas discussed at the
workshop, including the concept of valuing the bundle of HAP reduction efforts embodied in the
Clean Air Act from an economic perspective (i.e., the utility or value of peace-of-mind derived
from knowing that the public was protected from the bundle of effects, either known or
unknown, as opposed to the avoided health effects), the approach discussed in the manganese
case study, and the insurance concept suggested in Dr. Smith's comments.
One of the invited discussants, Dr. Bailar, summarized the situation aptly by saying that
benefit-cost analysis is an information-hungry process which we apply to an information-sparse
problem with respect to HAPs. Regulatory decision-making informed by benefit-cost analysis is
a precision-hungry process that at present we base on precision-sparse inputs. The workshop
provided a meaningful first step, and more discussion and collaboration between the risk
assessment and benefits assessment communities is needed.
-------
APPENDIX A: LIST OF WORKSHOP PARTICIPANTS
CHAIR AND MODERATOR
Dr. Michael T. Kleinman, College of Medicine, University of California, Irvine, CA.
CO-MODERATOR
Dr. Roy Alpert, Division of Environmental Health, University of Cincinnati, Cincinnati, OH.
KEY DISCUSSANTS
Dr. John C. Bailar III, Department of Health Studies, University of Chicago, Chicago, IL.
Dr. Trudy Cameron, Department of Economics, University of California, Los Angeles, CA.
Ms. Lauraine Chestnut, Stratus Consulting, Boulder CO.
Dr. James Cogliano, Office of Research and Development, U.S. Environmental Protection
Agency, Washington, D.C.
Dr. James DeMocker, Office of Air and Radiation, U.S. Environmental Protection Agency,
Washington, D.C.
Dr. A. Myrick Freeman, Department of Economics, Bowdoin College, Brunswick, ME.
Dr. Paul Locke, Pew Environmental Health Commission, Baltimore, MD.
Dr. Albert McGartland, Office of Policy, Economics and Innovations, U.S. Environmental
Protection Agency, Washington, D.C.
Dr. Dennis Paustenbach, Exponent, Menlo Park, CA.
Dr. V. Kerry Smith, Center for Environmental and Resource Economics Policy, Department
of Agricultural and Resource Economics, North Carolina State University
Raleigh, NC.
Dr. Jeanette Wiltse, Office of Water, U.S. Environmental Protection Agency, Washington,
D.C.
Dr. Lauren Zeise , California Environmental Protection Agency, Oakland, CA.
A-l
-------
PRESENTERS OF WHITE PAPERS
Dr. Bernard Goldstein, Environmental and Occupational Health Sciences Institute, Robert
Wood Johnson School of Medicine, Rutgers, Piscataway, NJ.
Dr. Lester Lave, Graduate School of Industrial Administration, Carnegie-Mellon University
Pittsburgh, PA.
Dr. Lorenz Rhomberg, Gradient Corporation, Cambridge, MA.
Dr. Bernard Weiss, University of Rochester, Rochester, NY.
SAB STAFF
Dr. Angela Nugent, Science Advisory Board, U.S. Environmental Protection Agency,
Washington, D.C.
A-2
-------
APPENDIX B: AGENDA
SAB/EPA Workshop on the Benefits of Reductions in Exposure to Hazardous Air
Pollutants: Developing Best Estimates of Dose-Response Functions
June 22-23, 2000
Westin Grand Hotel
West 2350 M Street, NW, Washington DC
Welcome, Brief Overview of Workshop, and Introductions Michael Kleinman, University
Of California, Irvine
9:15.
Background
Current approaches to Cancer and Noncancer Risk Assessment William Farland, ORD
History of cost/benefit analysis for air pollution in general Albert McGartland, EPA
for Section 812 of the Clean Air Act Amendments James DeMocker, EPA
10:30 am
Review of Workshop Scope and Purpose Michael Kleinman
10:45 am
Economist's Perspective on HAP Benefits Analysis Lester Lave, Carnegie Mellon
Under Section 812
11:15 am
Discussion of Morning Papers Michael Kleinman, Moderator,
Key Discussants
All Attendees
12:00 Noon
Lunch
1:00 pm
Case Study Presentations
(For each: 20 minute risk assessor presentation of issues, 30 minute core panel discussion, 30 minute general
discussion [comments from all attendees]) Roy Albert, University of Cincinnati,
Discussion Moderator
1:00 pm
Benzene Bernard Goldstein, EOHSI, Rutgers
Key Discussants
All Attendees
2:20 pm
Perclorethylene Lorenz Rhomberg, Gradient Corporation
Key Discussants
All Attendees
B-l
-------
4:00 pm
Manganese
5:20 pm Adjourn
June 23. 2000
Bernard Weiss, University of Rochester
Key Discussants
All Attendees
9:00
Recap of Day 1 and Review of Workshop Agenda for Day 2
Michael Kleinman
9:15
Discussion of Questions Before the Workshop
Roy Albert and Michael
Kleinman, Moderators
Key Discussants
All Attendees
(1) Proposed approaches for hazard assessments for selected HAPs
that would facilitate benefit assessments for those chemicals
(2) Expert discussants' views on whether it is possible to produce a
methodology for developing central tendencies and distributions
in hazard assessments for HAPs for use in benefits analyses
and how that might best be done
(3) How best to identify limitations and uncertainties in both risk
assessment methods and economic models
(4) Suggestions and priorities for a research agenda to address
identified gaps in available data and methods needed to conduct
HAPs related benefit analyses
Lead Discussants: Paul Locke
Laurie Chestnut
Lead Discussants: Dennis Paustenbach
Rick Freeman
Lead Discussants: John Bailar,
Kerry Smith
Lead Discussants: Lauren Zeise,
Trudy Cameron
11:30 am
Summary and Identification of Next Steps
Michael Kleinman, Moderator
12:00
Adjourn
B-2
-------
APPENDIX C: WORKSHOP PROSPECTUS DISTRIBUTED BEFORE THE
WORKSHOP
SAB/EPA Workshop on the Benefits of Reductions in Exposure to Hazardous Air
Pollutants: Developing Best Estimates of Dose-Response Functions
Purpose
Hazardous air pollutants (HAPs) have been the focus of a number of EPA regulatory actions,
which have resulted in significant reductions in emissions of HAPs. EPA has been unable to
adequately assess the economic benefits associated with health improvements from these HAP
reductions due to a lack of best estimate dose-response functions for health endpoints
associated with exposure to HAPs and also due to the air quality and exposure models for
HAPs available for use in benefits analysis. EPA is conducting two activities to develop a
proposed methodology to generate estimates of the quantified and monetized benefits of
reductions in exposure to HAPs. The first will be a workshop focusing on developing best
estimates of dose-response functions that relate changes in HAP exposure to changes in health
outcomes. The second activity will focus on (1) integrating these dose-response functions with
appropriate models of HAP concentrations and human exposure and (2) translating these into
economic benefits that would estimate changes in health risks resulting from regulations that
reduce HAP emissions.
The overall goal of these two activities is to identify methods for the Agency to consider using
in estimating changes in health risks resulting from HAP regulations that can be combined with
valuation functions to estimate monetized benefits of HAP reductions.
Risk assessments for HAPs have been developed to help decision makers set health-based
standards that are consistent with EPA's mission to protect human health. The quantitative
toxicity values from these assessments (that is, the cancer slope factors and the noncancer
reference concentrations and reference doses) are typically based on animal and epidemiologic
studies that involve higher exposures than those encountered in the environment. The gap
between environmental doses and study doses has led to toxicity values that can put a bound on
the actual risk without being able to provide a reliable central estimate or distribution of risks.
It is these latter terms (central estimates and distributions) that economists have traditionally
used to estimate the economic value of potential changes in risks.
In contrast, risk assessments for criteria pollutants have been based on epidemiologic and
clinical studies of exposures similar to those encountered in the environment. This has allowed
development of standard statistical confidence intervals and distributions. With this
information, economists have been able to develop economic benefit estimates for many health
endpoints related to criteria pollutants. Criteria pollutant benefit estimates have been feasible
C-l
-------
because of the availability of: (a) well-defined health endpoints such as hospital admissions or
premature mortality; (b) dose-response functions from epidemiological and clinical studies
which support estimates of risk reductions in terms amenable to economic valuation; (c)
reliable estimates of ambient concentration and population exposure change; and (d)
dose-response functions available from epidemiological and clinical studies in which the
exposures were similar to those being experienced in the ambient environment. Uncertainties
related to the health benefits of criteria pollutants have generally been represented by standard
confidence intervals based on measures of within and between study variation in the estimated
health effects.
While mortality from HAP-related cancer is a well-defined endpoint, there are very few
validated exposure-response relationships. For the many other potential health effects from
exposure to HAPs, such as changes in reproductive functions or mutagenic effects, there are
major information gaps in all aspects of risk assessment, as well as in exposure-response and
valuation. The focus of this workshop will be the development of best-estimates and
uncertainty characterizations for hazard and dose response functions for use in benefits
analyses of HAP regulations, with a focus on providing potentially useful data and tools to
support HAP-related benefit assessments, including national-scale program evaluations.
Expected outcomes from this workshop will include a report documenting: (1) proposed
approaches for hazard assessments for selected HAPs that would facilitate benefit assessments
for those chemicals; (2) expert discussants' views on whether it is possible to produce a
methodology for developing central tendencies and distributions in hazard assessments for
HAPs for use in benefits analyses and how that might best be done; (3) h ow best to identify
limitations and uncertainties in both risk assessment methods and economic models; and (4)
suggestions and priorities for a research agenda to address identified gaps in available data and
methods needed to conduct HAPs related benefit analyses.
Scope
The workshop will task expert discussants, who have knowledge and expertise in the fields of
economics, health science, and risk assessment as related to managing HAPs, to address the
following issues:
I. Whether it is possible to produce best estimates of the central tendencies and
distributions of the dose-response functions for a set of well-defined health endpoints
for each of the case-study HAPs for use in the second activity on air quality and
exposure modeling and how that might best be done.
II. How best to identify limitations and uncertainties in both risk assessment methods and
economic models with regards to changes in health risks from reductions in air toxic
emissions, especially in the following areas:
C-2
-------
A. Defining and characterizing best estimates and uncertainty distributions for
hazard and dose response functions for both cancer and non-cancer effects.
B. Defining the context for the use of conservative reference doses, including the
basis and methodology of the risk assessment.
C. Identifying potential uncertainties from extrapolating effects from "high dose"
occupational and animal exposure studies to lower level ambient HAP
concentrations.
D. Incorporating currently/typically available toxicity (both human and animal) data
sets into existing or modified economic benefit models.
E. Evaluating the usefulness of benefits models based on dose-response functions
derived from epidemiological studies as models for HAP benefit analyses.
III. Identifying gaps in existing knowledge and developing a proposed research agenda.
C-2
-------
APPENDIX D: STRATEGIES FOR BRIDGING THE GAPS BETWEEN ECONOMISTS
AND HEALTH SCIENTISTS
The Workshop resulted in many suggestions to address the gaps between economists and
health scientists to improve benefits assessments for HAPs. These suggestions included the
following:
1. Analyze the implications of current in vivo and in vitro tests for human toxicity for all
known human carcinogens. Compare laboratory results with human data; for chemicals
that are positive in the NTP bioassay, identify which are and which are unlikely to be
human carcinogens. (Lave)
2. Invest time and resources in developing expert judgment on the best estimate for a
chemical, like benzene, with substantial data on cancer and non-cancer effects and strong
hazard identification but limited data at low doses. (Goldstein)
3. Acknowledge that data sets have major uncertainties. Invest in statistical approaches to
characterize the distribution of uncertainties, even for chemicals with significant
uncertainty in their data sets. Approaches would establish explicit weights for different
assumptions associated with estimation of risk. Approaches would allow examination of
the contribution to overall uncertainty from various components. (Rhomberg)
4. Link endpoints in toxicology studies to adverse effects the public cares about.
Communicate these linkages and associated uncertainties to economists conducting
benefit assessments. (Weiss)
5. Estimate economic value of avoiding entry into a risk pool or shifts in the curve of
population's onset of disease. (Weiss)
6. Multiply the estimated number of cancer cases by a weighting factor that is determined
by the strength of the evidence. (Albert)
7. Educate congress, the public, and the news media about "the art of the possible" in risk
assessment and the limits of science. (Bailar)
8. Establish closer links between risk analysts and cost-benefit analysts. Establish cross-
training programs. Establish regular, weekly meetings on each project in which risk
analysis will be a significant element of a cost-benefit analysis. (Bailar)
9. Analyze the level of accuracy needed at each step of analysis to make most effective use
of limited resources. (Bailar)
10. Consider "bundling" or grouping HAPs that are similar, either according to health
endoint, chemical species, biologic mechanism or mode of action, or source categories.
(Bailar)
D-l
-------
11. Keep in mind that there were uncertainties not only in dose-response estimates that feed
into cost-benefit analysis, but also in all of other analytical steps (e.g., emissions,
dispersion, exposure, and valuation). (Cameron)
12. Consider a "top down" approach to benefits assessment, derived from individuals'
subjective assessments of what HAP regulation would be likely to achieve in terms of
health effects (or other effects) instead of aiming for "an unambiguous, objectively
calculated, bottom-up measure" calculated on a chemical-by-chemical basis. (Cameron)
13. Develop a benefits assessment based on a "better understanding of the relationship
between controls on emissions of HAPs and individuals' perceptions of safety or 'peace
of mind.'"(Freeman, Lave, Smith)
14. Sort HAPs into "bins" according to the adverse effects of concern, and conduct benefit-
cost analyses for different "bins." Distinguish chemicals listed due to their
carcinogenicity, from chemicals listed because they are systemic (non-carcinogenic)
toxicants, from irritants. For each group, develop a dose-response curve and build a
distribution of points around the dose-response curve (for both carcinogens and non-
carcinogens). (Paustenbach)
15. Invest in research to improve understanding and quantitative descriptions of how risk
may vary in a population. This research would produce better mean estimates of risk and
clearer understanding of the magnitude of risk borne by different individuals. (Zeise)
D-2
-------
Appendix E-1
"Current Approaches to Cancer and
Noncancer Risk Assessment:
Implications for Developing Best
Estimates of Dose-Response
Functions,"
Presented by Dr. William H. Farland
-------
An on the of
in to Air
22 and 23,
for
of
^
OS EM «rf JMf
H. Ph.D.,
for
of
U.S.
Bafioreii Center for Era
-------
"The quality of risk analysis will improve as the
quality of input improves. As we learn more
about biology, chemistry, physics, and
demography, we can make progressively better
assessments of the risks involved. Risk
assessment evolves continually, with
reevaluation as new models and data become
available."
-Science and Judgment in Risk
Assessment" (National Research
Council, 1994)
-------
Dichotomy
Cancer
Non-Cancer
Non-Threshold
rreversible
"Risk" value
+ Slope Factor
+ Unit Risk
* Risk-Specific
Dose
Threshold
Reversible
"Safety" value
+ RfD/RfC
+ ADI/TDI
+ MRL
-------
Systematic Characterization of Comprehensive
Chemical
Exposure
Concentration
"Dose"
Protective Exposure
Default
Exposure
Mechanisms
Disposition Models
Exposure
Mechanisms
Toiicant
Tissue
I
Mechanisms
Disposition Models
Toxicant-Target Models
Toxicological
Response
Response Qualitative
Response
Response
Predictive
Exposure
Mechanisms
Tlssyt
Response Quantitative
Mechanisms
Mechanisms
Disposition Models Toxicant-Target Models Tissue Response Models
-------
Exposure and Disease
vS>-<5>-<5>-<5>-<5>-<5>-<5>-<5>-<^^
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\^\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/
^
-------
o Emphasize full characterization
o Expand role of mode-of-action
information (and, therefore, biomarkers!)
o Use all information to design dose
response approach
Two step dose response assessment
-------
Characterization
of
and
-------
Definition:
Biologic markers are indicators
signaling events in biologic systems
or samples.
Three types:
Exposure
Effect
Susceptibility
-------
Mechanism of action:
Detailed molecular description of a key
event in the induction of cancer or other
health endpoints
Mode-of-Action:
Key events and processes, starting with
the interaction of an agent with a cell,
through functional and anatomical
changes, resulting in cancer or other
health endpoints
-------
How does the chemical
produce its effect?
Are there mechanistic data
to support this hypothesis?
Have other mechanistic
hypotheses been considered
and rejected?
-------
Address Uncertainty in Risk
Assessment:
• Comparative Structure Activity
Relationships (SAR)
• Relevance of animal data for extrapolation
• Shape of dose-response curve
Range of Observation
Range of Inference
Susceptibility of individuals/
subpopulations
-------
To show that a postulated mode-of-action is
operative, it is generally necessary to:
3 outline the sequence of events leading
to effects;
3 identify key events that can be
measured; and
3 weigh information to determine
whether there is a causal relationship
between events and cancer formation.
-------
Summary Description of Postulated
Mode-of-Action
Topics:
1. "Identify key events" (-> BIOMARKERS?)
2. "Strength, consistency, specificity
of association"
3. "Dose-response relationship"
4. "Temporal relationship"
5. "Biological plausibility and coherence"
Conclusion
-------
Examples:
Metabolism
Receptor-ligand changes
DMA or chromosome effects
Gene transcription; protein synthesis
Increased cell growth and organ weight
Hormone or other physiological
perturbations
Hyperplasia, cellular proliferation
-------
• Formaldehyde
• Methylene
Chloride
• d-Limonene
• Chloroform
• Dioxin
DMA crosslinks
Cell proliferation
Pharmacokinetics
Genetic polymorphisms
• «2-u-globulin, etc.
Cytotoxicity
Receptor-mediated
responses
-------
• BaP
• Amitrole
Melamine
Perchlorate
Vinyl Acetate
DMA reactive metabolites
Cell proliferation
Increased Thyroid
Stimulating Hormone (TSH)
Cell proliferation
Increased urinary pH
Irritation
Altered thyroid homeostasis
Cytotoxicity
Cell proliferation
-------
Risk Assessment of Perchlorate
Exposure
p
^
i
•
•
1
"
"W
•«
4^
•
•
•
^ ^
In
Biologically
!=• P
Dose
4^
i
•
•
\
'
"
\ in /'
% %
%% %
•
•
Eariy
» 1
Effect
^
i
•
• XT4>^T3
•
•
•
•
:: 1 J
Ez^EiiiEE
^.. .
Clini
» », Djsei
Fy net ion
i
cal
ise ^
A 4^ 4^
111
111
111
Thyroid x ^' _
* , / ^, Turre
'' fckl" Hhman
>rs y '"X Risk
^ •'' N- i,,- '' _. .. . , ^"' i,^ •••'"Children's
" vpe«topment,, ^ .,_Birth - ^ ,^_ Risk
* ^ •**•
w ^ ^
'
t + +
* + +
* + +
^ ^
-------
Exposure
cer
Internal
Dose
Effective
Dose
Early
Effect
Altered
I
Function
Clinical
Disease
Prognostic
I
1^1 JiJ &
^
M
M I
% * %
% %
' +*
-------
interpretation and
extrapolation of:
-------
of to
"
"
"
"
of
#
-------
10,000
Trials
0.20
12S.4F
0.00
0.00
0.00
0.00 0.03 1800.00 3605.00
Concentration Frequency
0.15
0.10
0.93
0.00
0.06
0.00
0.42 4.3! 1.00 128.52
Duration
0.05
0.00
25.00
0.05
0.00
55.78_ , ,.,.,. 108.70
50.00
Model Result
-------
extrapolation of:
•
-------
o Linear
o Sublinear
o Supralinear
o U-Shaped
-------
Probabilistic Estimate of
Upper Bound on Risk
Margin-of-Exposure (M-O-E)
Reference Dose (RfD)
Benchmark Dose (BMD)
NOAEL/LOAEL
-------
11
IB
C
S
C
O
£L
m
m
m
M
m
I
1
e
0.1
\
to
ED
10
-------
• Tumor data
• Pharmacokinetics and
metabolism data
• Data on effects of agent on
carcinogenic processes
-------
§
o
Q.
if!
£
o?
~r
X
/
/
from Heitzmann
and Wilson (1997)
.X
c
n
/ 6inc(high dose)
6jnc(Iow
V
fh
-------
CP
111
c
o
Q.
if!
from Heitzmann
and Wilson (1997)
-------
o Construct a biologically-based or case
specific model
o Link dose response curve for precursor
effect to dose response for tumor effect
o Use dose response for other effect in lieu
of that for tumor effect if it is judged to be
a better measure of potential risk
o Use to inform assessment of possible
dose response in range of extrapolation
-------
Assessment
O First step
+ Data in range of observation
Second Step
Evaluation in range of human
exposure (Extrapolation,)
-------
I
Q
5-i
o
I
+->
8
(H
O
Pi
o
10%
0%
Environmental
Exposure Levels
of Interest
Empirical
Range of
Observation
Range of
Extrapolation
ED
10
MOE-
Nonlinear default
Dose
-------
Current Risk Assessment Approaches Raise the
Following Issues:
"=> Characterization of subtle, low response
biomarkers; protective vs. predictive?
•=> Response biomarkers will be surrogates for
effect or multiple effects rather than the effect
of concern itself
•=> Additivity to background (exposure, response)
may be important to address where exposure
of interest lies on the dose-response curve
Outputs are likely to be ranges or distributions
-------
Development/validation of sensitive tools
aimed at understanding mode-of-action
Incorporation of "Framework" Concept
More Attention to Route-Specific/
Situation-Specific Characterizations
Addressing Sensitive Subpopulations
-------
Assessment
Refine estimates of dose to relevant targets
through use of biomarkers of exposure
Improve hazard characterization through
use of biomarkers of response with
mechanistic linkage to endpoints of concern
Strengthen inferences regarding the shape
of dose/response curves outside the range
of observation
Identify targets of opportunity for further
study in potentially sensitive human
populations
-------
Appendix E-2
"HAP Benefits Analysis in Section 812
Reports to Congress; Briefing for SAB/EPA
Workshop, June 22, 2000,"
Presented by Mr. James DeMocker
-------
HAP Benefits Analysis in §812
Reports to Congress
Briefing for SAB/EPA Workshop
June 22, 2000
Jim DeMocker, EPA/OAR
-------
•812 Benefit-Cost Analyses
Analytical Requirements
"(a)...The Administrator shall conduct a comprehensive
analysis of this Act on the public health, economy, and
environment... [which] should consider the costs, benefits
and other effects...[of] each standard issued for... (2) a
hazardous air pollutant listed under § 112, including any
technology-based standard and any risk-based standard..."
"(b).. .The Administrator shall assess how benefits are
measured in order to assure that damage to human health
and the environment is more accurately measured and
taken into account."
-------
•812 Benefit-Cost Analyses
Review Requirements
"(f)..-The Administrator shall appoint an Advisory
Council... [consisting of] recognized experts in... health
and environmental effects of air pollution, economic
analysis, environmental sciences, and other [appropriate]
fields."
"(g).. .The Council shall review... the data... the
methodology... and the findings of such report, and make
recommendations to the Administrator concerning the
validity and utility of such findings."
-------
812 Benefit & Cost Estimation
'Retrospective Study"
- Submitted to Congress October 1997
- Direct costs aggregated and fed to macro model
- Benefits by pollutant as data and models allowed
'Prospective Study"
- Submitted to Congress November 1999
- Direct costs estimated by title / major provision
- Benefits by pollutant as data and models allowed
-------
Retrospective Study
Stationary Source Pollutants
14 key HAPs from Cancer Risk Study (1990)
arsenic
asbestos
benzene
1,3 -butadiene
carbon tetrachloride
chloroform
chromium (VI)
dioxin
ethylene dichloride
ethylene dibromide
formaldehyde
gasoline vapors
product of incomplete
combustion (PICs)
vinyl chloride
-------
Retrospective Study
Stationary Source Method
Incidence change assumed proportional to emissions change
by = base year (85)
ty = target year (70, 75, 80, 90)
incidence (from CRS) by = base year (8.
-- activity (from macro model) ty = target year ('
-- population
z control efficiency (from CTGs, BIDs, regs, experts)
-------
Retrospective Study
Stationary Source Findings
Lower Bound
for Other HAPs
75 SO SS fO
Upper Bound
for Other HAPs
80 85
-------
Retrospective Study
Stationary Source Review Issues
Estimated incidence for vinyl chloride and asbestos
much higher than historical incidence
Cancer Risk Study designed for only rough order-of-
magnitude estimates
- Unit risk factors are upper-bound estimates
- Exposure estimates are typically upper-bound (MEI)
Control efficiencies assumed uniform across facilities
and 100% compliance with regulations
-------
Retrospective Study
Report to Congress Presentation
HAP benefits excluded from primary analysis -
described in Appendix
Quantitative analyses with caveats
- Stationary source cancer incidence reduction estimates
- Motor vehicle exposure reduction estimates
Qualitative discussions
non-cancer health effects
ecosystem effects
-------
Retrospective Study
Report to Congress
~ Health Research Recommendations ~
Address additional pollutants
Address mechanisms with pharmacokinetics
Address variations in human susceptibility
Address interactive effects of multiple exposures
Develop alternatives to cancer upper-bound methods
Develop D/R relationships for non-cancer effects
Develop methods for acute exposure effects
-------
Retrospective Study
Report to Congress
~ Exposure Research Recommendations ~
1 Expand data collection: control efficiencies, HAP
speciation, facility locations and operating
parameters
• Develop more comprehensive exposure models
• Refine uncertainty analysis methods
-------
Retrospective Study
Report to Congress
~ Ecosystem Research Recommendations ~
• Estimate levels of bioaccumulating toxics in media
• Correlate levels of bioaccumulating toxics with
exposures, concentrations, and adverse effects
• Develop wildlife correlate to RfD or D/R
relationship
• Address effects of mixtures
• Address additional ecosystems
• Address wetland species and functions
-------
Retrospective Study
Report to Congress
~ Valuation Research Recommendations ~
Address additional endpoints consistent with
kinds of damages expected
Initiate broad-scope economic valuation using
survey techniques
-------
Prospective Study
Methodology Alternatives Presented
— National Scale —
Assessment System for Population Exposure
Nationwide (ASPEN)
- Emissions inventory
* multiple pollutants
Air dispersion model
• point, area, and mobile source categories
- Exposure model (not completed)
-------
Prospective Study
Methodology Alternatives Presented
— National Scale —
Advantages
- Includes treatment of
• reactive decay (simplified)
• secondary formation (simplified)
• long-range transport (continental scale)
• wet and dry deposition (parameterized)
- Emissions/Dispersion well documented:
• sensitivity analysis
• model performance evaluation
• uncertainty analyses
-------
Prospective Study
Methodology Alternatives Presented
— National Scale —
Limitations
- National emission inventory uncertainties
- Gaussian model limitations
- Meso-scale transport not addressed (50 - 200 km)
- Re-suspension not addressed
- Not stochastic
- Spatial and temporal peaks not addressed
- Indoor sources not addressed
- Indirect exposures not addressed
-------
Prospective Study
Methodology Alternatives Presented
— Local Scale / Case Study —
Air Quality Integrated Management System
(AIMS)
Developed for Baltimore and planned for Houston
and Chicago
Integrates routinely collected data (measured air
quality, emissions, and meteorological data) and
dispersion modeling
-------
Prospective Study
Review Issues Raised
Resources for in-depth analysis for 188 HAPs
prohibitive: find priority HAPs
Unit Risk Factors are upper-bound estimates
Limited ambient monitoring data to validate
ambient concentration estimates
Exposure assessment limitations
- 50 km downwind distance for dispersion
- lack of attention to indirect pathways (e.g., Hg, dioxin)
- ASPEN preliminary performance evaluation concerns
-------
ASPEN Model Performance
1990 Carbon Monoxide
2500
2000
n
~
1500
o o
Ol
o
c
o
O
•a 1000
01
01
500
o _ c
• West R= 0.66
• Northeast R= 0.65
Southeast R= 0.38
o Central R= 0.23
500
1000 1500
Predicted Concentration (ug/m3)
2000
-------
Prospective Study:
Report to Congress Presentation
No quantified benefits
Expect benefits from MACT and incidental to
criteria pollutant control
Besides cancer inhalation impacts, other potential
benefits include reductions in:
- Non-cancer health effects
- Indirect non-inhalation exposure
- Ecological and welfare effects
-------
Prospective Study
Report to Congress Research Recommendations
• Workshops to address HAP benefits challenges:
- toxicology/risk assessment
- exposure assessment
economics
Investigate use of EPA's Air Toxics Data Archive
of measurement data from state / local programs
Explore whether "supersite" monitoring programs
can provide HAP ambient concentration data
-------
Future §812 Studies
Pondering potential scope, objectives, and
reference period for "812 III"
Detailed analytic blueprint to be developed, and
HAP Workshop outcomes will be considered
SAB Council and HEES will be asked to review
analytical blueprint prior to initiation of work
-------
-------
Retrospective Study:
Mobile Source Analysis
— Methods ~
Based on Motor Vehicle Related Air Toxics Study
(1993)
Exposure estimated for CO using measured
concentrations and HAPEM-MS
Exposure to HAPs assumed proportional to emission
factors
-------
Retrospective Study:
Mobile Source Analysis
— Methods and Data—
= ((AxC)- B)xM x Sx
VOC x HAP
E = exposure concentration
A = annual average CO ambient concentration (AIRS)
C = CO ambient to CO exposure concentration ratio (HAPEM)
B = CO background concentration (reported measurements)
M= fraction of CO emissions from mobile sources
S = scenario-to-control scenario CO emission factor ratio
VOC = VOC mobile emission factor by, scenario/year
HAP = HAP speciation factor for mobile source VOC, by scenario/year
CO = mobile source emission factor, by scenario/year
-------
Retrospective Study:
Mobile Source Analysis
— Findings —
te
ft
X
I
-------
Future 812
Tools Needed
Expanded air toxics monitoring data
- 90 new monitors by end of FYOO
- Air Toxics Data Archive to supplement AIRS with state
and local data
Improved emissions inventories
- 1996 National Toxics Inventory (NTI)
Evaluation/enhancement of air quality and
exposure modeling tools
Expanded risk data and improved methods
-------
Future 812
Risk Assessment Workshop
Current risk assessment state-of-the-art
- Probabilistic estimates for cancer
- Reference doses/concentrations for non-cancer
• More sophisticated D/R assessments for some criteria
pollutants
- Mixtures
Sum of upper-bounds for cancer
Hazard index for non-cancer
-------
Future 812
Risk Assessment Workshop
Recent trends in risk assessment
Cancer: mix of probabilistic (no threshold) and
reference concentrations (threshold)
Non-cancer: modeling and distributional approaches
- Dosimetry models focused on tissue concentrations
-------
Future 812
Risk Assessment Workshop
Potential sources of bias in risk estimates
- Linear high-to-low dose extrapolation
- Cross species scaling factor
- Treatment of untested chemicals and other data gaps
Latent effects
Use of most sensitive test results
Non-cancer uncertainty factors
Magnitude and severity of effects
Route-to-route extrapolation
Benchmark response rate (LED 10 instead of NOAEL)
Additive treatment of mixtures
-------
Future 812
Risk Assessment Workshop
Uncertainty in risk estimates
- Types
1 Causal link between exposure and effects
Magnitude of risk
Can use analysis of quantifiable uncertainty to develop
central risk estimate
Unquantifiable uncertainty may still lead to bias
• use of sensitive species
• consideration of non-relevant effects
-------
Future 812
Risk Assessment Workshop
Topics for discussion
- How to characterize a distribution of risk estimates as
an input to benefits assessment
- How to characterize the value of reducing exposure in a
reference dose framework: proportion of people above
RfD, contingent valuation, other?
- How to characterize benefits when uncertainty is great:
point estimate, range, other?
- Are some benefits better left unquantified?
-------
APPENDIX F-l
Estimating the Benefits of Abating Toxic Air Pollutants: What do Benefits Assessors Need
from Risk Analysis - and Why are they Unlikely to Get it?
White Paper by Dr. Lester Lave,
Graduate School of Industrial Administration,
Carnegie-Mellon University, Pittsburgh, PA.
F-l-1
-------
Estimating the Benefits of Abating Toxic Air Pollutants: What Do Benefits Assessors
Need From Risk Analysis - and Why are They Unlikely to Get It?
Lester B. Lave 5-22-00
Carnegie Mellon University
Abstract
I examine what benefits assessors require to estimates of benefits and costs of abating
hazardous air pollutants, why current toxicological information is not suited to benefits
estimation, and a research agenda that would lead to confident estimating of the benefits
of abatement. While it has many limitations and uncertainties, epidemiology data
provides information for benefits assessment. In contrast, regulatory toxicology is
focused on protecting people from harm, not on giving best (unbiased) estimates of the
harm to people from various exposures. Few toxicology tests have been validated in the
sense showing that they predict human toxicity. Other than reasoning from first
principles, we have no way of knowing which toxicology tests are best are predicting
human toxicity and, if so, how much. Fundamental changes in toxicology are required to
get data for benefits estimation. In my judgment, the pressure for estimating the benefits
and costs of abating hazardous air pollution is putting the right sort of pressure on the
regulatory environment These pressures will force changes in toxicology that will
improve identification and quantification of human toxicity.
Introduction
My first task is to inform risk analysts what economists need to estimate the benefits and
costs of abating hazardous air pollutants. That is a straightforward task. My second task
is to tell economists why the current toxicology information that they are getting is not
capable of estimating these benefits. My third task is to open a dialogue between risk
assessors and benefits estimators that might some day result in confident estimates of the
benefits. My final task is to suggest a research agenda leading to confident estimates of
the benefits of abating hazardous air pollutions. To illustrate the first task, I begin with a
review of a large EPA benefit-cost analysis.
Assessing the Benefits and Costs of the Clean Air Act
Assessing the benefits and costs of an environmental program, such as the retrospective
and prospective effects of the 1970 Clean Air Act, requires information beyond what is
factually available. Estimating the costs of the act requires estimating the private and
social costs, including abatement costs and opportunity costs, of the regulations. This
cost must be contrasted with the costs that would have been incurred if the Clean Air Act
and associated regulations did not exist. The actual costs incurred are difficult to
estimate; the costs that would have been incurred absent the act cannot be observed or
even estimated with confidence, since the world of no regulation does not exist. The cost
estimation is, by far, the easier part of the benefit-cost analysis.
The first step in assessing the benefits is to estimate the improvement in air quality
F-l-2
-------
resulting from the Clean Air Act. Unfortunately, there are not good inventories of the
ambient air quality or the amounts of each pollutant emitted even today, much less the
ambient air quality and amount emitted in 1969. Since the economy has grown since
1970, we need to estimate ambient air quality and how much emissions would occur at
each date, with and without the Clean Air Act. This task is similar to that for estimating
the costs of abatement.
The second step in estimating benefits is estimating the social benefits that result from
lower air pollution levels. The benefits include human morbidity and mortality,
visibility, odors, aesthetics more generally, damage to ornamental plants and crops, and
other effects from controlling air pollutants. In particular, the following steps are
necessary:
Identify each relevant category of harm - eliminate those that are "trivially" small.
Quantify the relationship between ambient air quality and each effect.
For each year, assess the changes in ambient air quality as a result of the regulations.
For each year, estimate the physical benefits of cleaner air in terms of the categories
identified in 1 and quantified in 2, e.g., premature deaths, cases of each disease,
quality adjusted life-years or disability-adjusted life-years, better visibility, etc.
Value these estimated physical benefits in dollars or compute the value of each relative to
the others and compute a benefits index.
Compute the net benefit (benefits minus costs) if the benefits are valued in dollars or the
cost-effectiveness (dollar costs divided by the benefits index) if only a benefits
index can be computed.
Many non-economists think that 5 is the hardest step, the one that cannot be done
rigorously. That is wrong. Methods have been developed to estimate these values.
Although there are important uncertainties associated with each method, valuation is not
the main source of uncertainty in estimating benefits.
Step 3 adds more uncertainty than step 5, in general. We don't know how the exposure
of the average American has changed as a result of the CAA. We have measurements of
the criteria air pollutants in a few hundred locations, but translating this information into
what people breath when they are outside is difficult. Still more difficult is estimating
the amount they breathe during the day, since they spend only a small amount of time
outside of buildings or motor vehicles. Additional problems are the health interactions of
pollutants, sensitized individuals, and local high pollution concentrations.
Estimating Effects on Humans: Using Epidemiology and Toxicology Information
But the greatest uncertainty is contributed by Step 2. Epidemiology studies the morbidity
and mortality in humans that result from exposure to toxicants. Unfortunately, there are
a host of problems with the usual epidemiology study. The data are almost always
observational rather than experimental. Epidemiologists are tempted to search for
associations that were not hypothesized before the study began. The data sets always
have important uncontrolled factors influencing the results. The data are imprecise or
incomplete because of people lost to follow-up or misdiagnosis. The dose (exposure) is
F-l-3
-------
almost never known with prevision, there are interactions with other environmental
exposures and genetic defects, and the full extent of reactions among people who are
exposed is unknown. Subtle reactions are almost impossible to identify. For example,
some epidemiologists use a relative risk of two as the criteria for having confidence in an
observed relationship.
Connoisseurs of epidemiology bemoan the misleading studies and quantification
difficulties. More than one has told me that they would prefer to rely on some other
approach. That statement reminds me of the story of someone seeking to hire a personal
assistant. With two persons to choose between, he interviewed the first and immediately
hired the second.
Despite these formidable difficulties, epidemiology observes morbidity and mortality in
humans. Toxicology generally studies the effects of a toxicant on a laboratory animal or
a cell culture. Occasionally, human volunteers to are exposed to the toxicant at levels
that are believed too low to harm the individual. For ethical reasons, the studies with
human volunteers keep the dose below the level that would be expected to cause even a
small adverse effect. Thus, there is no information on disease; one must extrapolate from
an observed effect (not adverse) from an acute dose to disease at a higher dose that is
usually present for a long period of time. Despite these formidable difficulties, studies
on humans obviate the need to extrapolate from rodents or from cell cultures. These
studies can provide precise measurement of the effects of the toxicant in these setting, but
say nothing directly about what will be the effect on human morbidity and mortality from
relevant exposures.
Problems such as lack of standardization of protocols are small compared to the central
problem of toxicology: In only a few cases have the results of in vivo or in vitro studies
been compared with human outcomes for the test substance. It is hard to know how to
interpret the results of a toxicology study. A well-done study will have high "internal"
validity, meaning that it can be replicated within this lab and even in other labs.
However, the study has little or no "external" validity in the sense of knowing its
implications for human health.
Regulatory toxicology seeks to protect humans against harmful exposure to toxicants.
Since it is difficult to draw inferences about the harm to humans from experiments with
rodents or cell cultures or to extrapolate from high doses to low doses, without knowing
the physiological mechanisms by which exposure to a toxicant causes disease,
toxicologists have made a set of "conservative" assumptions that are designed to give a
"plausible upper bound" to human toxicity. For example, for cancer the standard
assumption is to construct the 95% upper confidence level for the exposed rodents with
the steepest dose-response relationship. Other assumptions are made about the shape of
the dose-response relationship and the exposure of the population that intended to make
sure that the risk to humans is not underestimated.
F-l-4
-------
Hormesis
Another major issue is the large body of data showing that low level exposures to
toxicants may improve, rather than harm, health. The usual assumption for cancer risk
assessment is that even a single molecule of a carcinogen has a small chance of causing
cancer. A great deal of laboratory data on animals and even some data on humans
suggests that a tiny dose might improve health.
Hormesis becomes a dominant issue for policy in cases, such as benzene, where most
Americans are exposed to benzene at parts per billion concentrations. Most of the
estimated cases of leukemia from benzene exposure occur at a few parts per billion.
Hormesis suggests that concentrations at this level might improve health, rather than
cause leukemia.
Cancer Risk Assessment
Figure 1 illustrates a problem in interpreting toxicology data. These data summarize the
outcome of National Toxicology Program lifetime rodent cancer bioassay results for
about 1,000 chemicals. A first way of looking at the data is the concordance between rats
and mice: 70%. Thus, while there is general agreement between rats and mice, the
agreement is far from perfect. If we flipped a coin in order to predict the rat outcomes,
we would have a concordance of 50%.
Figure 1
Rats
Carcinogen? Yes No
Mice: Carcinogen?
Yes 35 (TP) 15 (FP)
No 15 (FN) 35 (TN)
Another view of the data is the ability of a mouse test to predict the carcinogenicity of
each chemical for rats. Viewed that way, when the bioassay is positive on mice and is
also positive for rats, that is a "true positive" or TP. Unfortunately, for some chemicals
the positive result for mice is negative for rats. This "error" is a "false positive" or FP.
When the bioassay is negative in mice and negative in rats, that is a "true negative" or
TN. Unfortunately, some chemicals are negative in mice but positive in rats, a "false
negative" or FN. The figure shows that, for chemicals that are positive in mice, 70% of
them are positive in rats (TP) and 30% are negative in rats (FP). Of chemicals that are
negative in mice, 70% are negative in rats (TN) and 30% are positive in rats (FN). The
usual interpretation of the NTP is that a chemical that is positive in either rodent species
is considered to be a possible human carcinogen. Thus, 65% of chemicals tested are
considered to be possible human carcinogens.
These NTP bioassay results are used to classify chemicals as likely human carcinogens.
Given the 70% concordance between rats and mice, what is the likely concordance
between rodents and humans? Since rats are more similar to mice than either are to
F-l-5
-------
humans, it seems likely that the concordance between rodents and humans will be less
than 70%. What proportion of the 65% of chemicals that are positive in rodent tests are
human carcinogens? How many false positives and false negatives are there likely to be
in this classification? As noted above, we know that there are some false positives (the
alpha-2-u globulin chemicals) and some false negatives (benzene). The NTP
interpretation is intended to minimize the number of false negatives, even at the cost of
additional false positives.
It seems unlikely that the concordance between rodents and humans is as high as 70%.
In fact, it is more than just possible that the concordance between rodents and humans is
no better than 50%, equivalent to a coin flip, which is a bit cheaper than spending more
than $1 million on a lifetime rodent bioassay.
This is relevant because of the limited data on toxicity in humans. For example, there is
epidemiology data connecting each chemical to a cancer for about two dozen chemicals
or groups of chemicals. Unfortunately, the concentrations that people were exposed to
are often highly uncertain and so there is only a remote idea of the dose-response
relationship. For the other 600 plus toxicants in the Toxic Release Inventory report, there
is at best toxicology data, generally in the form of rodent studies. Biologists warn that
extrapolating between species is perilous. Using toxicological data to estimate the risks
to humans from exposure to a toxicant does precisely that extrapolation. The
extrapolation begins with a leap of faith that the effect observed in rodents or in cultured
cells predicts human toxicity. The differences in anatomy and physiology between
humans and rodents means that many diseases/conditions are unique to rodents or to
humans. Still more uncertain is extrapolating human risk from the dose-response
relationship observed in rodents.
Non-cancer Risk Analysis
Non-cancer risk analysis proceeds by establishing the no observable effects level
(NOEL) in rodents and extrapolating to humans by using safety factors to account for
differences among species and for the most sensitive individuals. The practice assumes
that humans are the most sensitive species, despite considerable data showing that other
species are often more sensitive, e.g., dioxin in mice.
Several proposals have been published to estimate harm for exposure above the reference
dose or to make use of a combination of data and judgment. However, in none of this
work is there an attempt to give an unbiased estimate of the effect on humans. Past data
suggest that for both cancer and other health effects rodents sometimes are positive when
humans are negative and vice versa. Furthermore, when both are positive, sometimes the
rodents are more sensitive than humans and vice versa. These data suggest that if one
knew no more than the results of an in vivo experiment, one would be very uncertain
about the implications for humans, both in terms of whether the chemical is a human
toxicant and, if so, what is the potency.
I don't mean to suggest that benefits estimators are prissy, unused to highly uncertain
data. Indeed, anyone who has read the benefits assessment from the 812 report might be
F-l-6
-------
inclined to ask, as an epidemiologist did of me some years ago: "Are there any data so
bad that an economist would not analyze them?" However, the economist needs
unbiased estimates, even if they are highly uncertain. Toxicologists are not providing
unbiased estimates - that is the fatal flaw.
A Research Agenda
Risk analysis could be most helpful to benefits assessment if toxicologists performed the
following analysis:
For cancer, for all the known human carcinogens, examine whether a standard National
Toxicology Program bioassay would be positive. If there is some uncertainty,
what is the likelihood that a particular bioassay would be positive?
Compare the estimated dose-response relationship for rodents with the relationship for
humans, accounting for the uncertainty in the resulting estimates.
1. For chemicals that are positive in the NTP bioassay, which are not or are unlikely
to be human carcinogens?
2. For non-cancer endpoints, the same questions are relevant, although there is no
single standard for in vivo or in vitro tests. In particular, what is the concordance
between different species in the same test and across different in vivo and in vitro
tests? What is the concordance between test outcomes and human toxicity data
on each chemical?
The other issue is extrapolating from high human doses, observed in accidents,
occupational exposure, or tests with human volunteers, as well as extrapolating from high
doses in in vivo and in vitro tests to the low doses over long periods that most people
experience. In the absence of knowing the mechanism of action, one must rely on
assumptions about the nature of the dose-response relationship. The ED01 experiment
attempted to pin down the best dose-response relationship for cancer. Unfortunately, all
of plausible models performed about as well in explaining the observed data, even though
they had very difference implications for the effects at low exposures.
Summary and Conclusions
As currently practiced, regulatory toxicology cannot provide data to estimate the benefits
of abating hazardous air pollutants. No minor patches will provide these data.
Regulatory toxicology is built on a foundation of protecting humans from harmful
exposure to toxicants. It is saturated with implicit and explicit assumptions that lead to
"plausible upper bounds" rather than "best estimates" of the exposure-response
relationship.
If we knew the mechanisms of action by which toxicants harm humans, we could
establish standards that would protect them and also provide best estimates of the
exposure-response relationships to the benefits assessors. At present, few mechanisms of
action are known and it seems doubtful that we will ever know the mechanisms of action
for most toxicants.
Short of knowing the mechanisms, research can do much to clarify the qualitative and
F-l-7
-------
quantitative risks to humans of exposure. Toxicologists need to analyze the implications
of current in vivo and in vitro tests for human toxicity. They need to look for human data
to compare with laboratory results. I have no doubt that when this happens, we will find
that some popular tests predict human toxicity no better than flipping a coin. For these
tests, society is wasting its resources and using meaningless data to make regulatory
decisions. Other tests can be modified to increase their human predictivity. New tests
can be developed that are more predictive of humans.
Insisting on estimating the benefits of reducing exposure to hazardous air pollutants will
lead to better policy. More important, it should trigger a revolution in toxicology in
searching for laboratory tests that are more predictive of human toxicity.
F-l-S
-------
Appendix F-2
Benzene White Paper
White Paper by Dr. Bernard Goldstein
Environmental and Occupational Health Sciences Institute, Robert Wood Johnson School of
Medicine, Rutgers, Piscataway, NJ.
F-2-1
-------
Benzene White Paper
Bernard D. Goldstein, M.D.
Environmental and Occupational Health Sciences Institute
170 Frelinghuysen Road
Piscataway, NJ 08854
Telephone:732-445-0200
June 16, 2000
F-2-2
-------
Abstract
Benzene is a known cause of aplastic anemia and of human leukemia. At community air pollution
levels on which the benefits analysis for benzene control are to be based, there is no firm evidence
to support a non-neoplastic effect. Estimation of the leukemic effect at such levels requires
extrapolation across about three orders of magnitude of benzene dose. There is currently insufficient
evidence to depart in any direction from low dose linearity.
Introduction
Benzene has been chosen as one of three compounds to be Case Studies for the SAB/EPA
Workshop on the Benefits of Reductions in Hazardous Air Pollutants: Developing Best
Estimates of Dose-Response Functions. The goal of the Workshop is to discuss dose response
assessment methods for hazardous air pollutants (HAP) that are useful for assessing the benefits of
emission control measures. This document is intended to provide a background for this discussion.
The literature on benzene toxicity is perhaps as large as that for any of the compounds designated
as HAPs under the US Clean Air Act. This information has been extensively reviewed elsewhere
(Caprino and Togna, 1998; Goldstein and Witz, 2000; Snyderetal, 1993; Smith and Fanning, 1997;
Goldstein, 1977; Benzene '95 Conference, 1996; Krewski and Snyder, 2000). I will focus on those
studies that may be particularly useful for providing the information needed for economic analysis
of the benefits of reducing benzene exposure specifically related to the control of HAPs under the
1990 CAA Amendments. I have been asked to do this relatively late in the process. There has not
been time to go through a detailed analysis of the basis for the different risk assessments for
benzene, nor do I have the requisite expertise to clearly explicate the major differences in
mathematical modeling approaches. Perhaps this is an advantage.
The organizers of the workshop have commissioned an "Economist's Perspective" by Lester Lave,
a noted economist who has made maj or contributions to the economic analysis of air pollution health
effects. This is not the place to respond to all of the issues raised by his very provocative piece
which unfortunately demonstrates how poorly regulatory biological scientists have communicated
with economists, even ones active in the field of risk assessment such as Dr Lave. However, it does
lead to a recommendation.
To better understand the interface between regulatory risk assessment and economic benefit analysis,
we should rephrase the benzene-related question being asked. The current question is how to
estimate the benefit of partially reducing community benzene exposure now in the range of a few
parts per billion. Instead, I suggest that EPA develop a hypothetical example of an economic
analysis aimed at determining the benefits of reducing a putative workplace standard for benzene
of 30 ppm, TWA to a level of perhaps 15 ppm, TWA. Thirty ppm is a useful baseline because it
is in this range that there are data about actual leukemia risk and on non-cancer endpoints as well.
A target of 15ppm is suggested because it represents only a two-fold reduction to a level that is still
within the range of existing human and animal data. A two-fold reduction in outdoor benzene levels
is also presumably a reasonable outcome of the new MACT strategy for benzene. The goal of this
F-2-3
-------
exercise would be to obtain a better understanding of the biological uncertainties in the rich benzene
data base which impact on a benefit analysis when there is no need for extrapolation to much lower
exposure levels. Once this has been clarified, we can more readily address the extrapolation issues
relevant to the perhaps three orders of magnitude lower levels of benzene exposure needed for HAPs
benefit analysis.
An additional point about the history of HAP regulation relates to this Workshop. Two major
driving forces for the 1990 CAA amended approach to regulate HAPs were impatience at the
previous rate of EPA's regulatory approach and the impact of TRI data showing the many tons of
unregulated pollutants being released into the air . Using a process which required an initial
finding of likelihood of adverse effects, relatively few agents were previously regulated under
Section 112 of the CAA. Benzene was one of them and in fact there is a clear statement of the use
of risk assessment and of economic analysis in the 1984 benzene decision document (EPA, 1984).
This included a table describing the costs and the number of leukemia deaths averted for each of the
control approaches that were considered (Goldstein, 1985). Further, as only certain of these controls
were then imposed, one could relatively easily reconstruct the risk benefit criteria underlying the
decision.
In contrast, the present CAA lists more than 180 compounds to be regulated by EPA, in essence
shifting the regulatory burden from the government which had to make an initial finding of likely
adverse effects before listing, to industry who now must bring sufficient evidence that there are no
adverse effects in order to delist. Congress moved away from risk assessment as the primary basis
for regulation to a technology based approach in which risk considerations only come into play after
Maximum Available Control Technology (MACT) has been instituted. For almost none of these
previously unlisted compounds is there the rich data base available for benzene or for the other
compounds previously chosen for listing under Section 112. Further, for many of the compounds
for which there are ample human data, such as the alkyl benzenes, the data strongly suggest that no
measurable adverse effects are likely at community exposure levels. Yet Congress clearly intended
such compounds to be subject to MACT control irrespective of the lack of convincing data showing
adverse effects. It is inconsistent and perhaps disingenuous of Congress to both disavow risk
assessment as the primary basis for the regulatory control of HAPs while at the same time insisting
that EPA use risk assessment as a means of demonstrating the benefit of the technology-based
regulatory approach which it has imposed.
Health Effects of Benzene
Benzene has been known to produce destruction of the human bone marrow since the 19th Century.
In humans, and in all laboratory animals tested, benzene produces a dose-dependent destruction of
bone marrow precursor cells that are responsible for the production of mature red blood cells,
platelets, and granulocytic and lymphocytic white blood cells. This is accompanied by chromosomal
damage. The result is a decrease in all formed elements in the blood known as pancytopenia. A
severe form of pancytopenia, aplastic anemia, includes marked loss of bone marrow cellularity and
is a frequently fatal disorder.
F-2-4
-------
Benzene is also a known human carcinogen, indisputably causing acute myelogenous leukemia and
its variants (collectively called Acute Nonlymphocytic Leukemia - ANLL). It is also likely to cause
other hematological tumors. An intermediate diagnosis between benzene induced pancytopenia and
ANLL is myelodysplasia, a preleukemic condition characterized by morphological abnormalities
and an increase in number of bone marrow precursors representing a monoclonal expansion. Both
myelodysplasia and ANLL can occur without being preceded by clinically overt pancytopenia.
At concentrations well over 100 ppm (320 mg/m3) benzene also causes central nervous system
anesthetic-like effects common to alkyl benzenes and other VOCs. While an occupational hazard
in enclosed spaces, this non-hematological effect is clearly not pertinent to considerations of HAP
control and will not be discussed further. Based on relatively weak evidence in animal studies, more
information on the potential for benzene-induced developmental effects in humans would be
welcome, but again there seems to be no basis for considering such effects in the present document.
Benzene Exposure Levels Pertinent to Economic Analysis of the Impact of Control of HAPS
In order to provide a discussion of benzene toxicity pertinent to the purposes of this Workshop I
have briefly and incompletely reviewed the literature concerning expected outdoor community
concentrations of benzene. The focus is on the question of whether non-cancer hematological
effects might occur. Accordingly, the goal is to pick a level that would represent a high community
exposure that would be a reasonable target for assessing the health and economic impact of MACT
control measures. For neoplastic endpoints, if one assumes a linear risk then the absolute level is
not important - only the extent of reduction in benzene exposure and the size of the population is
needed to calculate the number of leukemia cases averted. But for non-cancer effects the
presumption of a no-effect level requires that some attempt be made to determine likely high
community exposure levels.
Community outdoor benzene levels even in reasonably polluted areas appear to range well below
10 ppb (32 • g/m3). Wallace and his colleagues have measured outdoor air benzene concentrations
in various parts of the United States as part of the Total Exposure Assessment Measurement
(TEAM) study. In an overview of these studies the mean outdoor air benzene concentration based
on backyard measurements of 175 homes in six urban areas was 6 ug/m3 (Wallace, 1991). The
highest levels in the TEAM study came from one of the two studies in Los Angeles where the
outdoor air concentrations appeared to range up to 30 ug/m3 with a geometric mean of 16 ug/m3
(Wallace, 1986). In a relatively polluted area of New Jersey, the mean levels were 4.1 ug/m3 at night
and 3.8 ug/m3 during the day. Lagrone (1989) reported outdoor benzene levels in a network of six
sites located in an industrial area of Houston ranged from 1.4-5.8 ppb, mean 3.6 ppb, (4.5 -
18.6* g/m3, mean 11.5 • g/m3) during the period September 1987 to March 1988. Johnson et al
(1991) used a variety of models to estimate incremental ambient benzene concentrations to receptors
living near seven bulk gasoline storage facilities in North Carolina. The highest modeled fenceline
annual average benzene incremental concentration was 2.1 ppb (6.1* g/m3). EPA reported that 1991
benzene concentrations in Lima, Ohio, ranged from 1.1 to 6.8 ug/m3, mean 2.6 ug/m3. EPA has also
concluded that the background concentration of benzene, attributable to long range transport and
non-anthropogenic sources, was 0.48 ug/m3 (Woodruff et al, 1998). Data from 97 samples taken as
F-2-5
-------
part of the NHEXAS study in EPA Region 5 shows a median benzene level of 2.9 ug/m3 and a 90th
percentile level of 5.6 ug/m3 (Clayton et al, 1999). There appears to be good evidence that ambient
benzene concentrations are decreasing. Based upon monitoring network data, the California Air
Resources Board (CARB, 1997) has estimated a population-weighted annual concentration of 14.7
ug/m3 benzene in 1982 and 2.3 ug/m3 in 1996.
Benzene Toxicology
To a toxicologist benzene is both fascinating and frustrating. It is a well studied compound that has
provided much insight into general toxicological mechanisms of action and is particularly relevant
to understanding target organ toxicity related to the bone marrow. Its metabolism has also been
thoroughly evaluated and, although complex, is reasonably well understood (Snyder and Hedli,
1996). We also know that it is one or more benzene metabolites, not benzene itself, that is
responsible for its hematological toxicity. Yet the linkage between benzene metabolism and
benzene hematotoxicity remains elusive. It is not at this time even certain that the toxicological
mechanisms by which benzene destroys bone marrow precursor cells leading to aplastic anemia are
the same mechanisms producing cancer of these cells. What we do know suggests that Occam's
Razor is dull, that there is not a single benzene metabolite producing a single mechanism of cell
damage and eventual mutation or cell death (Goldstein, 1990). Ratherthere are multiple metabolites
producing effects in multiple biological pathways that lead through a variety of mechanisms to
adverse effects (Chen and Eastmond, 1995B; Eastmond et al, 1987; Guy et al, 1991; Levay, 1992).
Understanding the relation between benzene metabolism and its mechanism(s) of toxicity is one part
of the puzzle that must be solved if we are to move away from the routine default assumption that
leads to linear extrapolation from high to low dose. A second part of this puzzle is to understand
the relation between the observed biological effects of benzene metabolites and the mechanism(s)
of carcinogenesis. The available information on these two parts of the puzzle does not always point
in the same direction. For example, there is evidence suggesting that the metabolism of benzene to
active intermediates saturates at higher doses which could mean that the dose response is
supralinear. There is also evidence suggestive of aneuploidy being an important mechanism of
benzene carcinogenesis, and it has been argued that such gross chromosome damage requires
multiple "hits" indicating that the dose response to lower benzene levels is sublinear. There are
counter arguments to each of the above.
Risk assessors have attempted to tease the biology out of the epidemiological findings. For example,
Crump (1994) using a weighted exposure approach has calculated the apparent latency period for
benzene leukemogenesis from the pliofilm cohort database and has inferred possible biological
explanations for why his calculations result in a longer latency period for leukemia than was
observed in radiation-exposed atom bomb survivors. Inferring biology from epidemiology can be
useful for hypothesis generation but needs to be approached cautiously as a basis for risk
assessment.
Current advances in molecular biology, including studies using these techniques in benzene-exposed
humans and animals, provide much promise for the eventual unraveling of the mechanisms of
F-2-6
-------
benzene hematotoxicity and leukemogenesis (Chen and Eastmand, 1995 A; Rothman et al, 1995; Xu
et al, 1998; Irons, 2000; Laskin, 2000; Mani et al, 1999; Smith and Rothman, 2000). But at the
present time it is difficult to observe a clear pattern, or even a clear directional signal, that would
permit a generally accepted biological basis for other than a classic linear model of benzene
leukemogenesis.
Risk Assessment for Hematological Neoplasms Caused by Benzene.
The most recent EPA update on benzene (EPA, 1997) derives two risk estimates from the pliofilm
cohort: a lifetime leukemia risk at Ippm (3.2 mg/m3) of 1.8 x 10"2 using an additive risk model, and
4.1 x 10"2 using a relative risk model. Both are based on linear low dose assumptions. These are
little changed from previous EPA risk estimates of 2.6 x 10"2, based primarily on the geometric mean
of four maximum likelihood risk estimates (EPA, 1985) or of an even earlier risk estimate of 2.2 x
10'2 risk at Ippm (3.2 mg/m3) (Goldstein, 1985). The EPA (1997) NCEA review of different
approaches based on the pliofilm data that use linear assumptions states that the risk at Ippm (3.2
mg/m3) ranges from 4.7 x 10"3 to 2.5 x 10"2. Assuming low dose linear extrapolation this translates
into a risk of 47-250 in a million for a 70 year lifetime exposure to 10 ppb benzene (3.2 mg/m3), a
reasonable upper bound for a community exposure level. Of note is that there is a reasonable
similarity between benzene risk assessments derived from the human and animal cancer data
(Goldstein, 1985).
There are three areas of uncertainty that are particularly pertinent to debates concerning the
appropriate risk of benzene-induced cancers: (1) the extent of benzene exposure of workers in
cohorts with an increased risk of ANLL, particularly the pliofilm cohort; (2) the appropriate shape
of the dose-response curve for extrapolating the carcinogenic potential of much lower level benzene
exposures; and (3) whether benzene also causes hematological cancers other than ANLL. The major
uncertainty is the shape of the dose-response curve and particularly whether there is sufficient
evidence to deviate from low-dose linearity.
(1) Extent of benzene exposure in cohorts with an increased risk of ANLL
One of the most thoroughly evaluated cohorts in the history of occupational epidemiology has been
that of pliofilm workers in two Goodyear plants in Ohio. The increase in leukemia incidence among
these workers put an end to any remaining doubt that benzene was a cause of ANLL (Infante et al,
1977). As benzene exposure levels had been reported within the then allowable 10 ppm TWA
workplace standard, OSHA attempted to impose an Emergency Temporary Standard of 1 ppm,
which was thrown out by the Federal court. During the formal rule making that followed, it became
apparent that the pliofilm workers had in fact been exposed to benzene levels well above the
standard. Identifying the actual exposure levels became particularly important to establishing the
new workplace standard, and to establishing the leukemogenic risk of benzene. Rinsky and his
colleagues at NIOSH performed what was then the most extensive retrospective cohort exposure
assessment (Rinsky et al 1981). Not surprisingly, they were forced to make numerous assumptions
as to past exposure levels. Particularly controversial was their assumption of what seemed to be
relatively low levels of exposure during the World War II period. As pliofilm production was an
F-2-7
-------
essential war industry, and a nationwide occupational health review made during this period
indicated a rather cavalier use of benzene in such industries, it seems reasonable that exposure levels
were significantly higher. Building on the work of Rinsky et al (1981), Crump and Allen (1984)
Paustenbach et al (1992, 1993) and Paxton et al (1994A) performed extensive reanalyses of the
exposure levels. Using the reported blood counts in these workers, my colleagues compared the
Crump and Allen with the Rinsky analyses and found that the former more closely predicted the
blood count variations (Kipen et al, 1988; Cody et al, 1993). But we pointed out that our findings
were not relevant to the absolute exposure levels, only to the relative time variations.
The NIOSH researchers have vigorously defended their exposure assessment (Utterback and Rinsky,
1995). A more recent exposure analysis of the pliofilm cohort has been performed by Schnatter et
al (1996) focusing on exposure levels of various worker subgroups in order to refine the risk
estimates. However, EPA (1997) concluded that the various exposure assessments for the pliofilm
cohort do not differ among themselves sufficiently to have a major impact on the 1985 estimate of
risk based upon the epidemiologic data alone. Note that the debate about benzene exposure levels
also has implications to mechanistic issues concerning dose rate and linearity.
Recent studies in China by scientists from the Chinese Academy of Preventive Medicine and the US
National Cancer Institute have provided another data base from which one can attempt to relate an
elevated risk of ANNL to workplace benzene exposure levels (Yin et al, 1987A, B, 1996A, B; Zhang
et al, 1996; Rothman et al, 1995, 96; Hayes et al, 1997). The number of leukemia deaths is
appreciably higher than that for the pliofilm cohort. A reconstructed exposure estimate for much
of the cohort has been reported (Dosemici et al, 1994). While a very useful exercise, the inherent
uncertainties in this dose reconstruction are already leading to controversy (Wong, 1998, 1999).
Preliminary review suggests that the resulting dose response estimates will be in the similar range
of the leukemia dose response estimates for the pliofilm workers. However, it is still possible that
ongoing prospective and retrospective studies of these heavily exposed workers will lead to more
refined estimates of dose response patterns. Of perhaps greater importance to the issue of low dose
linearity is the mechanistic information that may be obtained from study of these benzene-exposed
workers.
Understanding the dose portion of the benzene dose-response relationship requires knowledge of
the specific workplaces. For most industrial settings there tends to be large variations in the extent
of individual exposure which is often not apparent from usual industrial hygiene measurements.
Workers may be in a part of the refinery or chemical factory that is particularly prone to have a
leaking valve. Or there may be individual habits that should be prevented, such as using benzene
to wash off grease and grime from hands or clothes. Smaller and unregulated workplaces, which
seem to characterize the Chinese experience, may well have a larger degree of individual variation.
This variation is important, particularly for establishing leukemia risk, because fortunately only a
small percentage of even a highly exposed workforce develops ANNL. Area benzene
measurements may not reflect the actual exposure of those relatively few individuals, raising
questions as to the validity of risk assessments based on such measurements, particularly in
workplaces with highly variable exposure conditions.
F-2-8
-------
(2) The linearity of the dose response curve for leukemogenesis
Extrapolation of benzene carcinogenesis from the benzene levels observed in the pliofilm or Chinese
studies to the three or so orders of magnitude lower levels pertinent to community exposure levels
has been highly controversial. There have been numerous analyses that have argued that these
data, or others, provide evidence of a non-linear, sublinear or threshold for benzene leukemogenesis
(see for example Paxton, et al, 1994B: Cox, 1996; Schnatter et al, 1996; Wong and Raabe, 1995)
that would lead to a substantial decrease in the estimated risk for the lower level exposures relevant
to community benzene exposures. EPA (1997) has claimed that more than 100 risk estimates have
been presented, and that they vary by 6 orders of magnitude at 1 ppb. The key issue is whether the
extrapolation is assumed to be linear or non-linear. In essence, EPA has defended linear
extrapolation as the preferred approach unless there is adequate biological evidence supporting a
different dose response relationship.
The most extensively analyzed cohort of benzene exposed workers to date has been that of the
pliofilm workers (Infante et al, 1977;Rinsky etal, 1981,1987). A key issue is that with only 9 cases
of leukemia in the Rinsky, 1987 follow-up, versus 2.66 expected, there is very little stability in any
of the analyses. This follow up study may not have been warranted given the relatively short
latency period for ANLL as compared to solid tumors and the higher benzene exposures in the past.
In essence, additional follow up may dilute out the true effect and only add cases that are unrelated
to benzene exposure, thereby obfuscating the dose response issues even further.
Hayes et al (1997) did an extensive analysis of hematologic neoplasms in Chinese workers, reporting
on a cohort of 74,828 benzene-exposed and 35,805 unexposed workers. Their key finding was that
for workers historically exposed to benzene at levels of 10 ppm (32 mg/m3) or less there was an
elevated relative risk for all hematological tumors combined of 2.2 (95% CI 1.1-4.2). For ANLL
andmyelodysplasiatheRRwas3.2(95%CI 1.0-10.1). When the exposure levels were consistently
25 ppm or more, the RR for ANLL and myelodysplasia was 7.1 (95% CI 1.1-15.9). Riskfornon-
Hodgkin's lymphoma, an unusual tumor in China, was also significantly increased. The authors
cautiously note that the dose response curve for benzene-induced cancer from their study tended to
flatten out suggesting a supralinear curve. The cohort was also reported upon with slightly different
numbers (Yin et al, 1996B). This does response estimation has been criticized by Wong (1998,
1999) and by Wong and Raabe (2000).
The Health Council of the Netherlands reviewed benzene risk in 1987 and again in 1997 (Health
Council, 1997). In 1987 the conclusion was that benzene was a human carcinogen and that it
worked through a genotoxic mechanism. It was considered uncertain as to whether the mechanism
of action was stochastic or non-stochastic, i.e. without or with a threshold. A linear extrapolation
was chosen as it was deemed to be the more cautious approach. However, the Health Council
believed that a simple linear extrapolation was not warranted as it would produce "excessively low"
results that would be "overly safe". Accordingly they chose a factor of 100 higher than would
otherwise result from a linear extrapolation to a one in one million lifetime risk. This resulted in a
recommended exposure limit of 12 ug/m3 of benzene in outdoor air.
F-2-9
-------
Their more recent evaluation used a circuitous route to confirm the previously recommended level.
The Health Council's benzene committee again stated itself as being uncertain as to whether
benzene has a stochastic or non-stochastic mechanism of action. However, they were impressed by
a study of 208,000 petrochemical employees said to have an average exposure to 0.7 mg/m3 benzene
which they interpret as showing no increase in ANLL (Wong and Raabe, 1995). They extrapolated
this as being equivalent to 3 5ug/m3 lifetime to the general population. They further concluded that
this supported their earlier view that the exposure-response curve will be sublinear rather than linear.
As there was still uncertainty as to the exact extrapolation technique they left the 12ug/m3
recommendation intact as a one in one million risk.
(3) Does benzene cause hematological neoplasms other than ANLL?
One of the limitations in studying the effects of benzene in laboratory animals has been the difficulty
in developing an animal model of benzene-induced ANLL. However, studies in laboratory animals
have clearly demonstrated that benzene causes hematological neoplasms other than ANLL as well
as non-hematological neoplasms (Maltoni, 1983;NTP, 1984; Snyderetal, 1982). This has naturally
raised the question of whether benzene can cause cancers other than ANLL in humans (Savitz and
Andrews, 1997; Goldstein, 1990; Goldstein and Witz, 2000). I believe that the answer is most likely
yes, but still scientifically unproven, for a variety of hematological tumors including non-Hodgkins
lymphoma (NHL), multiple myeloma and acute lymphatic leukemia (ALL).
In each of these three tumors derived from lymphocytic cells there is some epidemiological support,
although controversial, as well as a strong element of biological plausibility. Lymphocytic cells
are particularly at risk to benzene toxicity, the lymphocyte count decreasing even more rapidly than
does the granulocytic count. Further, chromosomal effects are readily observed in the lymphocytes
of humans and animals with significant benzene exposure. And longer term exposures to benzene
causes lymphomas in laboratory animals.
This is not the place to enter into the details of the controversy concerning the epidemiology.
Briefly, Wong and Raabe and their colleagues have recently published two large meta analyses in
which they report no increased incidence of either multiple myeloma or of NHL (Bergsagel et al,
1999; Wong and Raabe, 2000). Both are seriously flawed by the fact that the populations under
study do not appear to have had a statistically significant increased incidence of ANLL, reflecting
the fact that benzene exposure for most workers in the petroleum industry is relatively well
controlled and that many workers in these cohorts are at little risk of benzene exposure (Goldstein
and Shalat, 2000; Bergsagel et al, 2000). It is unreasonable to ask the question of whether benzene
can cause NHL or multiple myeloma in a cohort in which there is not a clear signal of benzene
induced ANLL. The fallacy is similar to asking whether cigarette smoking can cause NHL in a large
cohort whose level of cigarette smoking is too low to cause a statistically significant increase in risk
of lung cancer.
An additional problem with their approach is exemplified by the findings of Rushton and Alderson
(1981). Their nested case control study of leukemia in British petroleum workers showed a positive
association between workplace benzene exposure and leukemia despite an overall SMR for leukemia
F-2-10
-------
of 0.95. Wong et al (1999) have published a nested case control study of leukemia, acute
myelogenous leukemia, multiple myeloma and kidney cancer in a cohort of petroleum workers
exposed to gasoline. As they did not find an increased risk of acute myelogenous leukemia or of all
leukemias in relation to benzene exposure, their negative findings for kidney cancer and multiple
myeloma are simply not relevant to the issue of whether benzene can cause these latter two cancers.
ALL is primarily a disease of children. While children are exposed to community sources of
benzene, their absence from the workforce precludes usual epidemiological approaches to the
question of whether benzene can cause ALL. About the most that we can reasonably conclude at
present is that it is highly unlikely that the potency of benzene in producing ANNL is exceeded by
its potency in producing any other human cancer.
A maj or issue in all of the extrapolation approaches from the pliofilm cohort is the lack of sensitivity
due to the relatively small numbers involved. These small numbers also impact on the usual
epidemiological approaches to determine if causality exists. For example, the four cases of multiple
myeloma observed in the initial pliofilm study (Rinsky, 1981) were not preferentially observed in
the more highly exposed work categories. But with only four myeloma cases, with one expected,
it is hard to put much credence in the lack of a dose-related distribution. Similarly, Wong and Raabe
(2000) have dismissed the finding of Consonni et al (1999) in which five cases of NHL were
observed (2.12 expected) because of a lack of a statistically significant upward trend.
To summarize the above, in my judgment it is very likely but scientifically unproven that benzene
causes hematological neoplasms other than NHL. A reasonable assumption is that this might lead
to a doubling of the overall cancer risk.
Risk assessment for non-cancer hematological effects
There is no question that benzene causes hematological effects other than cancer. High level
workplace exposures to benzene usually lead to more deaths from aplastic anemia than from ANLL.
Further, in large well studied cohorts in which there have been cases of aplastic anemia, there are
usually many more cases of pancytopenia with all degree of gradation from very mild to highly
significant. Clinical manifestations other than the laboratory findings include symptoms due to
anemia, an increased risk of infection due to a low white blood count, and an increased risk of
hemorrhage due to a low platelet count. In attempting to estimate the health costs, it should be noted
that there is a wide gap between the lower end of the statistically normal range of blood count values
and the much lower blood counts that are required for symptoms or for overt clinically recognizable
disease consequences that could be readily used for economic analysis. An individual with a mild
to moderate benzene-induced pancytopenia will likely be clinically unrecognizable unless he or she
happens to have routine blood counts.
For the purposes of the present exercise I will briefly attempt to distinguish among three levels of
benzene induced non-cancerous effect: 1) the exposure level that will produce a clinically
recognizable endpoint such as symptomatic anemia, infection or hemorrhage; (2) the exposure level
that will produce a lowering of blood count(s) below normal levels; and (3) the level of benzene that
might have any detectable hematological effects.
F-2-11
-------
(1) Symptomatic effects: Such effects undoubtedly require benzene exposures well above the 10
ppm (32 mg/m3) TWA workplace standard in effect for a few decades in the United States and
elsewhere. Literally millions of workers were subjected to routine blood counts on a quarterly to
annual basis. The reason the OSHA standard was decreased to its present 1 ppm level was solely
because of cancer concerns, not because the higher standard was leading to non-cancer
hematological disease. Recent data from China describe clinical aplastic anemia in factories with
exposure levels said to range from 93-1156 mg/m3 (Yin et al, 1987b). A not unreasonable
assumption is that clinically overt symptoms will not occur as a result of long term benzene
exposure to levels at the workplace below 100 mg/m3, or perhaps much higher.
(2) Bloodcount(s) below the normal range: There are a number of studies evaluating blood counts
in workers exposed to reasonably well defined levels of benzene, although in each study there is
some grounds for uncertainty as to whether the measured benzene levels are pertinent to the specific
individuals with low blood counts. In most cases the availability of blood count data is related to
surveillance of benzene-exposed workers. Complicating interpretation of these blood count data is
the fact that there are many reasons for variations in blood counts below the statistically normal
value, some related to normal biological variation, some to laboratory variation, and some to the
many other causes of low blood counts for reasons as diverse as viral infections and alcoholism.
Perhaps the most extensive study of workers exposed to benzene is that of Yin et al (1987b) who
found 2,676 cases of benzene poisoning, defined as a white blood count less than 4,000/mm3, in a
review of over 500,000 benzene-exposed workers in China. The geometric mean concentration in
50,255 workplaces was 18.1 mg/m3, and 64.6% of the workplaces had less than 40 mg/m3. From
their review of the data the authors conclude that cases of benzene poisoning may occur even in
factories with less than 40 mg/m3 benzene.
Gross chromosomal abnormalities in association with overt benzene hematotoxicity were originally
reported by Forni et al (1971; see also Forni, 1996). More recent findings of chromosomal
abnormalities using fluorescent in situ hybridization technique have been reported in Chinese
workers with benzene exposure above 31 ppm (99 mg/m3; Zhang et al, 1996).
(3) Any detectable hematological effects: Sensitive indicators of bone marrow toxicity have been
explored in animal studies aimed primarily at determining the mechanism of benzene
hematotoxicity. Mice are more sensitive to rats. Green et al (1981) looked at specific progenitor
bone marrow cells in mice and reported effects at inhalation exposure levels of 9.9 ppm (32 mg/m3),
but not 1.1 ppm (3.5 mg/m3) benzene, 6 hours per day for 5 days. Farris et al (1997) evaluated
similar endpoints in mice. They found effects at 100 and 200 ppm (320 and 640 mg/m3), 6 hrs a day
for up to 8 weeks, but not at 1,5, and 10 ppm (3.2, 16 and 32 mg/m3). Cytogenetic endpoints also
have been evaluated in laboratory animals.
There are a number of worker studies that have reported statistically significant changes in
hematological endpoints at relatively low benzene exposure levels. Ward et al (1996) reported on
the blood counts of the pliofilm workers and suggested that there may be no threshold for the
hematologic effects of benzene which may occur even at levels less than 5 ppm (16 mg/m3). On the
F-2-12
-------
other hand, Tsai et al (1998), based upon the lack of evidence of effects in hematological monitoring
results from 2475 employees at a petrochemical complex, questioned the need for this form of
surveillance. Khuder et al (1999) reported on a group of 105 petroleum workers exposed to 0.14
-2.08 ppm (0.45 - 6.6 mg/m3) benzene who over time had small but statistically significant falls in
certain blood counts. However, there were problems with this study, including a decrease in the red
cell mean corpuscular volume, a finding contrary to what is observed in benzene toxicity (Goldstein
and Cody, 2000). Nilsson et al (1996) reported findings suggestive of genotoxic effects in men
occupationally exposed to relatively low levels of benzene in the range of 0.1 ppm (0.3 mg/m3), but
there were other exposures as well. Multiple exposures also is a confounding factor in the report
of Carere et al (1995) of cytogenetic changes in Rome gasoline station attendants. There were also
inconsistencies in relation to benzene exposure levels.
In summary, there is no convincing evidence of any non-neoplastic hematological effects at benzene
levels in the range of current community air pollution levels.
Susceptibility issues
There is ample indirect evidence, as well as some direct evidence, of differences in susceptibility
to the hematological effects of benzene among individuals. Women are believed to be more
susceptible than men due to an average higher body fat leading to more benzene storage. There are
genetic polymorphisms governing the activity of many of the known steps in the benzene metabolic
pathway. One of the more intriguing recent findings in the field of genetic polymorphisms is the
observation by Rothman et al (1997) that the risk of decreased blood counts among Chinese workers
exposed to benzene increased seven-fold if the workers had two different phenotypic variations that
led on the one hand to increase the rate at which initial benzene metabolism produced hydroxylated
intermediates and on the other hand slowed the rate of detoxification of these metabolites. Workers
with only one of these variants had approximately a doubling of risk. There is also the suggestion
in the older literature that individuals with thalassemia were at increased risk of benzene toxicity,
an observation that is perhaps generalizable to any groups with increased bone marrow precursors
due to inherited anemias, e.g., sickle cell disease. Much of this work needs to be followed up before
it can be used in economic analyses.
F-2-13
-------
References
Benzene '95. (1995) An International Conference on the Toxicity, Carcinogenesis and Epidemiology of Benzene,
Piscataway, NJ. Environmental Health Perspectives. 104: (Supplement 6).
Bergsagel, D.E., Wong, O., Bergsagel, P.L., Alexanian, R., Anderson, K.C., Kyle, R.A., and Raabe, O.K. (1999)
Benzene and multiple myeloma: Appraisal of the scientific evidence. Blood, 95(4): 1174-1182.
Bergsagel, D.E., Wong, O., Bergsagel, P.L., Alexanian, R., Anderson, K.C., Kyle, R.A., and Raabe, O.K. (2000).
Benzene and multiple myeloma: Scientific evidence. Blood, 95(4): Letter.
Caprino, L., and Togna, G.I. (1998). Potential health effects of gasoline and its constituents: A review of current
literature (1990-1997) onto xicological data. Environmental Health Perspectives. 106(3): 115-125.
Carere, A., Antoccia, A., Crebelli, R., Degrassi, F., Fiore, M, lavarone, L, Isacchi, G., Lagorio, S., Leopardi, P., and
Marcon, F. (1995) Genetic effects of petroleum fuels: cytogenetic monitoring of gasoline station attendants. Mutation
Research. 332(1-2): 17-26.
Chen, H., and Eastmond, D.A. (1995a). Accelerated paper: Topoisomerase inhibition by phenolic metabolites: A
potential mechanism for benzene's clastogenic effects. Carcinogenesis. 16(10): 2301-2307.
Chen, H., and Eastmond, D.A. (1995b). Synergistic increase in chromosomal breakage within the euchromatin induced
by an interaction of the benzene metabolites phenol and hydroquinone in mice. Carcinogenesis. 16(8): 1963-1969.
Clayton, C.A., Pellizzari, E.D., Whitmore, R.W., Perritt, R.L., and Quackenboss, J. J. (1999) National Human Expo sure
Assessment Survey (NHEXAS): distributions and associations of lead, arsenic and volatile organic compounds in EPA
Region 5. Journal of Exposure Analysis and Environmental Epidemiology. 9: 3 81 -3 92.
Cody, R.P., Strawderman, W.W., and Kipen, H.M. (1993) Hematologic effect of benzene: Job-specific trends during
the first year of employment among a cohort of benzene-exposed rubber workers. J. Occup. Med. 35(8): 776-782.
Consonni, D., Pesatori, A.C., Tironi, A., Bernucci, L, Zocchetti, C., andBertazzi, P.A. (1999). Mortality study in an
Italian oil refinery: Extension of the follow-up. American Journal of Industrial Medicine. 35:287-294.
Cox, L.A. Jr. (1996) Reassessing benzene risks using internal doses and Monte-Carlo uncertainty analysis. Environ.
Health Perspect. 104(Suppl 6): 1413-29.
Crump, K.S. (1994). Risk of benzene-induced leukemia: a sensitivity analysis of the pliofilm cohort with additional
follow-up and new exposure estimates. J.Toxicol.Environ.Health. 42(2 ): 219-242.
Crump, K., and Allen, B. (1984). Quantitative estimates of risk of leukemia from occupational exposure to benzene.
OSHA Docket H-059b, Exhibit 152 (Appendix B). Washington, D.C.
Dosemeci, M., Li, G., Hayes, R.B., Yin, S., Linet, M., Chow, W., Wang, Y., Jiang, Z., Dai, T., Zhang, W., Chao, X.,
Ye, P., Kou, Q., Fan, Y., Zhang, X., Lin, X., Meng, J., Zho, J., Wacholder, S., Kneller, R., and Blot, W. (1994). Cohort
study among workers exposed to benzene in China: II. Exposure. American Journal of Industrial Medicine. 26(3): 401-
411.
Eastmond, D.A., Smith, M.T., and Irons, R.D. (1987) An interaction of benzene metabolites reproduces the
myelotoxicity observed with benzene exposure. Toxicol. Appl. Pharmacol^ 91: 85-95.
Environmental Protection Agency. (1997). Carcinogenic effects of benzene: an update. EPA/600/P-97/001A 1-43.
Environmental Protection Agency. 40 CFR Part 61 National emission standards for hazardous air pollutants; benzene
F-2-14
-------
emissions frommaleic anhydride plants, ethylbenzene/styrene plants, and benzene storage vessels; proposed withdrawal
of proposed standards, [AD-FRL-2523-6], 49 FR 8386, March 6, 1984.
Farris, G., Robinson, S.N., Gaido, K.W., Wong, B.A., Wong, V., Hahn, W.P., and Shah, R.S. (1997). Benzene-induced
hematotoxicity and bone marrow compensation in B6C3F1 mice. Fundamental and Applied Toxicology. 36: 119-129.
Forni, A. (1996). Benzene-induced chromosome aberrations: A follow-up study. Environmental Health Perspectives.
104(Suppl6): 1309-1312.
Forni, A.M., Cappellini, A., Pacifico, E., and Vigliani, E.G. (1971) Chromosome changes and their evolution in
subjects with past exposure to benzene. Arch. Environ. Health. 23: 385-391.
Goldstein, B.D. (1977). Hematotoxicity in humans. J. Toxicol. Environm. Health (suppl.) 2:69-105.
Goldstein, B.D. (1985) Risk assessment and risk management of benzene by the Environmental Protection Agency.
Banbury Report 19: Risk Quantitation and Regulatory Policy.
Goldstein, B.D. (1990) Is exposure to benzene a cause of human multiple myeloma? Annals of the New YorkAcademy
ofScience,609:225-334.
Goldstein, B.D.and Cody, R. (Letter) Assessment of complete blood count variations among workers exposed to low
levels of benzene. J Occup Environ Med. 42:2-113-114, February, 2000.
Goldstein, B.D.and Shalat, S.L. (2000) The causal relation between benzene exposure and multiple myeloma. Blood,
95(4): 1512-4. Letter
Goldstein, B.D. and Witz, G. Chapter 4: Benzene. In: Environmental Toxicants: Human Exposures and Their Health
Effects. Second Edition. Ed. M. Lippman, John Wiley and Sons, Inc., pp.121-150, 2000.
Green, J.P., Snyder, C.A., Lo Bue, J., Goldstein, B.D. and Albert, R.E. (1981). Acute and chronic
dose-response~Effects of inhaled benzene on multi potential hematopoietic stem (CFU/S) and granulocyte/macrophage
progenitor (GM/CFU/C) cells in CD/1 mice. Tox. Appl. Pharm^ 58:492-503.
Guy, R.L., Hu, P., Witz, G., Goldstein, B.D. and Snyder, R. (1991) Depression of iron uptake into erythrocytes in mice
by treatment with the combined benzene metabolites p-benzoquinone, muconaldehyde and hydroquinone. J. Appl.
Toxicol 11: 443-446.
Hayes, R.B., Yin, S.-N., Dosemeci, M., Li, G.-L., Wacholder, S., Travis, L.B., Li, C.-Y., Rothman, N., Hoover, R.N.,
and Linet, M. (1997). Benzene and the dose-related incidence of hematologic neoplasms in China. Journal of the
National Cancer Institute. 89: 1065-1071.
Health Council of the Netherlands: Committee on the Evaluation of the Carcinogenicity of Chemical Substances. (1997).
Benzene. 1-44.
Infante, P.P., Rinsky, R. A., Wagoner, J.K., Young, R.J. (1977). Leukemia in benzene workers. Lancet. 1:76-78.
Irons, R.D. (2000) Molecular models of benzene leukemogenesis. In: Special issue onbenzene. Envir. Health Perspect.
(Supplement to be published). Krewski, D., and Snyder, R. (editors).
Johnson, T.R., Roy, A.P., andCapel, J.E. (1991) Estimation of incremental benzene exposure associated with seven bulk
gasoline storage facilities in North Carolina. Health and Environmental Sciences Department, American Petroleum
Institute. Report, 1-36.
F-2-15
-------
Kipen, H.M., Cody, R.P., Crump, K.S., Allen, B. and Goldstein, B.D. (1988) Hematologic effects of benzene: A
thirty-five year longitudinal study of rubber workers. Toxicology & Industrial Health^ 4:411-430.
Khuder, S.A., Youngdale, M.C., Bisesi, M.S, Schaub, E.A. (1999) Assessment of complete blood count variations
among workers exposed to low levels of benzene. J. Occup. Environ. Med 41:821-826.
Krewski, D., and Snyder, R. (editors). (2000) Special issue on benzene. Envir. Health Perspect. (Supplement).
Lagrone,F.S. (1989). Assessing community exposure to selected volatile indicator compounds. Journal of the American
College of Toxicology. 8(5): 897-904.
Laskin, D.L. (2000) Nitric oxide as a mediator of benzene-induced hematosuppression and toxicity. In: Special issue
on benzene. Envir. Health Perspect. (Supplement to be published). Krewski, D., and Snyder, R. (editors).
Levay, G. (1992) Potentiation of DNA adduct formation in HL-60 cells by combinations of benzene metabolites.
Proceedings of the National Academy of Science of the United States of America. 89: 7105-7109.
Maltoni, C., Conti, B. , and Cotti, G. (1983). Benzene: A multipotential carcinogen. Results of long-term bioassays
performed at the Bologna Institute of Oncology. American Journal of Industrial Medicine, 4: 589-630.
Mani, C., Freeman, S., Nelson, D.O., Vogel, J.S., and Turteltaub, K.W. (1999) Species and strain comparisons in the
macromolecular binding of extremely low doses of [14C] benzene in rodents, using accelerator mass spectrometry.
Toxicol. Appl. Pharmacol. 159(2):83-90.
Nilsson, R.I., Nordlinder, R.G., Tagesson, C., Walles, S., and Jarvholm, E.G. (1996). Genotoxic effects in workers
exposed to low levels of benzene from gasoline. American Journal of Industrial Medicine, 30: 317-324.
National Toxicology Program. (1984). NTP Technical Report on the Toxicology and Carcinogenesis Studies of
Benzene (CAS #71-43-2) NF344/NRats andB6C3Fb Mice (Gavage Studies) NTP TF 289. U.S. Department of Health,
Education and Welfare.
Paustenbach, D. J., Bass, R.D., and Price, P. (1993). Environmental Health Perspectives Supplements, 101(Suppl. 6):
177-200.
Paustenbach, D.J., Price, P.S., Ollison, W., Blank, C., Jernigan, J.D., Bass, R.D., and Peterson, H.D. (1992) Re-
evaluation of benzene exposure for the plioform (rubberworker) cohort (1936-1976). Journal of Toxicology and
Environmental Health, 36: 177-231.
Paxton, M.B., Chinchilli, V.M., Brett, S.M., and Rodricks, J.V. (1994a). Leukemia risk associated with benzene
exposure in the pliofilm cohort: I. Mortality update and exposure distribution. Risk Analysis, 14(2): 147-154.
Paxton, M.B., Chinchilli, V.M., Brett, S.M., and Rodricks, J.V. (1994b). Leukemia risk associated with benzene
exposure in the pliofilm cohort: II. Risk estimates. Risk Analysis, 14(2): 155-161.
Rinsky, R., Smith, A.B., Hornung, R., Filloon, T.G., Young, R.J., Okun, A.H., andLandrigan, PJ. (1987). Benzene and
leukemia: An epidemiologic risk assessment. New England Journal of Medicine. 316: 1044-1050.
Rinsky, R.A., Young, R.J., Smith, A.B. (1981) Leukemia in benzene workers. Am J. Industrl. Med. 2:217-245.
Rothman, N., Hayes, R.B., Haas, R. etal. (1995). Benzene induces gene-duplicating but not gene-inactivating mutations
at the Glycophorin A locus in exposed humans. Proc. Natl. Acad. Sci. USA, 92:4069-4073.
Rothman, N., Li, G., Dosemeci, M., Bechtold, W.E., Marti, G.E., Wang, Y., Linet, M., Xi, L., Lu, W., Smith, M.T.,
F-2-16
-------
Titenko-Holland, N., Zhang, L., Blot, W., Yin, S., and Hayes, R.B. (1996). Hematotoxicity among Chinese workers
heavily exposed to benzene. American Journal of Industrial Medicine, 29:236-246.
Rothman, N., Smith, M.T., Hayes, R.B., Traver, R.D., Hoener, B., Camleman, S., Dosemeci, M, Linet, M, Zhang, L.,
Xi, L., Wachholder, S., Lu, W., Meye, K.B., Titenko-Hollanad, N., Stewart, J.T., Yin, S., and Ross, D. (1997) Benzene
Poisoning, A Risk Factor for Hematological Malignancy, is Associated with the NQ01 609C->T Mutation and Rapid
Fractional Excretion of Chloroxazone. Cancer Res^ 15:2839-2842.
Savitz, D., and Andrews, K.W. (1997). Review of epidemiologic evidence onbenzene and lymphatic andhematopoietic
cancers. American Journal of Industrial Medicine, 31(3): 287-295.
Schnatter, A.R., Nicolich, M.J., and Bird, M.G. (1996). Determination of leukemogenic benzene exposure
concentrations: Refined analyses for the pliofilm cohort. Risk Analysis, 16(6): 833-840.
Smith, M.T., and Fanning, E.W. (1997). Meeting Report: Report on the Workshop Entitled: "Modeling Chemically
Induced Leukemia - Implications For Benzene Risk Assessment. Leukemia Research, 21(5): 361-374.
Smith, M.T., and Rothman, N. (2000) Biomarkers in the molecular epidemiology of benzene-exposed workers. In:
Special issue onbenzene. Envir. Health Perspect. (Supplement to be published). Krewski, D., and Snyder, R. (editors).
Snyder, C.A., Goldstein, B.D., Sellakumar, A., Bromberg, L, Laskin, S. and Albert, R.E. (1982) Toxicity of chronic
benzene inhalation: CD-I mice exposed to 300 ppm. Bull. Environm. Contain. Toxicol^ 29:385-391.
Snyder, R., and Hedli, C.C. (1996) An overview of benzene metabolism. Environ. Health Perspect. Suppl., 104:1165-
1171.
Snyder, R., Witz, G. and Goldstein, B.D. (1993) The toxicology of benzene. Environmental Health Perspectives^
100:293-306.
Tsai.S.P., Bennett, J.M., Salesman, C.N., Ryan, T.E., Gilstrap, E.L., and Ross, C.E. (1998). Medical surveillance for
hematological disorders among active and retired refinery workers. Journal of Environmental Medicine. 40(5): 475-480.
Utterback, D.F. and Rinsky, R.A. (1995). Benzene exposure assessment in rubber hydrochloride workers: A critical
evaluation of previous estimates. American Journal of Industrial Medicine, 27:661-676.
Wallace, L.A. (1986). The California Team Study: Breath concentrations and personal exposures to 26 volatile
compounds in air and drinking water of 188 residents of Los Angeles, Antioch, and Pittsburg, CA. Atmospheric
Environment, 22(10): 2141-2163.
Wallace, L.A. (1991). Comparison of risks from outdoor and indoor exposure to toxic chemicals. Environmental Health
Perspectives, 95: 7-13.
Ward, E., Hornung, R., Morris, I, Rinsky, R., Wild, D., Halperin, W., and Guthrie, W. (1996). Risk of low red or white
blood cell count related to estimated benzene exposure in a rubberworker cohort (1940 -1975). American Journal of
Industrial Medicine, 29: 247-257.
Wong, O. (1998). Re: Benzene and the dose-related incidence of hematologic neoplasms in China. J.Nat. Cancer Inst^
90: 469-471.
Wong, O. (1999). A critique of the exposure assessment in the epidemiologic study of benzene-exposed workers in
China conductedby the Chinese Academy of Preventive Medicine and the US National Cancer Institute. Regul. Toxicol.
Pharmacol. 30: 259-267.
F-2-17
-------
Wong, O., and Raabe, O.K. (1989). Critical review of cancer epidemiology in petroleum industry employees, with a
quantitative meta-analysis by cancer site. American Journal of Industrial Medicine. 15: 283-310.
Wong, O., and Raabe, O.K. (1995) Cell-type-specific leukemia analyses in a combined cohort of more than 208,000
petroleum workers in the United States and the United Kingdom, 1937-1989. Regul. Toxicol.Pharmacol. 21(2): 380-4.
Wong, O., and Raabe, O.K. (2000). Non-Hodgkin's lymphoma and exposure to benzene in a multinational cohort of
more than 308,000 petroleum workers, 1937 to 1996. JOEM. 42(5):554-567.
Wong, O., Trent, L., and Harris, F. (1999) Nested case-control study of leukemia, multiple myeloma, and kidney cancer
in a cohort of petroleum workers exposed to gasoline. Occup. Environ. Med. 56:217-221.
Woodruff, T.J., Axelrad, D.A., Caldwell, I, Morello-Frosch, R., and Rosenbaum, A. (1998) public health implications
of 1990 air toxics concentrations across the United States. Environmental Health Perspectives. 106(5):245-251.
Xu, X., Wienchke, J.K. , Niu, T., Wang, M, Watanabe, H., Kelsey, K.T., and Christiani, D.C. (1998). Benzene
exposure, Glutathione S-Transferase Theta homozygous deletion, and sister chromatid exchanges. American Journal
of Industrial Medicine. 33: 157-163.
Yin, S.-N., Li, G.-L., Tain, F.-D., Fu, Z.-L, Jin, C, Chen, Y.-J., Luo, S.-J., Ye, P.-Z., Zhang, J.-Z., Wang, G.-C, Zhang,
X.-C., Wu, H.-N., and Zhong, Q.-C. (1987a). Leukemia in benzene workers: a retrospective cohort study. British
Journal of Industrial Medicine. 44: 124-128.
Yin, S.-N., Li, Q., Liu, Y., Tain, F., Du, C., and Jin, C. (1987b). Occupational exposure to benzene in China. British
Journal of Industrial Medicine. 44:192-195.
Yin, S.-N., Hayes, R.B., Linet, M.S., Li, G.-L., Dosemeci, M., Travis, L.B., Zhang, Z.-N., Li, D.-G., Chow, W.-H.,
Wacholder, S., Blot, W.J., and The Benzene Study Group. (1996a). An expanded cohort study of cancer among
benzene-exposed workers in China. Environmental Health Perspectives. 104(Supplement6): 1339-1341.
Yin, S., Hayes,R.B.,Linet,M.S.,Li, G.,Dosemeci,M., Travis,L.B., Li, C.,Zhang,Z.,Li,D.-G., Chow, W., Wacholder,
S., Wang, Y., Jiang, Z., Dai, T., Zhang, X., Lin, X., Meng, J., Ding, C., Zho, J., and Blot, W. (1996b). A cohort study
of cancer among benzene-exposed workers in China: overall results. American Journal of Industrial Medicine. 29:227-
235.
Zhang, L., Rothman, N., Wang, Y., Hayes, R.B., Bechtold, W., Venkatesh, P., Yin, S., Dosemeci, M., Li, G., Lu, W.,
and Smith, M.T. (1996). Interphase cytogenetics of workers exposed to benzene. Environmental Health Perspectives.
104(Supplement6): 1325-1329.
F-2-18
-------
APPENDIX F-3
Challenges in Projecting Human Health Impacts from Exposures to Perchloroethlyene
White Paper by Dr. Lorenz Rhomberg,
Gradient Corporation, Cambridge, MA
F-3-1
-------
Challenges in Projecting Human Health Impacts from
Exposures to Perchloroethylene
Lorenz R. Rhomberg, Ph.D.
Principal Scientist
Gradient Corporation
Cambridge, MA
1 Introduction
When analysis of the toxic effects of chemicals is applied to the task of assessing benefits of regulations
that limit exposure, the goal is to estimate the impact of changes in exposure regimes on changes in the
burden of ill health in the exposed population. This differs from the aim of traditional regulatory risk
assessment, which is to define exposure levels that can be deemed "safe," or at least that can be found to
pose no more than "acceptable" risks. That is, the usual methods seek to define dose levels without
pronounced impacts, not to estimate or to characterize the impacts that may occur.
Not surprisingly, traditional methods are ill suited to the estimation and description of toxic effects to be
expected when chemical exposures approach and exceed levels that can be assuredly ruled safe. An
often-mentioned issue is that traditional risk assessment methods are "conservative" in that they deal with
uncertainties in the inferential process by making estimates or assumptions unlikely to underestimate risk,
thereby tending to overestimate risk, at least on average. Such biases distort the assessment of benefits
gained from avoided exposure.
F-3-2 Gradient Corporation
-------
Two further issues are perhaps as important, however, and may be more difficult to remedy. First, owing
to their focus on defining doses without unacceptable effects, existing approaches often say little about
what specific toxic phenomena are to be expected at exposures exceeding "safe" levels. Second, because
they focus on individual risks to benchmark individuals with defined "high-end" exposures, existing
approaches are not geared to estimating population risks among a large group of subjects with varying
levels of exposure that may fluctuate in time or consist of occasional high-exposure episodes.
To undertake benefits assessment, some new approaches to analysis of toxic effects will have to be
considered. This paper attempts to examine the challenges and to consider modifications to risk analysis
methods that may help to address some of the questions. To ground the discussion in the context of data
that are available for actual toxic agents, the example of perchloroethylene is used.
It is best to begin by defining goals, even if they represent ideals that we are unlikely to achieve in
practice. In order for economic analysis to measure the benefits of regulations that restrict exposure to a
potentially toxic agent, it must have estimates of the burden of ill health in the exposed population as it
would be expected to exist when the regulation is applied as well as when it is not applied. The
differences between these constitute the avoided health impacts, and the values placed on them (which,
thankfully, it is not my task to address) largely constitute the benefits of the regulation. Clearly, at least
one of these scenarios (with the regulation or without) is hypothetical, and so even in the ideal case we
cannot rely solely on observation. Modeled projection of health impacts to be expected in a population
under hypothetical exposure scenarios is a necessary part of the analysis.
1.1 Needs
It would seem that the ideal toxicological analysis would provide characterization of the following:
1. What specific responses are engendered by exposures to the toxic agent?
2. For responses that are graded, how severe is the response? How does severity progress over
time?
3. When (in the course of an ongoing exposure) do responses arise? How long do ill effects endure?
4. In whom do responses arise?
These bear some discussion. First, to put value on a case of toxicity avoided, it helps to specify the effect
in question, since different effects (and different severities) have different impacts on the quality of life.
Exiting methods typically eschew making statements about the specific nature of toxic effects in humans
that are extrapolated from animal studies. Animal carcinogenicity is assumed to indicate a human risk for
some type of cancer, but this is not necessarily expected to manifest itself as the same type of cancer seen
in the animals. Noncancer toxicity assessments define doses that appear to avoid all adverse responses
seen among experimental animals, and the most sensitive of these is deemed the "critical effect," but it is
not specified which effects are to be expected in humans in exposures that exceed "safe" levels. Ideally,
then, methods for benefits assessment should aim at making more specific projections about the nature of
the toxicity to be expected in sufficiently exposed humans. They should recognize that several toxic
effects may be at issue, not solely the one that was used to set the acceptable dose in the regulation being
examined.
Similarly, existing methods do not project when during the course of an ongoing exposure the adverse
effects will become manifest. To place a value on a case of toxicity, however, one would want to know
when in life it appears, how long the state of ill health endures, whether it changes in severity, whether
the disease fully or partly regresses upon cessation of exposure, how much the length of life is shortened,
F-3-3 Gradient Corporation
-------
and how a change in exposure at some midlife point (resulting, say, from the imposition of a regulation)
changes the likelihood of response. In short, one would like not just dose-response relationships, but
descriptions of response as a function of dose-rate and time, including description of the consequences of
non-constant dose rates. Exposure-dependence of the course of disease in any cases engendered is also of
interest.
It should be clear that the concern is for population risks, not just individual risks (which are often the
focus of traditional assessments). We seek to characterize all the effects as they are (or would be) realized
in an actual human population of interest. The hypothetical 70-year fence-line resident, the pica child, the
worker laboring 45 years at a degreasing tank, and other standardized individual scenarios of exposure
that define benchmarks of individual risk in regulation-setting assessments are not at issue, not just
because they are "high-end" exposures but because they represent but a few individuals among the many
in the population whose collective benefits we wish to address.
Ideally, we would want to describe not only the full frequency distribution of exposure levels, but also
when and by whom the various exposures are experienced, since exposures at different ages or in
different patterns over lifetime will differently affect the likelihood of responses (and we may wish to
place different value on responses occurring at different times or in people with different prior states of
health). Multiple sources and pathways of exposure exist, and people change their geographic locations
and local exposures on timescales ranging from minutes to years. When a regulation is phased in, or
when an agent persists in the environment even after controls are imposed, the exposures will change year
by year, and this time pattern may be important to characterize to gauge accrual of benefits from the
exposure restriction.
These facts make for major challenges to exposure assessment. (Exposure methods are not my focus, but
the issues should not be overlooked.) Creating a complete inventory of the individual histories of
exposure in an entire diverse population may seem a daunting task, but considerable progress has been
made in approaching such a description using simulation modeling. In this approach, the events and
settings that lead to exposure in a population are described as random variables, and a large set of
simulated life histories can be assembled (ILSI 1998) that describes the diversity of experiences in a
whole population.
1.2 Uncertainty
Existing methods in risk assessment for projecting human risks from experimental observations of
toxicity in animals are highly uncertain. Even use of epidemiological studies entails uncertainty in
characterization of exposures, in description of responses, and in generalization from the study population
to the more general population of interest. New methods that attempt to make more detailed
pronouncements regarding the nature of endpoints and the timecourse of their manifestation while
acknowledging the complexity of the distribution of human exposures are bound to be still more
uncertain.
Any demand that an analysis of benefits cannot be undertaken until impacts of exposure can be projected
with confidence dooms the enterprise. It also misses the point. While we do our best to project outcomes
with precision, uncertainty cannot be avoided, only characterized. It is the fact that outcomes are
uncertain that makes them risks. The assessment of the costs and benefits of regulation can be regarded
as a problem in decision under uncertainty—we have to decide how much to spend to control exposures
in the face of uncertainty about how much benefit (in terms of reduced health impact) we will receive.
The decision to incur regulatory costs is deemed a good one for society if the mathematical expectation of
F-3-4 Gradient Corporation
-------
the uncertain benefits exceeds that of costs. The expectation is not the single most likely value, but rather
the average over possibilities, each weighted by its likelihood of being true.
Seen from this point of view, the characterization of uncertainty in the projection of health effects is
central to the analysis. The characterization of risk consists of specifying an array of possible outcomes
or courses of events, each element of which is associated with the likelihood of its occurrence and the
consequences should it indeed occur (Kaplan and Garrick 1981). In the present case, the likelihoods
constitute our relative confidence in the alternative projections of health impacts. We want to avoid the
"upper bound" and "worst-case" nature of much of existing methodology, but at the same time we should
not seek only single "best estimates" (such as maximum likelihood curve fits). Instead, we should seek to
characterize the distribution of possibilities.
2 Perchloroethylene
Perchloroethylene ("Perc," CAS No.127-18-4) is a high production-volume chlorinated solvent used as a
chemical intermediate, as a solvent and degreasing agent, and as the primary solvent in drycleaning
operations. Perchloroethylene is moderately volatile; without containment and measures for vapor
recovery, use and disposal can result in considerable release of vapor to the atmosphere. Spills and leaks
during storage have resulted in cases of contaminated soil and groundwater. Because the resulting
exposures to workers and the general public lead to concerns for potential human health effects, the use
and disposal of perchloroethylene is subject to regulation aimed at limiting workplace concentrations and
releases of the chemical to the environment. The mandated controls can be costly, and it is of interest to
establish how much impact on the health of the human population is avoided through their application.
A full review of the exposures to perchloroethylene and a complete characterization of its toxicologic and
epidemiologic database are beyond the scope of this paper. The following overview, drawn from IARC
(1995), EPA (1985, 1991), ATSDR (1997), OEHHA (1999) and other sources, gives a perspective on the
available information that is sufficient for the present discussion of methodological issues.
Exposure: Worldwide annual production of perchloroethylene (which has declined somewhat in recent
years) is in the hundreds of thousands of tons. Some 55% is used for drycleaning, 23% as a chemical
intermediate (mostly for CFC production, which is declining), and 13% for liquid and vapor degreasing,
with other uses including fabric treatment and paint stripping. Sampled air concentrations vary
considerably in degreasing facilities, but means are often on the order of 10-100 ppm (parts per million)
with some individual air samples in the 1,000 ppm range. Occupational exposures in drycleaning
facilities are on the order of 10-50 ppm (IARC 1995).
Ambient air levels are much less and are reported here in parts per billion (1000 ppb = 1 ppm); they vary
somewhat with season and are generally higher in urban than in rural air. Levels of 0.2 to 2 ppb are
usually found in outdoor urban air. Indoor levels are often higher, sometimes tenfold outdoor levels.
Peak levels in apartments above drycleaning establishments have been measured at 1000 ppb and higher.
Off-gassing from drycleaned clothes can lead to temporarily high levels in automobiles (about 2000 ppb
but with reports up to 300,000) and in homes (about 400 ppb).
Pharmacokinetics and Metabolism: Perchloroethylene is readily absorbed after inhalation or ingestion.
Much of this is exhaled unchanged, but on the order of 30-50% is metabolized at low exposure levels in
both rodents and humans. Most of this metabolism is via an oxidative pathway, but at exposure levels
higher than about 100 ppm (such as in rodent lifetime bioassays), the oxidative pathway becomes
increasingly saturated. This leads to proportionally higher metabolism by a glutathione-conjugation
F-3-5 Gradient Corporation
-------
pathway (which is still small in absolute terms) and higher exhalation of unmetabolized compound.
Glutathione conjugates can be further metabolized in kidney to reactive, apparently genotoxic
compounds, but oxidative metabolites (and perc itself) do not appear to be genotoxic. Several
pharmacokinetic models of perchloroethylene metabolism exist; they agree in broad outline but differ in
detail, especially regarding rodent-human differences in the extent of the conjugative pathway.
2.1 Observations of Toxicity in Humans
Neurological effects have been seen in populations occupationally exposed to bouts of high
perchloroethylene concentration in air, and the nervous system seems to constitute the most susceptible
target in humans. Overt symptoms such as headache, nausea, and ataxia are not seen in experiments at
doses below 100 ppm, and these are fully reversible. Subtler pre-clinical neurophysiological and
neurobehavioral effects such as changes in electroencephalograms, visual-evoked potentials, color vision
discrimination, and tests of coordination or reaction time show detectable influence of exposure at levels
between 15 and 100 ppm, although these, too, are reversible upon cessation of exposure. No clear
evidence suggests permanent neurological effects from chronic occupational exposure, but some studies
report detection of significant differences in memory and reaction time.
Case studies exist of workers exposed to very high levels in industrial accidents (e.g., a worker found
unconscious in a pool of solvent) in which serious liver or kidney damage occurred, but in such cases
there is apparent full recovery within weeks. As with neurological effects, subtler pre-clinical changes
that are considered markers of potential toxicity are seen in some studies of workers with exposures in the
20-30 ppm range. Many of these are elevations in serum concentrations of certain liver-cell enzymes
(SGOT, SGPT, GOT) that are taken to signal some loss of integrity or increased permeability of liver
cells, and hence possible beginnings of hepatotoxicity. It is typical for these quantitative measures to be
within normal range in all subjects yet the means for exposed and unexposed groups are statistically
different.
Some studies suggest slightly increased rates of spontaneous abortion or menstrual complaints in women
with occupational exposures, and some studies suggest longer times to conception in couples with one or
the other parent exposed. No associations with stillbirth, low birthweight, or malformations have been
seen.
Several occupational epidemiological studies of carcinogenic effects have been conducted of dry cleaning
workers and those exposed in settings where degreasing activities lead to elevated air concentrations.
Various inconsistent small elevations of one or another type of tumor have been reported (lymphopoietic,
female genital, bladder, kidney, breast), but the only one showing any consistency is esophageal cancer.
This effect (SMR 2.1 and 2.6) was seen in two drycleaning employee cohorts (but only in black men in
one of them). A case-control study of esophageal cancer showed a non-significant association with
employment in drycleaning. Esophageal cancer is subject to influence by smoking and alcohol use.
Moreover, perchloroethylene is not the only chemical exposure for many of the workers in these studies.
2.2 Observations of Toxicity in Experimental Animals
Many of the noncancer effects seen in humans are seen in animals as well, but often at higher doses and
for more overt and frankly toxic versions of the effect (since subtle effects are difficult to detect). Thus,
animals acutely exposed to over 1000 ppm showed ataxia and anesthesia as well as altered psychomotor
functions. Effects on brain weight were seen in rats at 600 ppm for 4 or 12 weeks. High exposures also
produce liver and kidney toxicity, and the biochemical markers such as serum enzymes also appear at
F-3-6 Gradient Corporation
-------
exposures on the order of 25 ppm. Some effect on litter size and survival during lactation were seen at
lOOOppm.
In lifetime carcinogenicity bioassays, perchloroethylene by gavage (NCI 1977) and by inhalation (NTP
1986) increased hepatocellular carcinomas in male and female mice. The NCI study has been questioned
because the perchloroethylene used was stabilized with epichlorohydrin, itself an animal carcinogen.
Inhalation in rats led to an increase in mononuclear cell leukemia in both sexes, although response was no
higher at the high than at the low dose. In addition, treated male rats had a few renal tubular cell
adenocarcinomas that, although not statistically elevated compared to controls, were considered
lexicologically significant owing to their historical rarity.
None of these animal tumor responses is without some controversy regarding its applicability as indicator
of potential human risk. Mice of the strain tested are particularly prone to such tumors, and they appear
at high levels even in controls. A major metabolite of perchloroethylene, trichloroacetic acid, induces
proliferation of peroxisomes in mouse liver cells at high doses, and the damage or oxidative stress these
cause may be involved in the induction of tumors, although other evidence questions the role of
peroxisomes in hepatocarcinogenesis and the correlation of their induction with liver tumor induction has
counterexamples. Humans have very little peroxisomes induction, even at high exposures, and the
background rate of liver cancer is much lower than seen in mice. Meanwhile, trichloroacetic acid
administered to mice in drinking water or experienced as a metabolite of trichloroethylene (which is
similar in toxicology and metabolism to perchloroethylene) causes similar liver tumors at doses below
those inducing peroxisomes and without inducing evident cell proliferation.
Similarly, the rat strain studied is prone to mononuclear cell leukemias, a tumor type with no clear
analogue in humans (it is splenic, and human leukemias originate in marrow). The rat controls have high
responses, although the rate is observed to vary among studies. Male rats can develop kidney tumors
from some chemicals that inhibit degradation of a male rat-specific protein (a2u-microglobulin) that
accumulates in renal tubule cells, causing toxicity. This syndrome is unique to male rats and is
considered irrelevant to human risk (since humans lack the mechanism altogether). Perchloroethylene
metabolites appear to cause this phenomenon in male rats, but only at doses higher than those in the NTP
bioassay, suggesting that a different mechanism is responsible. On the other hand, bioassay-level
exposures to perchloroethylene do induce kidney toxicity, probably as a result of the kidney's ability to
further metabolize products of the conjugative pathway into reactive compounds (which also may be
genotoxic). But this phenomenon, including the kidney toxicity, is seen in mice as well, and mice do not
have elevations in kidney tumor risk. There is evidence that the conjugative pathway and the activation
of metabolites in kidney happen in humans, but the quantitative extent is unclear.
3 Projecting Cancer Risks
If we want to assess the benefits of limiting perchloroethylene exposure in terms of avoiding cancer risks,
the first question to face is whether perc is a human carcinogen at all. One possible stance is to conclude
that evidence is insufficient to treat this compound as a human carcinogen, and hence there is no cancer
risk among exposed people (and thus no benefit from restricted exposure). Even if we feel that this is the
single best-supported conclusion, however, there is some probability that we are wrong, and if we are, the
cancer risk that may exist is overlooked. By the same token, it would be a mistake to put all our credence
in an analysis that assumes that perc is a human carcinogen, ignoring the substantial probability that any
risks so calculated are illusory.
F-3-7 Gradient Corporation
-------
Current methods force just such an either-or decision, with the decision process couched in the weight-of-
evidence determination in hazard identification. In the case of perchloroethylene, the weight-of-evidence
regarding human carcinogenicity is particularly muddled. IARC (1995) has called perc a 2B "probable
human carcinogen" based on what it judges to be "limited" epidemiological evidence and "sufficient"
animal evidence. EPA has withdrawn a former B2 classification, and the SAB has declared perc to be on
the borderline between B2 and C. For the purposes of benefits assessment, our purpose should not be to
resolve the hazard question, but to figure how best to hedge our estimates of cancer risks to account for
the ambiguity. At present, there is no rigorous analytical scheme for doing this, so we need to rely on
some kind of expert judgment. For sake of argument, I propose to put 10% weight on the possibility that
perchloroethylene does indeed pose a human cancer risk (at some levels of exposure relevant to the
assessment), and 90% on the possibility that it does not. My judgment attempts to account for the
inconsistency of results among epidemiological studies, the likelihood that exposures to other agents or
confounding by smoking or alcohol apply, the inconsistency among animal cancer results and lack of
concordance with observations in humans, and the lack of biological hypotheses for why esophageal
cancer in particular should be caused by perchloroethylene.
A variant of this approach would be to make separate judgments about each potential basis for a human
cancer risk estimate, i.e., a judgment about the esophageal cancer, about the bladder cancer, about the
hematopoietic cancers, etc. seen among the human studies, as well as judgments about the liver cancer,
leukemias, and kidney tumors in the animal studies. Each weight could then be multiplied by the study-
specific estimate of risk (made contingent on its presumed relevance). This appropriately allows some
(very small) probability that, say, both the mouse liver tumors and the human-study bladder tumors are
indicating some actual human cancer risk from perc.
At this point it is probably wise to emphasize the distinction between using such a hedging approach for
setting a regulation in the first place and for estimating the benefits of a regulation set by some other
reasoning. In my view, some degree of conservatism and precaution in setting allowable exposures is
legitimate. What we get for our money is not just the reduction of health impacts, but some degree of
assurance that we have done enough to protect public health. Nonetheless, when the question is the
estimation of what the regulation has accomplished, what is needed is our best attempt to make objective
estimates of the relative likelihoods that various levels of benefit have been achieved. Such an analysis
informs not only the expected benefits (the mean over possibilities) but also the assessment of how much
assurance we have in fact achieved.
The next question is to ask what the cancer potency is in humans, contingent on our provisional
consideration that there is one. The problem most often pointed to in this realm is that current methods
for describing dose-response relationships define "upper bound" risks rather than central estimates. As
noted previously, the solution is not to use the single best-fitting dose-response equation (the maximum
likelihood estimate), since this fails to express the variety of more-or-less reasonable dose-response
relations and does not in general reflect the expected value of the risk.
Instead, a useful approach is to conduct a bootstrap analysis of the dataset. In this simulation-based
approach, a large number of alternative datasets are generated by resampling the original data (with
replacement), and a best-fitting curve is generated for each iteration. This expresses the variation in low-
dose potency to be expected as a result of the experimental error inherent in a limited number of
observations, and the mean of the distribution gives an unbiased estimate of the expected value over the
various possible values, with each possibility weighted by its likelihood of occurrence.
F-3-8 Gradient Corporation
-------
Although this kind of experimental error is what is allowed for in the upper-bound calculations of
traditional methods, such error does not constitute the only, or even the primary, source of uncertainty in
estimates of human low-dose cancer potency. There are many analytical choices made in the projection
of animal-based risk estimates to humans. Notably, these include the choice of dose-response model to fit
(which ideally should reflect understanding of underlying biological mechanisms of the agent's toxic
action) and the means for determining the lexicologically equivalent exposures in the experimental
animals and in humans. Such factors should be thought of as aspects of model uncertainty, since they
reflect not alternative realizations of some underlying distribution, but rather our uncertainty as to what
structure for the analytical approach gives the best projections.
Several studies have attempted to address this kind of uncertainty by an extension of the "hedging"
approach described above. Each analytical choice is expressed as a stated set of alternatives, and the
alternatives are then given weights to reflect the perceived relative plausibility of the approaches they
embody. In a simulation approach, one can then iterate the analysis many times, each time choosing one
of the alternatives for each factor with a likelihood proportional to the weights they have been assigned.
The resulting distribution of outcomes gives a description of the array of possible overall analytical
answers and their relative plausibility. McKone and Bogen (1992) applied this approach to
perchloroethylene cancer risks from contaminated drinking water, although they gave equal weights to all
the alternative datasets and analytical methods considered. Thompson and Evans (1997) built on this
approach to consider cancer risks from perc use in dry cleaning. A major advantage of such analysis is
that it allows examination of the contribution to overall uncertainty from the various components, and it
lends itself to value-of-information analysis that seeks to define how investment in research efforts to
reduce key uncertainties can be expected to pay off in informing regulatory decisions. (They found that
the expected value of perfect information about perchloroethylene's potency exceeds that about
exposures.)
Evans et al. (1994) applied a more extensive version of this approach to the description of the
carcinogenic potency of chloroform. They used a panel of experts to provide weights on the various
analytical choices, and they allowed for the weights placed on alternatives for one factor to be contingent
on choices for other factors. They found a wide but not unreasonable distribution of implied potencies.
The then-existing EPA potency estimate fell at a high percentile of the estimates, as is appropriate for an
upper bound, but the whole distribution provides perspective on the expected amount of benefit that limits
on chloroform exposure could be thought to achieve.
This process of elaboration of possible alternatives could be drawn out indefinitely, so one has to devise
an approach that captures the main sources of uncertainty and describes them adequately for the purposes
at hand. In the case of perchloroethylene, we have several (poor) choices of datasets to analyze (and
hence a large weight on the notion that none of them is applicable), several alternative pharmacokinetic
models, each of which could be subject to a characterization of the uncertainty distribution of its estimates
of values of several different dose measures (reflecting different perchloroethylene metabolites in
different tissues, in mice, rats, and humans), with different dose-response approaches to be considered in
view of judgments about mechanism of carcinogenic action. Clearly, the approach is not easy to
implement, but simplified versions could be used to give a reasonable view of the uncertainty about
projections of human cancer risk.
Once we have such projections, we need to deal with the fact that they are unspecific about the kind of
cancer to be expected in humans as well as the time of appearance of any tumors that are in fact caused.
At present, there is no very satisfactory method for specifying these, but it is worthwhile considering how
important it really is to do so. If we assume that most cancers have roughly similar impact on length and
F-3-9 Gradient Corporation
-------
quality of life, and if we assume that induced cancers appear with the same distribution over ages as the
general burden of background cancers, we will probably not be far off.
4 Projecting Noncancer Health Effects
Many of the issues just discussed regarding cancer risks apply to noncancer risks as well, but there are
some additional questions to be considered.
First, "noncancer" toxicity is a catchall category, and a single chemical may cause several different kinds
of noncancer effects. In the traditional assessment process, a critical effect is identified as the basis of
setting an exposure below which no adverse responses are expected, but above such a level, various
toxicities may be caused, and as doses increase, the number of endpoints that may become important may
increase, as effects with higher and higher population thresholds come into play. For instance,
moderately high doses of perchloroethylene may cause neurological effects, and still higher ones may
cause these plus renal toxicity. We must therefore keep in mind that a series of parallel endpoint-specific
assessments is necessary, and not just an assessment of the endpoint on which the RfD is based.
Second, noncancer endpoints vary considerably in their severity. This is always part of the debate about
"adversity" that arises when one is defining the critical effect. A benefits assessment must consider the
fact that avoidance of some effects that are not frankly adverse may nonetheless have some value (albeit
less than might be ascribed to a more severe effect). It may be legitimate, therefore, to assess endpoints
that would not be considered a basis for an RfD, but nonetheless affect quality of life. For example,
avoidance of headaches and dizziness from perc inhalation may be validly considered as benefits of
regulation of workplace levels, even if they are not strictly "toxic" effects.
Third, since severity can vary a good deal, it becomes especially important to identify the nature of the
toxic effects that may be engendered. As with cancer assessment, traditional methods do not specify what
effects may be risked at doses above those deemed "safe," and it is not generally presumed that humans
will have the same toxic effects as those seen in experimental animals, but such presumptions are
necessary for benefits analysis to gain specificity.
Fourth, unlike cancer, the severity (and not just the frequency) of response increases with increasing dose.
Much toxicity data is expressed in quantal form (with or without an effect of a given grade), and the
increasing health impact of higher doses on those individuals showing effects may not be readily
described. Since people vary in their tolerance of exposures to agents, at some doses, some individuals
will respond and others will not. At higher doses, more people in an exposed population will respond, but
those who already responded at a lower concentration will have more severe effects at a higher one. As a
consequence, the mix of severity of responses will vary with dose.
Fifth, unlike cancer, which once started becomes autonomous and independent of the dose that caused it,
noncancer effects may (or may not) be dependent on continued exposure. For exposures that can be
avoided, and for toxic effects that become evident with relatively short latency, it may be that sufferers of
moderate symptoms remove themselves from exposure and limit the impact on their health. (Of course,
the need to do so might be considered a non-health impact to which value might be ascribed.) For
example, someone experiencing mild neurological symptoms from perc exposure on the job might seek
reassignment. On the other hand, an effect on a pregnancy outcome provides no opportunity to detect and
avoid a developing problem.
F-3 -10 Gradient Corporation
-------
Sixth, in a similar vein, different endpoints will have different latencies, typical durations, tendencies to
progress or resolve, and different degrees of recovery or reversibility being possible. The impact on
quality of life will depend heavily on whether the effect appears early or late in life, whether it is
permanent or reversible, and whether it gets worse with time, with or without continued exposure. These
are not matters treated in traditional assessments of noncancer risk.
Seventh, many noncancer endpoints are defined and measured in terms of markers or indicators of effects,
but the endpoints themselves are not the primary concern. For example, the effects of perchloroethylene
on finger-tapping frequency or color discrimination are examined because these objective tests are
thought to be measurable manifestations of underlying neurological impacts. The benefit of restricted
exposure is not in better finger-tapping ability or fineness of color discrimination, but in freedom from the
underlying neurotoxicity that these markers are presumed to reflect. The quantitative connection of
marker effects with the impairments of the underlying system being affected are not always very clear.
Eighth, the reason that differing levels of response are seen at different doses for effects presumed to have
a threshold is that different individuals have different tolerances, or individual thresholds, or degrees of
reserve capacity. Those who respond at the lowest doses will be those in the population with the least
reserve capacity. It may be that such people are very nonrandomly distributed over age and other
demographic categories, and it may be that those prone to response are prone because of pre-existing ill
health or marginal health, and their change in health state may be different than is assumed if effects are
thought to fall randomly on the exposed members of a population.
Finally, traditional approaches to noncancer risk assessment make little attempt to characterize the
quantitative changes in probability of response with changing dose levels. The focus is on finding
NOAELs or benchmark doses—doses substantially without effect—rather than to map the shape of the
dose-response relationship. Moreover, the means to extrapolate effects from animals to humans are not as
well developed as for cancer assessment. The extrapolations are covered by "uncertainty factors" that act
in part to make extrapolation corrections (to human equivalent doses or to particularly sensitive humans)
and in part to allow for case-by-case uncertainty about how big an extrapolation correction to make. That
is, the analysis is more of a safety assessment than a risk assessment, and impacts of exposures above the
RfD are not readily characterized.
Methodological changes are needed that make noncancer risk analysis capable of explicit estimation and
extrapolation. This requires separating the two roles of the uncertainty factors into unbiased estimates of
extrapolation and additional allowances for uncertainty in those extrapolations. One promising approach
is to use (in place of fixed uncertainty factors) empirical distributions over many agents of the magnitude
of extrapolation needed. Baird et al. (1996) have explored such an approach. In current ongoing work, I
and my colleagues (Sandra Baird, John Evans, Paige Williams, Andrew Wilson) are further developing
this approach for the assessment of reproductive and developmental toxicity of ethylene oxide in humans.
We use empirical distributions over many chemicals of species differences in lexicologically equivalent
doses for noncancer effects as well as empirical information about interindividual variation in sensitivity
to arrive at unbiased estimates (with characterization of uncertainty) of the human dose-response
relationship. Such results are suitable for making estimates of impacts of exposures at different dose
levels, including those above traditionally defined reference doses. The result of this analysis is a set of
distributions of the uncertainty in doses expected to lead to different levels of response in an exposed
human population, the kind of assessment that is needed for analysis of benefits of regulation for
noncancer endpoints.
F-3 -11 Gradient Corporation
-------
5 Conclusions
Risk analysis in support of benefits assessment is different in aims from analysis for the setting of
regulatory levels as currently practiced. It needs to be focused on estimation of effects, not on the
bounding of regions of exposure where one can be very confident that unacceptable impacts are not to be
expected. Accordingly, methods of risk analysis for benefits assessment need to be somewhat different.
There are many profound challenges, but I have tried to show that they are approachable, at least in
concept. The appropriate analyses are not quick or easy, and there is no minor tweak to existing methods
that will make them fully applicable. Having laid out attempts to define the ideal analysis, perhaps
simpler versions that are more readily conducted will become evident.
It is important to distinguish the task of estimating actual health effects (and the uncertainty about that
estimation) from the task of setting regulatory levels. The difficulty in estimating benefits should be clear
from the above discussion. A good deal of judgment is necessary, and there is likely to be controversy in
specific cases about the weights to be put on alternative possible estimates of the health effects
engendered by an exposure. This makes it difficult to use analysis of benefits and costs to define what
acceptable exposure levels should be. This being said, there is value in using such analysis to gauge how
much value is gained from regulation, and how much uncertainty there is about the magnitude of such
gain.
F-3 -12 Gradient Corporation
-------
Acknowledgements
Leslie Beyer and Eric Dube of Gradient Corporation contributed to the summary of toxicity of
perchloroethylene. Production of this paper was supported by the U.S. Environmental Protection
Agency's Office of Air and Radiation under Order No. OD-6263-NALX.
References
ATSDR (Agency for Toxic Substances and Disease Registry) 1997. Toxicology Profile for
Tetrachloroethylene: Update. US Department of Health and Human Services, Public Health Service,
Atlanta, GA.
Baird SJS, Cohen JT, Graham JD, Shlyakhter AI, Evans JS. 1996. Noncancer risk assessment: a
probabalistic alternative to current practice. Human Ecol. Risk Assessment 2:79-102.
EPA (US Environmental Protection Agency) 1985. Health Assessment Document for Tetrachloroethylene
(Perchloroethylene). (EPA/600/8-82/005FA) US Environmental Protection Agency, Office of Health and
Environmental Assessment, Washington, DC.
EPA (US Environmental Protection Agency) 1991. Response to Issues and Data Submissions on the
Carcinogenicity of Tetrachloroethylene (Perchloroethylene). Review Draft. (EPA/600/6-91/002A) US
Environmental Protection Agency, Office of Research and Development, Washington, DC.
Evans JS, Gray GM, Sielken RI Jr., Smith AE, Valdez-Flores C, Graham JD. 1994. Use of probabalistic
expert judgment in uncertainty analysis of carcinogenic potency. Regul. Toxicol. Pharmacol. 20:15-36.
IARC (International Agency for Research on Cancer) 1995. IARC Monographs on the Evaluation of
Carcinogenic Risks to Humans. Vol.63. Dry Cleaning, Some Chlorinated Solvents and Other Industrial
Chemicals. World Health Organization, International Agency for Research on Cancer, Lyon, France.
ILSI (International Life Sciences Institute) 1998. Aggregate Exposure Assessment: An ILSI Risk Science
Institute Workshop Report. Risk Science Institute, International Life Sciences Institute, Washington, DC.
Kaplan S, Garrick BJ. 1981. On the quantitative definition of risk. Risk Anal. 1:11-27.
McKone TE, Bogen KT. 1992. Uncertainties in health-risk assessment: an integrated case study based on
tetrachloroethylene in California groundwater. Regul. Toxicol. Pharmacol. 15:86-103.
NCI (National Cancer Institute) 1977. Bioassay of Tetrachloroethylene for Possible Carcinogenicity.
Publication No. 77-813. National Cancer Institute, Department of Health and Human Services. Bethesda,
MD.
NTP (National Toxicology Program) 1986. Toxicology and carcinogenesis studies of tetrachloroethylene
(perchloroethylene) (CAS No. 127-18-4) in F344/N rats and B6C3FJ Mice (Inhalation Studies). National
Toxicology Program Technical Report TRY 311. National Institutes of Health.
OEHHA (Office of Environmental Health Hazard Assessment) 1999. Public Health Goal for
Tetrachloroethylene in Drinking Water. Review Draft. Pesticide and Environmental Toxicology Section,
Office of Environmental Health Hazard Assessment, California Environmental Protection Agency.
F-3 -13 Gradient Corporation
-------
Thompson KM, Evans JS. 1997. The value of improved national exposure information for
perchloroethylene (perc): a case study for dry cleaners. Risk Anal. 17:253-271.
F-3 -14 Gradient Corporation
-------
Appendix F-4
Calculating the Economic Benefits of Reductions in Manganese Air Concentrations
White Paper by Dr. Bernard Weiss,
University of Rochester, Rochester, NY
F-4-1
-------
Assessing Benefits of Reductions in Manganese Air Concentrations
Bernard Weiss
Department of Environmental Medicine
University of Rochester
School of Medicine and Dentistry
Dr. Bernard Weiss
University of Rochester Medical Center
Rochester NY 14642
Tel: 716-275-1736
Fax: 716-256-2591
e-mail:
F-4-2
-------
Introduction
Manganese presents a conundrum for risk assessment because it is both
an essential nutrient and a potent neurotoxicant. Its neurotoxic properties have
emerged almost exclusively from inhalation exposures, although some
epidemiological data suggest that high concentrations in drinking water may be
associated with neurological impairment. Several kinds of occupations expose
workers to inhaled manganese, the most prominent being mining, ore-crushing,
and ferro-manganese production. Mining for manganese ore provides the best
documented association owing to the high levels of Mn02 dust encountered in
the process.
Table 1 lists some of the characteristic signs and symptoms of
manganese neurotoxicity. Some, like bradykinesia, are also distinguishing signs
of Parkinson's disease. Others, like the kind of emotional lability marked by
abnormal laughing (and crying), are distinctive for manganese. In South
American mining communities familiar with manganese intoxication, such a
syndrome has earned the label, "locura manganica," or manganese madness,
often viewed as the first stage of the full syndrome of manganese intoxication.
The most suitable animal model for research into manganese
neurotoxicity is the nonhuman primate. Because of the unique organization of the
primate brain, other animal models, such as rodents, are not as satisfactory,
although they may yield useful information about neurochemical processes.
Figure 1 depicts these differences as the relationship between dose and
measure and shows, roughly speaking, a difference in sensitivity of close to two
orders of magnitude between primates and rodents. One factor that may account
for some of the discrepancy is the lack of advanced tests for motor function in the
rodent studies comparable to the effortful response criterion used by Newland
and Weiss (1992) in trained monkeys.
Although the motor signs exhibited by Mn miners correspond in part to
those seen in Parkinson's disease (PD), enough differences are visible to
question the widely-held proposition that Parkinson's and manganism are
virtually identical. Barbeau (1984) suggested that the syndrome more closely
resembled a dystonia than classical PD, a point of view also supported by Pat et
al (1999) and others. Neuropathological observations support this distinction.
With manganism, the main evidence of degeneration is seen in the globus
pallidus, with less severe damage in the striatum (putamen and caudate nucleus)
and in the substantia nigra pars reticulata. In contrast, the primary lesions seen in
PD lie in the substantia nigra pars compacta and consist of depigmented and
missing neurons, viewed as the dominant morphological markers of PD,
accompanied by Lewy bodies, which consist of abnormally aggregated proteins
found largely in dopaminergic neurons and recently shown to also contain the
protein alpha-synuclein.
F-4-3
-------
Convincing evidence of globus pallidus involvement also comes from
magnetic resonance imaging (MRI) data. Because manganese is a paramagnetic
metal, it modifies the return of protons to their original orientation after
displacement by a strong magnetic field. These shortened times can then be
used to produce different degrees of brightness in the calculated image that are
related to local manganese concentration. The images and plots published by
Newland et al (1989) and Newland and Weiss (1992) show the highest
concentrations in the vicinity of the globus pallidus in exposed monkeys. MR
images of an arc welder who had been exposed in the process of repairing and
recycling railroad track made of manganese steel alloy also showed localized
deposition in the globus pallidus (Nelson et al, 1993).
Ingested manganese is closely regulated by the gut. Inhaled manganese
bypasses the gut, and can enter the brain in two ways. First, as described by
Tjalve and Henriksson (1999), the olfactory pathways provide a direct path into
brain tissue. Rats given 54Mn intranasally accumulated the metal in a variety of
brain structures, including the basal ganglia. Primates exposed by inhalation to
trace amounts of 54Mn showed a rise in brain levels that peaked at about 40 days
(Newland et al, 1987). The manganese disappeared from brain much more
slowly, with half-lives of 223 to 267 days in the two monkeys studied. 54Mn was
detected in the lungs for 500 days after exposure, suggesting that they served as
a reservoir for uptake into brain (Figure 2). Although these data may also reflect
some storage in bone, as noted by Andersen et al (1999), they indicate the
strong possibility that long residence times in the lung provide a continuing
source of brain exposure. This may be a special problem for young children in
areas where dense vehicular traffic deposits manganese-laden dust. As with lead
(cf., Lanphear et al, 1998), typical children's activities in high dust areas will
expose them to elevated levels of both inhaled and ingested manganese, and
Dorman et al (2000) have recently shown that neonatal rats administered
manganese orally may be at greater risk for Mn-induced neurotoxicity than adult
rats.
Most of the data pertaining directly to the benefits issue come from
occupational studies. Table 2 gives the details of some of the important studies
that attempted to relate exposure to neurobehavioral endpoints. The mean blood
concentrations of exposed workers, except for Chia et al (1993), hover near 10
• g/L, with their controls at one-half to two-thirds that value. Chia et al, however,
studied a population in Singapore whose dietary habits undoubtedly differed from
those in the other studies. Table 3 compares the results of a number of studies
based on neurobehavioral endpoints. What is most evident there is the apparent
sensitivity of motor function tests to manganese exposure, a result consistent
with the evidence showing the main sites of deposition to lie in the basal ganglia,
particularly the globus pallidus. Table 4 (Lucchini et al, 1999) offers more recent
data from the population studied by Lucchini et al (1995). It too shows that mean
control blood values are two-thirds of exposed values, meaning that an elevation
F-4-4
-------
of one-third above baseline accounts for the performance differences between
the two populations of workers. Also, note the closely overlapping ranges. Figure
3 plots the relationship in this population between air concentration and blood
level in the work environment. Two features deserve comment. One is that even
negligible air concentrations are associated with blood levels, as noted above,
not overwhelmingly different from much higher concentrations. The second is
that, at least in these workplaces, the distribution of exposure, as the authors
note, is log-normal, with most workers clustered at the low end.
Worker populations present special problems for risk assessment. The
healthy worker effect, a notorious confounder in epidemiological investigations,
reduces the accuracy with which occupational data can be extrapolated to
groups such as children, the elderly, or other especially susceptible populations.
Moreover, standards such Threshold Limit Values and Permissible Exposure
Limits are based on 8-hour days and 40-hour weeks rather than continuous
environmental exposure. To more directly determine potential manganese
toxicity in the general population, Mergler et al (1999) undertook a community
study in southwest Quebec. The subjects ranged from 20 to 69 years of age and
had not experienced any workplace exposures. The entire study sample of 297
subjects was about equally divided between men and women.
Table 5 presents the blood values. They show slightly higher levels in the
women than in the men, but totally overlapping ranges. The investigators
administered the most extensive series of neurobehavioral tests ever used to
study manganese, and based most of their analyses on a separation of subjects
based on blood levels. A value of 7.5 • g/L served as the dividing concentration.
Age was chosen as a second dichotomous variable separating subjects below
and above 50 years of age.
The neuropsychological measures adopted by Mergler et al (1999) and
that documented evidence of adverse effects are listed in Table 6. The first four
are indices of motor function and the first three are described at length by Beuter
et al (1999). The motor function measures yielded convincing relationships, but
their most interesting features are their dependence on age. Figure 4 displays
the interaction between manganese blood level (above or below 7.5 • g/L) and
age (above or below 50 years) for the index used to describe performance on a
task requiring the subject to alternate strikes with a stylus at two spatially
separated targets. This pattern, showing a persuasive influence of age, is
consistent with most of the data from this study.
Neurodegenerative diseases, like most other degenerative diseases, are
typically diseases of aging, with both incidence and prevalence rising with
advancing age. One useful way to contemplate the potential impact of neurotoxic
chemicals is to evaluate how they might shift the relationship between
prevalence and age. A model of how even small shifts in a population distribution
F-4-5
-------
can incur large public health costs is seen in Figure 6. It depicts the
consequences of a 3-point or 3% shift in mean IQ score, the kind of shift
produced by small elevations in lead exposure. It shows that even that small a
shift produces a significant increase in the number of individuals classified as
mentally retarded. It incurs massive expenses in remedial care and education,
but also produces a significant decrease in the number of individuals in the
superior range (e.g., IQ>130). Figure 7 shows that even a 1% leftward shift, or
one IQ point, is a significant societal burden. In its evaluation of the benefits
stemming from the removal of lead from gasoline, EPA, basing its calculations on
the relationship between IQ score and lifetime earnings, estimated benefits
approximating one trillion dollars.
A variation of this logic can be applied to manganese given the
assumption that it can contribute to neurodegenerative disease. First, consider
Figure 8, which depicts the reduction in nerve cell density with age that occur in
certain brain structures. McGeer et al (1988) plotted the relationship between age
and nerve cell number in the substantia nigra (SN). A key pathological marker of
PD is loss of pigmented neurons in one part of SN. Figure 8 demonstrates that
an acceleration of this natural loss by 0.1 % annually will, over several decades,
produce what might be termed premature aging of this structure. If the natural
course of aging produces a loss of 40% by age 73, say, an additional
acceleration of 0.1 % will incur such a loss about ten years earlier.
Assume exposure to an agent that produces such a superficially minor
acceleration. Figure 9 shows the consequences for the prevalence of PD of
accelerations of 5 and 10 years respectively. The consequences are hardly
minor. Table 7 takes the prevalence figures on which Figure 9 is based, and,
from the projected US age distribution (US Census) in 2005, shows the baseline
rates of PD and their estimated medical costs, and compares them to what would
be expected if the age distribution were to be shifted by five years. The
differences are considerable, and would result from an acceleration of functional
loss of less than 0.1% annually (see Figure 8). For the age group 60-64, the
increment in annual costs is over 700 million dollars.
Would this be a reasonable model for manganese? Or, put another way,
what evidence is there to support a contribution by manganese exposure to PD
or other neurodegenerative diseases?
One source of evidence is manganese poisoning, which confirms that
manganese is a powerful neurotoxicant, producing the kinds of clinical signs,
largely irreversible, listed in Table 1. A second source of evidence comes from
detailed studies both of communities and of workers indicating that exposed
populations displaying no signs of clinical disease can nevertheless be shown to
suffer from neuropsychological deficits detected by appropriate testing
procedures. But this kind of evidence is not specific to PD.
F-4-6
-------
What is specific to PD, however, is both research and experimental data
implicating the central nervous system structures targeted by manganese in PD.
To incorporate these results into a benefits analysis first requires some probing
into the potential relationship between manganese and neurodegenerative
disease. It will be especially illuminating to examine how it might relate to PD
because it is a clear example of a relationship with age. As noted earlier, the
globus pallidus, on the basis of both chemical analyses and MRI, appears to
accumulate manganese in greater quantities than other basal ganglia structures
and is the site of lesions produced by manganese in appropriate doses. Although
neuropathology does not indicate manganese-induced damage to the structure
directly implicated in PD, the pars compacta of the substantia nigra, a great deal
of evidence links its function with the globus pallidus.
One measure of the critical role played by GP in PD is the burgeoning
literature on attenuation of PD symptoms by pallidal surgery or stimulation.
Electrical stimulation of the internal pallidum may reduce the fluctuations
associated with medication such as L-dopa, and permit a reduction in dosage.
Pallidal surgery is now an accepted method for bringing substantial relief to PD
patients. In addition, electrophysiological studies indicate a role for the globus
pallidus in the resting tremor displayed by PD patients. Figure 5 shows the
linkages among various basal ganglia structures and emphasizes the lack of
isolation among them.
One conclusion to be drawn from this information is that what are called
extrapyramidal diseases possess commonalities arising from their intimate and
extensive structural and chemical interconnections. Damage to one component
of the basal ganglia almost surely is bound to exert influence on functions
subserved by other structural components, as in the overlapping symptoms of
PD and Alzheimer's disease. In addition, the disabling effects of pharmacological
therapies for PD, such as the dyskinesias resulting from L-dopa, are improved by
pallidal stimulation, another source of evidence for the intimate links between GP
and SN. A neat piece of evidence come from an experiment with monkeys
(Zhang et al, 1999) rendered hemi-parkinsonian by an injection into the right
carotid artery of MPTP (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine), originally
discovered as a contaminant in designer drugs that produced parkinsonian signs
in drug addicts. Figure 10 shows that monkeys with pallidal damage resulting
from MPTP were less responsive to the amelioration of PD signs than monkeys
lacking evidence of damage.
Both community surveys (Mergler et al, 1999) and studies of worker
populations (e.g., Apostoli et al, 2000) suggest that relatively small increments in
manganese blood levels are associated with significant diminutions in
neurobehavioral function. If these functional indices are assumed to reflect
deficits in brain function, and if we pair these deficits with the recognized declines
in brain compensatory capacity associated with aging, slight elevations in
F-4-7
-------
airborne manganese might produce a small, but medically and economically
significant shift to an earlier onset of neurodegenerative diseases such as
Parkinson's disease.
Small" and "significant" need to be seen in context. An aging population is
beginning to confront us with difficult medical and economic choices, and
the most overwhelming problem is certain to be neurodegenerative
diseases. In evaluating the potential contributions of environmental
neurotoxicants to this problem, a simple calculation will prove illuminating.
If the entrance of 30 patients into institutional care is delayed by one year,
the savings amount to over one million dollars.
Reference List
Andersen ME, Gearhart JM, Clewell HJ . Pharmacokinetic data needs to support
risk assessments for inhaled and ingested manganese. Neurotoxicology
1999;20:161-71.
Apostoli P, Lucchini R, Alessio L . Are current biomarkers suitable for the
assessment of manganese exposure in individual workers? Am J Ind Med
2000;37:283-90.
Barbeau A . Manganese and extrapyramidal disorders (a critical review and
tribute to Dr. George C. Cotzias). Neurotoxicology 1984;5:13-35.
Beuter A, Edwards R, deGeoffroy A, Mergler D, Hudnell K . Quantification of
neuromotor function for detection of the effects of manganese.
Neurotoxicology 1999;20:355-66.
Chia SE, Foo SC, Can SL, Jeyaratnam J, Tian CS . Neurobehavioral functions
among workers exposed to manganese ore. Scand J Work Environ Health
1993; 19:264-70.
Dodel RC, Singer M, Kohne-Volland R, Szucs T, Rathay B, Scholz E, Oertel WH.
The economic impact of Parkinson's disease. An estimation based on a
3-month prospective analysis. Pharmacoeconomics 1998; 14:299-33312.
Dorman DC, Struve MF, Vitarella D, Byerly FL, Goetz J, Miller R . Neurotoxicity
of manganese chloride in neonatal and adult CD rats following subchronic
(21-day) high-dose oral exposure. J Appl Toxicol 2000;20:179-87.
Durif F, Lemaire JJ, Debilly B, Dordain G. Acute and chronic effects of
anteromedial globus pallidus stimulation in Parkinson's disease. J Neurol
Neurosurg Psychiatry 1999;67:315-22.
Fearnley JM, Lees AJ. Ageing and Parkinson's disease: substantia nigra regional
selectivity. Brain 1991 ;114:2283-301.
Fine J, Duff J, Chen R, Chir B, Hutchison W, Lozano AM, Lang AE. Long-term
follow-up of unilateral pallidotomy in advanced Parkinson's disease. N
Engl J Med 2000;342:1708-14.
Goyer R. Toxic effects of metals. Amdur MO; Doull J; Klaassen CD, eds.
Casarett and Doull's Toxicology. 4th ed. New York: Pergamon Press;
1991. pp. 623-80.
Gross RE, Lozano AM . Advances in neurostimulation for movement disorders.
F-4-8
-------
Neural Res 2000;22:247-58.
Hauber W, Lutz S . Dopamine D1 or D2 receptor blockade in the globus pallidus
produces akinesia in the rat. Behav Brain Res 1999; 106:143-50.
Hochberg F, Miller G, Valenzuela R, McNelis S, Crump KS, Covington T, Valdivia
G, Hochberg B, Trustman JW. Late motor deficits of Chilean manganese
miners: a blinded control study. Neurology 1996;47:788-95.
Huang CC, Chu NS, Lu CS, Chen RS, Calne DB . Long-term progression in
chronic manganism: ten years of follow-up. Neurology 1998;50:698-700.
Kessler II. Epidemiologic studies of Parkinson's disease. 3. A community-based
survey. Am J Epidemiol 1972;96:242-54.
Kim Y, Kim KS, Yang JS, Park IJ, Kim E, Jin Y, Kwon KR, Chang KH, Kim JW,
ParkSH, et al. Increase in signal intensities on T1-weighted magnetic
resonance images in asymptomatic manganese-exposed workers.
Neurotoxicology 1999;20:901-7.
Lanphear BP, Matte TD, Rogers J, Clickner RP, Dietz B, Bornschein RL, Succop
P, Mahaffey KR, Dixon S, Galke W, et al. The contribution of
lead-contaminated house dust and residential soil to children's blood lead
levels. A pooled analysis of 12 epidemiologic studies. Environ Res
1998;79:51-68.
Lombardi WJ, Gross RE, Trepanier LL, Lang AE, Lozano AM, Saint-Cyr JA .
Relationship of lesion location to cognitive outcome following
microelectrode-guided pallidotomy for Parkinson's disease: support for the
existence of cognitive circuits in the human pallidum. Brain 2000;123(Pt
4):746-58.
Lozano AM, Lang AE, Hutchison WD. Pallidotomy for tremor. Mov Disord
1998; 13(Suppl 3): 107-10.
Lucchini R, Apostoli P, Perrone C, Placidi D, Albini E, Migliorati P, Mergler D,
Sassine MP, Palmi S, Alessio L . Long-term exposure to "low levels" of
manganese oxides and neurofunctional changes in ferroalloy workers.
Neurotoxicology 1999;20:287-97.
Lucchini R, Selis L, Folli D, Apostoli P, Mutti A, Vanoni 0, Iregren A, Alessio L .
Neurobehavioral effects of manganese in workers from a ferroalloy plant
after temporary cessation of exposure. Scand J Work Environ Health
1995;21:143-9.
McGeer PL, Itagaki S, Akiyama H, McGeer EG. Rate of cell death in
Parkinsonism indicates active neuropathological process. Ann Neurol
1988;24:574-6.
Mergler D, Baldwin M, Belanger S, Larribe F, Beuter A, Bowler R, Panisset M,
Edwards R, de Geoffrey A, Sassine MP, et al. Manganese neurotoxicity, a
continuum of dysfunction: results from a community based study.
Neurotoxicology 1999;20:327-42.
Nelson K, Golnick J, Korn T, Angle C . Manganese encephalopathy: utility of
early magnetic resonance imaging. Br J Ind Med 1993;50:510-3.
Newland MC . Animal models of manganese's neurotoxicity. Neurotoxicology
F-4-9
-------
1999;20:415-32.
Newland MC, CecklerTL, Kordower JH, Weiss B . Visualizing manganese in the
primate basal ganglia with magnetic resonance imaging. Exp Neural
1989; 106:251-8.
Newland MC, Cox C, Hamada R, Oberdorster G, Weiss B . The clearance of
manganese chloride in the primate. Fundam Appl Toxicol 1987;9:314-28.
Newland MC, Weiss B . Persistent effects of manganese on effortful responding
and their relationship to manganese accumulation in the primate globus
pallidus. Toxicol Appl Pharmacol 1992; 113:87-97.
Pal PK, Samii A, Calne DB . Manganese neurotoxicity: a review of clinical
features, imaging and pathology. Neurotoxicology 1999 1999;20:227-38.
Papa SM, Desimone R, Fiorani M, Oldfield EH . Internal globus pallidus
discharge is nearly suppressed during levodopa-induced dyskinesias. Ann
Neural 1999;46:732-8.
Shinotoh H, Snow BJ, Hewitt KA, Pate BD, Doudet D, Nugent R, Perl DP,
Olanow W, Calne DB . MRI and PET studies of manganese-intoxicated
monkeys. Neurology 1995;45:1199-204.
Tjalve H, Henriksson J . Uptake of metals in the brain via olfactory pathways.
Neurotoxicology 1999;20:181-95.
Vitarella D, Wong BA, Moss OR, Dorman DC. Pharmacokinetics of inhaled
manganese phosphate in male Sprague-Dawley rats following subacute
(14-day) exposure. Toxicol Appl Pharmacol 2000; 163:279-85.
Weiss B . Vulnerability of Children and the Developing Brain to Neurotoxic
Hazards. Environ Health Perspect 2000;108(Suppl 3):375-81.
Zhang Z, Zhang M, Ai Y, Avison C, Gash DM. MPTP-lnduced pallidal lesions in
rhesus monkeys. Exp Neural 1999; 155:140-9.
F-4-10
-------
Appendix F-4 Figures and Tables
F-4-11
-------
Primates
W
W
W
W
w
I—
100
BA in CD, GP <7>
Hyperactive > action tremor > clumsy <7)
Action tremor (7)
- - Unsteady gait, clumsy (8)
-- Hyperactive, uncoordinated CO)
- - Dl binding <6>
Reduced raelopride, nomifensin bindingCH)
€3oTitt"a€tut*€, u*MSG&rdittsti®n (IS)
-• Intention tremor (6)
Tremor, ujneoordlnation (4)
-f DA in Caudate C4)
j- Enhanced MRI: Striatum (8)
T Action tremor (1)
Bradykinesia, rigidity (2)
- - Incomplet* effortful (1)
MRI: Psllidum and Pituitary (1)
JO 100
KmlciltH
Doee of
DA, GABA <11)
-Alter dopsmln* (10)
-Altered DA, (9)
-Decreased GABA, iacrems«cj (8)
rlncraaae GABA in (7)
[•Increased TYK hydroxylase (1)
HPecreased in {6)
f-Transient incresse-loconacitor activity CS>
AChE (4)
rTransient inerease-loeomotoi" activity (8)
^Altered DA, NE (2)
Altered TRY, DA, ME (!)
rrmirt
Figure 1. Relationships between administered manganese dose and indices of
neurotoxicity in primates and rodents (Newland, 1999).
F-4-12
-------
10,000
•S 1,000--
1GJ
SO
100
150 200 250 300 350
Days Since Exposure
400 450
500
Figure 2. Radioactivity in the head after inhalation of Mn in two monkeys (M.
nemestrina). From Newland et al, 1987.
F-4-13
-------
o
o
o
0 10O 200 3OO 4OO 5QQ 7OO
{in
Figure 3. Relationship between air and blood manganese levels in exposed
workers (Apostoli et al, 2000)
F-4-14
-------
EKM: Irregularity
.us •
.04 -
.03 -
.02 -
.01 ~
0
-.01 '
I
T
41
A
5 1 57
83 83
<50 >50
age category (years)
(Mn:p 7.5 ug/L. O = MnB < 7.5
ug/L. From Mergler et al, 1999.
F-4-15
-------
PALUOUM
gr-'j Primary
g^j Supplementary
fi. J Premotor
V4-i
Ip} Pf«ironiiil
S Limble
Figure 5. Globus Pallidus connections. PUT=putamen; SM=sensorimotor circuits.
F-4-16
-------
IQ Distributions:
Effects of a 3% Shift in the Mean
48 IM> Sft
«.ft-l
I »-«-
I
£ set
s.oo-^
Upper chart shows an IQ
distribution with a mean of
100 and SD of 15. The dark
area represents the 2.3%of
the population below 70. The
light area represents those
with IQs below 100 and above
70.
The lower chart depicts an IQ
distribution with a mean of
97. Here, 3.2% of the
population falls below 70. IQ
of 100 is shown on both
charts.
Figure 6. Consequences for classification of Mentally Retarded (IQ<70) of a 3%
shift in the IQ distribution. From Weiss, 2000.
-------
Figure 7. Proportion of Individuals in Retarded Range (IQ<70) with Different
Population Mean IQ Scores
0.05
0.04
0
I 0.03
O
t:
g- 0.02
Q.
0.01
0
, — j
—
i — I
—
'
— I
95 96 97 98 99 100 101 102 103 104 105
IQ
-------
Reduction in Cell Number with Age
rc
6
M-
o
c
o
'-E
o
Q.
2
o_
0)
a:
JO
"a>
O
McGeer et al
XtraO.1%
HIHXtra 0.3%
10 20
30
40 50
Age
60 70 80 90
Figure 8. Cell loss with age in substantia nigra
F-4-19
-------
1400
1200
o 1000
o
o"
o
g. 800
Q
Q.
"S
g 600
c
TO
400
200
A Baseline
-------
% of
to the
62 5
Figure 10. Comparison of therapeutic effectivenes of L-dopa in hemiparkinson
monkeys with and without collateral pallidal damage.
F-4-21
-------
Table 1. Signs and symptoms of manganese neurotoxicity
• Abnormal gait
• Impaired coordination
• Abnormal laughter
• Expressionless face
• Weakness
• Bradykinesia
• Somnolence
• Dysarthria
• Difficulty walking
• Clumsiness
• Lack of balance
• Muscle pains
* Diminished leg power
F-4-22
-------
Table 2. (Mergler and Baldwin, 1997)
Demographics and Internal Exposure Parameters for a Number of Comparable Studies of Active Workers
Reference
(exposed + controls)
Roels et al. (1987)
(141 + 104)
Wennberg e(o(. (1991)
(30 -t- 90)
Roels et al. (1992)
(92 + 101)
Chia et al. (1993)
(17 + 17)
Mergler et al. (1994)
(74 + 74)
Lucchini et al. (1995)
(n = S8P
Type of plant
Ma oxide and salt
Steel smelting
Dry alkaline battery
Mn ore milling
Ferro and silico
alloy production
Ferroalloy plant
Mean age
of exposed
34.3 ± 9.6
(19-59)
46.4
(19-63)
31.3 ± 7.4
(22-49)
36.6 l 12.2
43.4 ± 5.4
(32-58)
38.9
(20-53)
Years of
exposure,
mean ±SD
(range)
7.1 ±5.5
(1-19)
9.9
5.3 ±3.5
(0.2-17.7)
7.4 ±4.3
(1-14)
16.7 ± 3.2
(1-17)
(2-28)
MnBof
exposed,
geometric
mean
(ng/100 ml)
1.22*
—
0.81»
2.53
1.03*
1.19*
MnBof
controls,
geometric
mean
(ptg/lOO ml)
0.49
—
0.68
2.33
0.68
0.60
MnUof
exposed,
geometric
mean
ip, g/g cr)
1.59«
—
0.84*
6.1 fig/liter
0.73
2.8 jig/liter
MnUof
controls,
geometric
mean
(>i.g/S cr)
0.15
—
0.09
3.9 ^.g/liter
0.62
1.7 jig/liter
1 Lucchini et al. compared 19 low-exposure workers to 39 more highly exposed workers.
* The authors report significant differences between those exposed and control (P <0.05).
F-4-23
-------
Table
3. (Mergler and
Baldwin, 1997)
Results of Neurofunctional Assessment from Comparable Studies of Active Manganese-Exposed Workers
Reference
(exposed + controls)
Siegl and Bergert (1982)
(25 + 21)
RoelsefoZ. (1987)
(141 + 104)
Wennberg a al. (1991)
(30 + 90)
Wennberg et al. (1992)
(30 + 90)
Rods eioi. (1992)
(92 +101)
Chiae(o2.(1993)
(17 + 17)
Merglereioi. (1994)
(74 + 74)
Beutereioi. (1994)
(10 + 10)
Lucchini et al. (1995)
Motor
functions
—
4
4
—
—
4
4
—
4
Hand Response
steadiness speed
4
4 4
4
- -
4 4
— —
4 n.s.
_ _
— n.s.
Other
Diadocho- intellectual Olfactory Mood
kinesia Memory functions sensitivity state
_ _ _
4
— 4 n,s.
4 _ —
— n.s. —
4 4
— n.s. 4
4
4 4
- -
— -
- -
4
— —
- -
T 4
— -
" "
' Lucchini et al. compared 19 low-exposure workers to 39 more highly exposed workers.
-------
Table 4. Blood manganese levels (|jg/L) in exposed and control workers
(Lucchini et al, 1999).
Exposed
Control
Mean
9.71
6.00
Median
9.00
6.00
Range
4-19
2-9.5
Table 5. Blood manganese levels (|jg/L) in the Quebec community study
(Baldwin et al, 1999; Mergler et al, 1999)
All
Women
Men
Mean
7.50
7.90
7.00
Median
7.70
6.60
Range
2.5-15.9
2.8-15.9
2.5-13.9
-------
Table 6. Neuropsychological Measures used in the Quebec Community Study
•Eye-hand coordination
•Rapid pointing movements
•Tremor frequency and amplitude
•Neurological exam indices
•Learning and recall
•Mood
•Psychological symptoms
•Body sway
Table 7. Projected annual medical costs of Parkinson's disease, to age 74,
based on age-related prevalence comparing baseline age distribution with onset
accelerated by 5 years.
Age
(2005)1
35-39
40-44
45-49
50-54
55-59
60-64
65-69
70-74
Number
(x1,000)
20,082
22,634
22,230
19,661
16,842
12,848
10,086
8,375
Base PD
928
1,556
8,876
28,210
40,300
72,619
102,514
Base Cost
(x$1,000)2
12,584
21,099
120,359
382,528
546,680
984,714
1,390,090
+5PD
823
1,584
9,810
23,593
52,827
92,506
123,458
119,218
+5 Cost
(x$1,000)
11,160
21,479
133,024
321,277
716,334
1,254,381
1,674,090
1,616,596
Difference
(x$1,000)
11,160
8,895
111,925
200,918
333,806
707,701
689,376
226,506
1US Census Projections
2 Based on Dodel et al (1998) @$13,560/yr
F-4-26
-------
APPENDIX G: WRITTEN SUBMISSIONS FROM KEY DISCUSSANTS
Expert panelists and the Workshop Co-Moderator, Dr. Roy Albert, provided the
following written comments after the workshop on the questions discussed on June 23, 2000:
G.I. Comments of Dr. Roy Alpert, Division of Environmental Health, University of
Cincinnati
a) A Way to Characterize Overall Carcinogen Risk: If the EPA's Air Office is going to
estimate the monetary benefits of regulating carcinogenc HAPs, how will it deal with agents
whose probability of being a human carcinogen is less than certain? One can estimate the
number of cancer deaths by use of a dose response function and an exposure estimate on the
assumption that the agent in question is a human carcinogen. But suppose the evidence does not
permit us to say that it is definitely a human carcinogen? What then? One possibility is to
multiply the estimated number of cancer cases by a weighting factor that is determined by the
strength of the evidence. Suggested values for weighting factors are given in the table below.
The sue of the square of the weighting factors gives greater separation between strong and week
evidence.
EPA Category
A
Bl
B2
C
C
Descriptor
Definite
Probable
Probable
Possible
Possible
Weighting
Factor
1.0
0.75
0.5
0.25
0.12
(Weighting
Factor)2
1
0.56
0.25
0.06
0.01
Level of
Evidence
Sufficient
human and
anumal
2 species
postiive and
some human
2 species
positive
1 species both
sexes
1 species 1 sex
b) A Method for Best-Estimate and Uncertainty Characterization of Carcinogen Risks.
The stated purpose of the meeting was the development of a berst-estimate and uncertaty
characterization for hazard and dose-response functions fo use in benefit analysis of HAP
regulations. One possibility is offered below. The proposed method avoids large downward
extrapolations to vanishingly small levels of risk.
G-l
-------
The EPA's approach to carcinogen risk estimation was developed by its Carcinogen
Assessment Group (CAG) in the mid 1970s. The first method involved the use of the lowest
statistically signfiicant data point as the point of departure for a linear non-threshold
extrapolation to zero dose/zero incidence. Shortly after that, because of the emphasis on the
conservative approach to risk assessment, the linear non-threshold extrapolation sued the 95%
upper confidence limit of the lowest statistically significant data point as the point of deparrture
for the extrapolation to zero dose/zero incidence. The risk estimates obtained this way were
described as "plausible upper limit estimates, i.e., ones that were not likely to be highter than the
true risks but could be lower even to a considerable extent." After a few years, the statisticians
objected to the procedure, saying that all the data points above the lowest statistically significant
point were being wasted. After much discussion, the use of the multistage model was then
agreed upon and introduced by CAG and remains the method used up to the present.
The multistage model appealed to the statisticians because it is flexible enough to
accommodate almost any data set and it has biological plausibility, since cancer is multistage in
its development. Furthermore, the multistage model has a low-dose linear non-threshold
component. This low-dose linear non-threshold component depends on the assumption that the
carcinogen in question behaves like whatever it is that causes tumors in control animals (the
background). The 95% upper confidence limit for the low-dose linear non-threshold
extrapolation was carried over to the multistage model. In all the variants of the oinear
extrapolation, the background tumor incidence was subtracted from the incidence obatained with
each of the carcinogen doses so that the extrapolation could be extended to the zero dose/zero
incidence. No consideration was taken of the statistical uncertainty in the background incidence.
A more realistic estimate of risk with confidence limits can be made while retaining the
essential features of the established EPA approach. The multistage model can be retained on the
basis of the original rationale for using it. The low-dose linear non-threshold component can
also be retained on the grounds that carcinogens probably act the same way as the cuases of
background tumors. However, in the proposed approach we do not subtract out the background
tumor incidence in order to extrapolate down to zero dose/zero incidence. We extrapolate down
to the background tumor incidence. This follows logically from the assumption that the
carcinogen in question acts like whatever it is that causes background tumors. The dose
response curve is readjusted so that the administered dose is an increment to the equivalent
background dose as determined by the slope of the linear component of the multistage model.
The risk can be expressed as either the absolute incremental risk or the relative incremental risk.
For example a 1% absolute risk increment on a 5% background risk would be a relative
incremental risk of 20%.
There are two statistical uncertainties. The first deals with how well the data fit the
extrapolation model. The second uncertainty relates to the background cancer level. The overall
uncertainty of the risk estimates would be the combination of the two uncertainties.
The above approach was published in the context of time to tumor data rather than life-
time incidence. But the principle is the same. The reference is as follows:
G-2
-------
Albert, RE. and B Altschuler, 1976. Assessment of Environmental Carcinogen Risks in Terms
of Lifen Shortening. Environmental Health Perspectives 13: 91-94.
G.2. Comments of Dr. John C. Bailar III, Department of Health Studies, University of
Chicago, Chicago, IL.
How best to identify limitations and uncertainties in both risk assessment methods and
economic models.
Here I will summarize some of the important limitations and uncertainties, put them in a
context, and offer some suggestions about how to deal with them.
Cost-benefit analysis is an information-hungry process, which we must apply to an
information-sparse problem. This can be done, and it will be done, but the results will not be
pretty.
Further, regulatory decision-making based on cost-benefit analysis is a precision-hungry
process that we must necessarily base on precision-sparse inputs. Again, this can and will be
done, but the results will not be pretty.
The two fundamental problems are the enormous burden of work required to deal with
189 HAPs and the great uncertainty inherent in the estimates on both sides of this process - the
costs and the benefits — for each one of them. It may be close to impossible for conscientious
economists and conscientious risk assessors to do an even marginally competent job for each
HAP. It may be even more difficult to remain honest about the real level of uncertainty. The
more we have learned about any particular hazard, the more complex we have found it to be.
This seems to be a general phenomenon, and we should adapt to it. There is no reason to think
that any of the 189 HAPs is basically any simpler then benzene, though benzene looks quite
complicated because we know a great deal about it.
We never think we know as much as the users of our analyses demand, and we never
really know as much as we think we do. Uncertainty to three orders of magnitude is the norm in
risk assessment. When that is compounded with the deep and numerous problems of cost-benefit
analysis the uncertainties may very well rise to six orders of magnitude. This is in part because
we need to estimate marginal effects, both costs and benefits, and these marginal effects are
small differences in sometimes quite large basic numbers.
Some of the sources of uncertainty are:
a) Missing, incomplete, and inaccurate records of human exposure; for ambient
HAPs, these gaps are enormous, and we know even less about probable levels of
future exposures
b) A nearly complete absence of information about co-exposures
G-3
-------
c) Poor and incomplete records of human health outcomes, including the whole
range of biases recognized by epidemiologists (non-random dropouts, healthy
worker effects, recall biases, changing concepts and tools for diagnosis of disease,
and all the rest)
d) Unknown mechanisms of action, even when toxic endpoints have been identified,
and the consequent difficulties of predicting what will happen at biologically low
exposures
e) Targets that are moving over time, including the effects on future costs and
benefits of increasing life span and increasing time in which delayed toxicities
can appear, decreases in competing causes of illness and death, and the different
and changing medical implications of various illnesses and impairments
f) The over-simplifications imposed by the regulatory context: Human disease rarely
has one cause, and one cause does not always produce a single disease, so that we
try to satisfy unifactorial regulatory demands in a highly multifactorial situation.
g) Numerous extrapolations, which (for animal studies) include: animal to human,
high to low dose, one route of administration to another, constant lifetime to
intermittent and/or irregular exposure, and uniform and protected laboratory
environments to highly diverse and unprotected human situations
h) Large standard errors, sometimes from necessarily small samples, in many of the
critical inputs
i) Unpredictable and poorly understood environmental transport, including
meteorology, hydrology, and many other things
j) Possibly important synergies in biologic effects and a gross lack of understanding
about two critical issues: what is in specific mixes of air pollutants, and how
variable the mixes can be from one setting to another. Such information does not
seem to be available for any subset of the HAPs.
k) Allocating the "blame" for bad outcomes that are in fact the result of synergies
among exposures (e.g., when we cost out the extra risks and expenses from the
synergy between asbestos and tobacco smoke, which industry should get the
hypothetical bill?)
1) The non-linearity of many impairments; for example, one can lose fifty percent of
liver capacity and never know it, so that loss of the first ten percent may have a
value vastly different from the value of the last ten percent.
m) Unknown levels of regulatory compliance, and predictably incomplete efforts to
G-4
-------
monitor compliance for 189 HAPs
n) Big and fundamental questions about which things we are to value and about how
we are to attach specific values to those things
o) All the difficulties of trying to place dollar valuations on various kinds of
incommensurate health outcomes, including death, and the even greater
conceptual difficulties of placing values on markers of exposure or effect when
signs or symptoms of illness have not appeared
p) The sheer volume of 189 HAPs, which will impose great demands for scarce
technical talent as well as resources, and which will certainly force the adoption
of means to keep those demands within reachable limits
q) Estimates of the costs of compliance with a new regulation are notoriously prone
to error, usually but not always in the direction of gross over-estimation of what it
will cost polluters to clean up their act
More fundamentally, different people will value things in different ways. Whose
valuation counts? Will we take what people say the first time we ask them, or try to educate
them before they give us their values? Will we give special weight to the valuations of people
who have had the outcomes in question and understand them? What about substitution effects?
What about benefits forgone? Will we assign the same value to every death or every illness of
given severity? Willingness to pay is hardly a meaningful metric for someone who has barely
enough to get by anyway.
Are dollars even the right metric? There are questions about equity, there are differing
and non-linear utilities, there are major questions about what to exclude as externalities. There
are discount rates and intergenerational effects to account for, as well as distributional effects
more generally, and there may well be important transaction costs.
There are theoretical answers to all or nearly all of these points, but each application of
theory requires the use of inputs that are to some extent uncertain. I was quite serious about the
six orders of magnitude of uncertainty. We can be honest about that uncertainty, bury our heads
in the sand, and see the special interests take over the process, or we can abandon our scientific
and technical integrity, lie about the uncertainty, and ultimately lose our credibility and our
claim to special standing as scientists. Or, we can come to grips with it as a serious challenge,
deal with it directly and honestly, and do what we can to make sure that users of our analyses
understand the fundamental falsehood of any claims (including tacit claims) that some other
approach is better.
Some recommendations that may be constructive:
a) There is a need for substantial education about the art of the possible. We need to
G-5
-------
educate congress, the public, and the news media; in fact, every group or person who
encounters these issues. Perhaps most, we need to educate ourselves, so that we have
realistic expectations of what we can in fact accomplish.
b) There is an evident need for very much closer links between risk analysts and
cost-benefit analysts. Each person with a significant technical or managerial role on
either side of this divide should spend at least six months working in the other program.
(When I came to this point in the meeting itself, I heard snickers from the audience. As
well as I could tell, they came from a few persons on each side of the divide, though there
may be no better way to understand the real problems, and to learn how to help solve
those problems, than to wrestle with them yourself. I fear that some persons may not
recognize the career advantages of knowing both sides. Heavy pressure from higher
levels may be needed to implement this change.)
c) I recommend also that there be regular, weekly meetings on each project in which risk
analysis will be a significant element of a cost benefit analysis, to assure full
communication and understanding about what is needed, what can be provided, and how
to adjust for the inevitable gaps between these. Passing written reports back and forth
will not do the job.
d) There is a need for serious attention to the level of accuracy needed at each step of the
process of risk analysis / cost benefit analysis. Does it matter if we are off by 20%?
Two-fold? Ten-fold? One thousand-fold? Analysts and managers rarely address these
questions in any serious way (perhaps because they do not get beyond the correct
recognition that greater accuracy is always better, ceteris paribus), and yet they are
critically important here because of the need to make most effective use of limited
resources and to balance countless compromises and trade-offs.
e) We need an organized, almost assembly-line approach to risk assessment if we are to
deal with all 189 HAPs. The need for standardized procedures to deal with the HAPs
inevitably leads to the need for "bundling", though this too will require some
compromises, and general solutions may not always fit well. (Perhaps Procrustes had the
right idea.) Bundling according to health endpoint or cause of death might advance the
purposes of hazard identification. Bundling according to chemical species could advance
exposure estimation. Similarly, dose-response and sensitivity studies might be stronger if
we bundle by biologic mechanism or mode of action. Finally, regulatory considerations
may fit best with bundling by source categories. The last of these is apparently favored
by the economists because it is related directly to their task, but the value of the other
axes of classification in the risk assessment phases may outweigh the value of using
source categories in the cost-benefit analysis. It may be that we could somehow combine
these other axes with source categories to gain some of the advantages of both.
f) The axis of bundling needs more study than has been evident here, since it has
implications for the inputs to the cost-benefit analysis, which should follow the same
G-6
-------
pattern. If the bundling is by source categories, testing should also be by source
categories, with a focus on the study of the complex mix that comes from any one source,
rather than its components. I understand the technical and scientific objections to this,
but if we take those objections too seriously, they will simply undermine the whole
rationale for bundling by source categories. One cannot have it both ways. These
matters require serious study before EPA adopts an approach based on source categories
as a critical axis of classification.
In summary, we need radical solutions; tinkering with methods, approaches, and
mind-sets now on the shelf will not do the job. The technical and scientific issues of cost-benefit
analysis of regulatory control of the 189 HAPs are daunting, and any conceivable result will
inevitably carry an enormous margin of uncertainty. It is important that everyone involved in
this process, including all users of the results, understand that this true and that it is unavoidable.
This is not an indictment of either risk analysis or cost-benefit analysis, both of which are
critically important in collecting, analyzing, and interrupting what is or will be on the record. It
is certainly better to collect, organize, and interpret what we can than to simply give up and
proceed on blind faith that some step is or is not justified. However, we must not expect to
produce, or allow others to expect from us, a level of precision and certainty that the process is
unable to deliver.
G.3. Economics and Toxicology: Results of a Dialogue on the Prospects for Assessment of
Benefits from Regulation of Hazardous Air Pollutants. Comments of Dr. Trudy Cameron,
Department of Economics, University of California, Los Angeles, CA.
For some time now, there has been a degree of acrimony between some economists and
some toxicologists. This acrimony concerns the nature of the information being supplied by
toxicologists to economists for use in the congress!onally mandated task of valuing the
non-market benefits of environmental regulations. From the economist's point of view, the
problem can be characterized as "Why don't they just give us the information we need? Why are
they being so uncooperative?" From the toxicologist's point of view, the problem can be
characterized as "Why are they asking us for something that is impossible to provide? Why are
they being so unreasonable." The recent EPA workshop should reduce this acrimony, and help
us focus on the task at hand, by emphasizing the insight that we do not live in a perfect
(research) world.
a) What do economists need, in a perfect world, to calculate benefits? People's demand
for regulation of hazardous air pollutants is what economists call a "derived demand." People
are frequently viewed as demanding environmental regulation not for its own sake, but mainly
because the regulation may achieve a reduction in health risks.
Economists are accustomed to dealing with a wide range of derived demands. For
example, few people demand electricity for its own sake. Instead, we are willing to pay for
electricity because of the services that can be provided by electrical appliances. To understand
willingness to pay in a derived demand context, it is helpful to consider the chain of relationships
G-7
-------
that form the connection between willingness to pay and the underlying good. To continue the
electricity example, consider somebody who uses electricity to heat water. Formal modeling of
willingness to pay for a kWh of electricity for water heating will depend on the individual's value
of a gallon of hot water and on the efficiency of the water heater (namely, how much electricity
is required to produce a gallon of hot water). There are only two functions in this particular
chain: (1) how much hot water can be produced from a kWh of electricity, and (2) how much
utility (i.e. happiness, satisfaction) the individual gets from a gallon of hot water. Since utility is
not directly quantifiable, we measure it by how much money the individual is willing to give up
to get that increase in utility.
In the case of derived demand for hazardous air pollutant regulation, there are rather
more functions involved in the process of characterizing how a given environmental regulation
concerning hazardous air pollutants will ultimately affect individual utilities. (The final step,
monetization of regulatory benefits, requires the conversion of a given improvement in
individual utility levels into an equivalent income difference. This is the topic of a future
workshop, so we will leave this final benefit in terms of utility.)
First, we need to outline some of the constituent functions and identify the argument of
each function that is of key interest. Each of these functions will of course be multivariate and
subject to uncertainty.
(1)
Emj =
Emissions Em from firm j depend upon the firm's inputs and technology,
Ij
Ak = Ak(Emj,...) Ambient concentrations in region k depend upon emissions of all
contributing firms j
Exi = Exi(Ak,...) Exposure of individual i depends on ambient concentrations in their region
k
Di=Di(Exi,...)
Ci = Ci
Si = Si
Dose received by individual i depends on exposure
Cases of health effects for individual i depends upon the individual's dose
Symptoms of individual i depend upon whether they are a victim of health
effects
Utility levels are probably most directly influenced by symptoms
(compromised function, life expectancy, etc.)
Conceivably, each of these relationships could be studied independently. Since
controlled experimental data are rare (and may not reflect the true empirical derivatives in the
G-8
-------
field), a considerable amount of modeling will be necessary. By formally modeling all of the
factors that determining a particular outcome variable, we take other covariates into account and
minimize "omitted variables bias" in the estimated slopes. Specifically:
(2)
Emj = Em(Ij,...) - Emissions of a particular pollutant by a particular firm will depend on the
firm's inputs, production technology and abatement efforts, including MACT. It may depend
upon factor prices (including the prices of precursors of polluting emissions) and upon the prices
of the firm's outputs.
Ak = Ak(Emj,...) - Ambient concentrations in a particular region will depend upon the
emissions of all firms that contribute to these ambient levels of pollution and upon the "fate and
transport" (transfer coefficients) for each firm. Transfer coefficients depend upon weather
conditions, season, prevailing winds, the nature of the pollutant, and other factors.
Exi = Exi(Ak,...) - Exposure to the pollutant of individual i will depend upon the individual's
behavior and patterns of activity, including avoidance behaviors.
Di = Di(Exi,...) - Dose actually received will depend upon exposure and other factors.
Ci = Ci(Di,...) - Cases as a function of dose level will depend upon the individual's
socioeconomic status, current health status, age, gender, and other factors (such as the duration
of exposure or cumulative exposure).
Si = Si(Ci,...) - Symptoms, given that the individual develops a case of the health effect, will
depend on the individual's metabolism, access to treatment, age, current health status.
Ui = Ui(Si,...) - Utility may depend only upon the spectrum of symptoms the individual does or
does not experience. However, it is possible that utility will be affected even if this particular
individual is completely asymptomatic. Perhaps knowledge of exposure creates fear, or the
exposure and incidence of cases for other people affects utility. There can be both "use" and
"nonuse" demands for reliefer prevention of symptoms (including fatal cancers).
Studied in isolation, each of these functions presumably has some explicit approximate
mathematical form, the parameters of which must be determined from empirical studies. It is not
too far off the mark to suggest that each function in the list above is the province of a distinctly
different discipline.
Each of the above functions can be embedded into the next. People do not so much want
environmental regulation because they want firms to meet MACT standards. Rather, they want
environmental regulation because of what it means for themselves and their families. Thus, in a
simple model, utility levels depend directly on symptoms being experienced, and indirectly on
each of the contributing factors, all the way back to li (which we might interpret as abatement
G-9
-------
technology), if the EPA should choose to regulate at that level. Consider one possible
characterization of individual utility.
(3)
Ui = Ui ( Si (Ci (Di (Exi (Ak (Emj (Ij )))))))
The partial derivatives that we need to know in order to "construct" the effect on utility of
a change in abatement technology, Ij appear in the following expression:
/ A\
(4)
dT T
Ui
••u;
"Si
"Si
"Q
••Ci
••D;
• •
DI
••Exj
••Ex;
"Ak
"Ak
•Enij
••Enij
'I
*JT
If the EPA regulates individual firm emissions:
/c\
(5)
JT T
dU;
•Ui
"Si
• •s
^I
• •c
^1
"Ci
••D;
••DI
••Ex!
••Ex;
• *A
Ak
• *A
Ak
••En^
If the EPA regulates ambient concentrations:
(6)
dU;
0
"Ui
• •
"Si
"Si
• •
• «r
^i
"Ci
• •
•«i
••D:
• •
••Exj
••Ex;
• • • • i
"Ak
• • »/w • •
In the easiest possible world, each of these partial derivatives would be a nonstochastic
scalar. This implies extreme linearity in each respective function. But it is likely that most, if
not all, of the constituent functions outlined above are rather nonlinear. It is also possible that
the derivative of any one outcome with respect to one cause (say dQ/dD; = the effect of a change
in dose of benzene on the incidence of cancer) depends on the doses of other HAPs. There can
be interactions between stressors that influence the effect of a change in the level of any one
stressor.
The benzene, perchloroethylene, and manganese case studies focus primarily on the
• C;/» D; link in this very long chain of partial derivatives. But if we are trying to assess the
G-10
-------
welfare effects of regulation-induced changes in Ij (or Emj or Ak), these effects depend upon all
of the intervening partial derivatives, not just one of them.
But the utility function proposed above is just one of a number of possibilities. It
assumes that individuals derive utility from environmental regulation ONLY insofar as it affects
the symptoms they experience. It is an economist's job, however, to try to discern just what it is
that contributes to individuals' utility levels. It is possible that an individual's utility level
depends on emissions Emi not only indirectly via the symptoms they experience from the health
effects these emissions create, but also directly on emissions levels. Perhaps the utility function
looks more like this:
(7) Ui = Ui ( Si (Ci (Di (Exi (Ak (Emj (Ij )))))), Ci, Di, Exi, Ak , Emj , Ij)
If this is the way an individual's utility level is determined, then the individual may
derive an increase in utility from a decrease in emissions or ambient concentrations even if this
change in ambient concentrations produces absolutely no health effects (• C/'D; = 0)!
Individuals are allowed to derive utility from whatever they want. There is no justification for
considering only that utility derived directly from health symptoms.
b) Economists prefer to attempt to value the things that enter most directly into people's
utility functions. In the environmental regulation context, this usually means symptoms, such as
"days of eye irritation," or "days of moderate cough," or even "statistical lives lost." In the chain
of partial derivatives outlined above, we would prefer to explore how people are willing to trade
off dollars for changes in the level of symptoms. We might ask them directly what they would
be willing to pay, or we might ask them to choose among policies that involve different costs and
different levels of symptoms. The dollar metric is merely an intermediate device to capture how
much of other things they would be willing to give up in order to achieve a reduction in some set
of symptoms.
Focusing on the value of symptom changes reduces the dimensionality of the problem in
many cases. This is analogous to the way market researchers sometimes reduce the vast number
of different automobiles on the market to a much smaller number of attributes. Each auto can be
characterized not by its make and model and year, but by the bundle of attributes that it
represents (curb weight, acceleration, MPG, age, number of seats, etc.). The advantage of this
method is that if we study the market prices of autos as a function of the bundle of attributes each
represents, we can infer how market price depends on attributes. Then, if confronted by a new
make and model, with a specified set of attributes (preferably within the range of attributes
observed in the estimating sample) we can figure out approximately what people would be
willing to pay for the new vehicle.
To reduce the 188 HAPs to a smaller set of spanning "symptoms," these symptoms need
to be defined rather grossly, of course. Rather than trying to come up with distinct benefits
estimates for changes in the level of each of the 188 distinct HAPs, we would instead endeavor
to infer the incremental value of changes in each of a smaller set of symptoms. A one-unit
G-ll
-------
change in the concentration of a particular HAP would then need to be quantified in terms of
what that means for what matters to people: the suite of symptoms they are experiencing (e.g.
eye irritation, sore throat, cough, fatigue, for some compounds, and more serious endpoints for
other compounds). Policy changes with respect to HAPs that result in a bundle of symptom
changes will make it necessary to ascertain whether the effects of distinct symptoms on utility
are additive, subadditive, or superadditive.
We would like to stock the research "shelf with a set of estimates concerning people's
willingness to pay to avoid increments in each of the set of symptoms that are the usual suspects
in HAP assessments. We would also like to be able to say something about interactions among
sets of symptoms. If we can build up this inventory, then we have some hope of reconstructing
willingness to pay for a reduction in the amount of a particular HAP or set of HAPs that may not
have been studied explicitly. This process is known as "benefits transfer." We do not imagine
that it would be possible or sensible to conduct a separate economic analysis of willingness to
pay to reduce each one of the 188 HAPs. Instead, we would like to study only a few of them
explicitly (and bring in results on willingness to pay for symptom reduction from benefits
assessments for criterion pollutants or other applications).
Economists have become more and more confident over the years about how and when
they can come up with reasonable point and interval estimates for reductions in symptoms. But
we must rely entirely on other disciplines to convert a proposed environmental policy into
changes in a set of symptoms for some segment of society. What can we do if other disciplines
are unable to specify point and interval estimates for the rest of the partial derivatives? This is
only a dead end if we restrict utility to be derived only from indicators of health status.
But keep in mind that even if the most likely magnitude of the dose-response function
derivative is "zero" at current ambient concentrations, this does not mean that individuals cannot
experience a direct increase in utility simply from knowing, for example, that the ambient
concentration of a suspected HAP has been reduced.
Unlike toxicologists, who rightly expect to be able to identify a mechanism that explains
how an increase in the dose of some toxicant contributes to changes in an individual's health
status, economists expressly do not seek to figure out how a change in symptoms leads to an
effect on individual utility levels. "There is no accounting for tastes." We do not care about
WHY people derive utility from something, only that they do. The variant of the utility function
that emphasizes only symptoms is consistent with someone having a value system where actions
are based on "hypothetical imperatives" in the sense of Kant (i.e. IF you want to achieve this
end, then you must take this action). People who care directly about emissions or ambient levels
might be viewing HAP control as more of a "categorical imperative" in Kant's terminology (i.e.
you must take this action). Individuals can have any of a wide variety of philosophies or value
systems driving their individual utility functions. Economists just take these as they are and
concentrate on the task of how to aggregate these into a measure of collective social welfare that
respects these individual utilities.
G-12
-------
c) Why do economists want to know about central tendencies? Can't they figure out
what they know from the information about the 95th percentiles???
Since my logistic dose-response curves would be so untidy as to obscure the point, I will draw
linear dose-response relationships.
If the conditional distribution of responses at a particular dose had the same shape as the
conditional distribution at any other dose (only with a different expected value), then the change
in the conditional expectation of the response distribution for a given change in the dose would
be identical to the change in the 95th percentile of the response distribution for the same change
in the dose. It would not be necessary to know the central tendency.
Are the conditional distributions of responses identical at all dose levels? This is an empirical
question. It may be convenient to assume that they are, so that the needed slope (derivative) is
not only constant, but the same
response
(R)
95th percentile
of f(RlD)
E[F!|D]
5th percentile
of f(R|D)
dose [D]
E[R|D]
at all percentiles of the
conditional distribution.
However, without evidence to
support this rather heroic
assumption, it is more
reasonable to allow for a
conditional distribution of
responses that varies across
dose levels. Without drawing
the precise shapes of each of
these conditional distributions,
the lines that connect the 95th
percentiles, the expected
values, and the 5th percentiles
could just as easily look like
the second diagram. Here
(below), the assumption of
linearity in the percentiles is
retained, but even this may be
untenable.
5th percentile
of f(R|D)
dose(D)
-------
In this more general case, it is clear that the derivative implied by the relationship
between expected response and dose will be quite different from the derivative implied by the
relationship between the 95th percentile and dose.
Knowledge of the profile of the 95th percentiles of the conditional distribution of
response, given dose, is insufficient to determine the complete distribution. Even a normal
distribution requires two moments to identify its exact shape. Usually, these are the mean and
the variance. But you could equally well pin down a specific normal distribution with
information about its 95th and 5th percentiles (because the distribution is symmetric, this implies
the location of the mean, and the range between these percentiles implies the dispersion. The
95th percentile alone is insufficient to identify the distribution, even if it was known to be
normal (or lognormal.) To be able to identify a distribution based on a single percentile, we
need a one-parameter distribution. An example would be the exponential distribution. The
location and scale of an exponential distribution would be completely conveyed by the
information in its 95th percentile. But can we assert that the conditional distribution of
responses, given dose, is exponential. Not without evidence.
d) Why do economists need conditional distributions of response, given dose? If we are
going to do a rigorous assessment of the overall effects on benefits from environmental
regulation, we need to know both the central tendency and the dispersion in each of the partial
derivatives that must be multiplied to produce an estimate of utility gains from HAP regulation.
Since the product of random variables is unlikely to have an analytically tractable distribution,
simulation methods will typically be required to generate a distribution for benefits that allows
the reporting of not only best estimates, but also comprehensive error bars.
e) Utility levels under uncertainty. The problem with a utility function that is linear in an
uncertain symptom is that it does not allow for risk aversion with respect to that symptom. A
deviation from the expected symptoms that results in lesser symptoms may not have nearly the
G-14
-------
utility (U)
negative effect on utility of a
deviation in from expected
symptoms that results in worse
symptoms. The loss of utility
when the outcome differs from its
expected value may be very much
asymmetric on either side of the
expected value of the outcome.
There has not yet been much
empirical work on alternative
kinds of utility function for use
under uncertainty. Further
research on how to implement
benefit-cost analysis under
uncertainty in more general
contexts is clearly needed.
Relevant considerations include
how individual subjective
uncertainly is distinct from
scientific uncertainty or
disagreement with respect to the
ultimate symptom differences to be expected from a HAP regulation.
Risk averse with respect to symptoms?
E[U]< U(E[S])
symptoms (S)
The associated figure shows a utility function with respect to symptoms, U(S), for an
individual who is risk-averse with respect to symptoms. We need to know the distribution of
symptoms in order to figure out the distribution of possible utilities associated with these
symptoms. Note that a utility function characterized by risk aversion means that a symmetric
distribution for S is converted into a skewed distribution for U. The important result is that
expected utility E[U] is not simply equal to utility at the expected level of symptoms. It is likely
to be less. How much utility is lost due to uncertainty about the level of symptoms depends on
the shape of the distribution of S and on the shape of the utility function.
Economists have a pretty good theoretical framework for Benefit-Cost Analysis under
uncertainty. A succinct exposition of the theory in the case of objective uncertainty over a
binary outcome was presented by Graham (1981). For simple utility functions (e.g. those which
are linear in the levels of any factor that is uncertain) it is a straightforward matter to generalize
that model to the case of a continuum of possible outcomes (such as the uncertain response to a
dose of a HAP, or more directly, the uncertain symptoms from a dose of HAP). When utility is
depends upon the level of a symptom AND upon the squared deviation of the symptom from its
most likely value, the mathematics is simple and expected utility depends on the expected level
of the symptoms and on the variance in symptoms around this expected level.
G-15
-------
f) But we do not live in an ideal world. Unfortunately, the perfect ingredients for a full
benefit-cost analysis under uncertainty are not available. Some of the key derivatives in the
chain are unknown and probably unknowable at finite research cost. How, then, can an
economist begin to measure the benefits that people derive from regulation of hazardous air
pollutants? Rather than the "bottom-up" approach of building the benefits from derived demand
for regulation via demand for reduced health risks, we can consider a more "top-down"approach.
This approach accommodates both limited scientific information and individual consumers'
subjective assessments of the key partial derivatives (or even just the products of these partial
derivatives). I make the following suggestions for a research strategy based on what I am
currently attempting to do in the case of derived demand for climate change mitigation policies
Cameron (1998). Understanding and measuring the demand for expensive climate change
mitigation programs has much in common with the problem of understanding and measuring the
demand for HAP regulations. Both issues are characterized by uncertainty (incomplete or
ambiguous scientific information, controversy among experts, competing corporate and
advocacy group positions on the science, varying levels of expert "credibility," and varying
degrees of public interest cross-sectionally) and long latency periods.
The individual's subjective distributions on the magnitudes of the partial derivatives in
the chain are a product of the interaction between their experiential knowledge and the
information that they receive from outside sources. For example, if the individual knows
somebody who died of cancer who worked in a dry-cleaning establishment, this knowledge will
be combined with whatever expert information to which the individual has been exposed to yield
that individual's subjective conditional distributions of responses in the dose-response
relationship between perc and cancer.
Thus, the proper characterization of individuals' willingness to pay for HAP regulation
depends on their subjective assessments of all of the relevant partial derivatives (either
individually or in compounded form). An appropriate research strategy would elicit from each
individual their subjective assessments of what HAP regulation would be likely to achieve in
terms of health effects (and other effects). If he or she desires it, the individual should have
access to summaries of whatever expert information is available, including the fact that this
information is complete, if that is the case. The individual's updated subjective assessment
should then be established, and then his or her preferred choices among a set of policy
alternatives should be elicited. These policy alternatives should differ not only in terms of their
costs to the individual, but also in terms of the degree of protection provided against the
consequences the respondent expects without regulation. This method accommodates consumer
utility from HAP regulation no matter how it arises, whether through avoidance of measurable
health effects or simply through "existence" demand for a non-toxic environment. Economists
do not presume to question where utility enhancements come from, only whether they exist.
This method of assessing the benefits of HAP regulation is likely to rely upon stated
preference techniques (i.e. contingent valuation or its generalizations), since it is hard to imagine
actual referenda being held on alternative HAP policies. Fortunately, researchers are
G-16
-------
understanding more and more about the limitations and idiosyncracies of stated preference
research and how to minimize these.
It is important to keep in mind that economic research concerning the apparent value of
HAP regulations conditional on the public's understanding of the risks these compounds present
does not mean that we have to make policy based on widely held misperceptions about the true
risks of HAP concentrations. The idea is to model benefits as an explicit function of perceived
risks. Once this function is understood, it is then straightforward to replace the subjective risks
with scientifically supportable levels of objective risk and re-calculate the implied level of
benefits that would accrue to each individual if their beliefs were consistent with the science.
Note that the science can still involve uncertainty, and the values that individuals are "predicted"
to hold for HAP regulation should definitely be conditional on the extent of uncertainty (either
individual subjective uncertainty, or scientific uncertainty).
g) A smattering of philosophy. Benefit-Cost Analysis, as it is interpreted in most
contexts today, is understood to be based upon a utilitarian (Benthamite) social welfare function.
It is important to keep in mind that this particular social welfare function is not the only game in
town, although most economists in the U.S. are steeped in the utilitarian tradition because it
makes the benefit-cost problem tractable and it does have a number of desirable properties. (See
Kolstad (2000).)
Imposing an environmental regulation would be a "no-brainer" if it made nobody worse
off and at least somebody better off. Then nobody would be opposed to it and at least one
person would be in favor. Environmental regulations are controversial only if the beneficiaries
gain at someone else's expense. Some individual utilities will go up (or the regulation would not
be demanded) but other people's utilities will go down. This is true, for example, if the
beneficiaries of the regulation are not the same people that bear the costs.
Suppose there are some winners from regulation and some losers. If the winners win big
enough to be able to compensate the losers for their losses, then it would be possible to achieve
unanimity about the desirability of a regulation.
Just before WWII, Nicholas Kaldor and John Hicks proposed that the secondary
consideration of the distributional consequences of some proposed reallocation of resources can
be separated from the primary discussion of whether the net change in utility is positive overall.
Even if compensation does NOT take place, if the "benefits" in terms of gained utility for some
members of society exceed the "costs" in terms of lost utility for other members of society, then
there is an argument that the proposed reallocation is a good idea for society as a whole. It is a
second-order issue to then consider whether the distributional consequences of the reallocation
are sufficiently undesirable to preclude the reallocation on distributional grounds. This so-called
"compensation principle" says that if a proposed reallocation would create more gains than
losses, then it is socially desirable, even if no compensation occurs.
G-17
-------
A Social Welfare Function (SWF) takes the utility levels of individuals and combines
them in some fashion to yield a single-dimensioned scalar number for something called "social
welfare." Different value systems lead to different candidates for the SWF.
Utilitarian (Benthamite): W(ul,...,uN) = • ;•• fU;, • ;•• «0
Aggregate social welfare is a weighted sum of the utility of each individual in society. The
weights are positive, but need not be equal. We are usually most interested in welfare changes
from resource reallocations: • W/*X (where x is a determinant of individual utilities, ui ). This is
a way of denoting the "net benefits" from resource reallocation "•x." If these net benefits are
positive, the utilitarian (Kaldor-Hicks compensation principle) opinion would be that the
reallocation is a good idea. This is Benefit-Cost analysis, as conventionally practiced.
Egalitarian: W(ul,...,uN) = • fU; - • - f [u; - mini^)] * > 0.
In this SWF, society cares about the total amount of utility, • ;•% but also about the degree of
inequality. If everybody enjoyed identical utility, the term • ffu; min;(u ;)] would be zero. The
negative sign indicates that social welfare is decreased by departures between individual utilities
and the lowest individual utility level. The • "parameter dictates the weight on distributional
issues (i.e. inequality).
Rawlsian: W(u 1,... ,uN) = min;(u ;)
A society is only as well off as its least fortunate member.
i) A reminder: Arrow's Impossibility Theorem (AIT). This was one of the
discoveries for which Kenneth Arrow won the Nobel Memorial Prize in Economics in 1972
(shared with Sir John Hicks of Kaldor-Hicks fame). The result was part of his Ph.D. dissertation.
To paraphrase its useful result: There is no "ideal" way to combine individual preferences into a
social choice mechanism. Since then, people have spent a lot of time tweaking the conditions,
trying to figure out under just what modified conditions you CAN produce a nice tidy theory of
social decision-making.
But this impasse cannot stop the frequent need to make policy decisions regarding the
allocation of resources in some "best" fashion. In practice, the Kaldor-Hicks compensation
principle is often used, with ex post consideration of the severity of the distributional
consequences. The AIT just reminds us that this decision is not necessarily the only, or
necessarily the best, way of deciding about resource reallocations.
Some Criticisms of the Utilitarian Approach to Decisionmaking:
1) Do we each have a utility function that adequately and consistently represents our
preferences, especially over time. Tastes change. Preferences can be manipulated by
information campaigns (e.g. advertising, public service announcements, "education").
G-18
-------
2). Decisions according to the benefits and costs experienced by current members of
society are suspect if not all of the affected individuals are taken into consideration.
Specifically, future generations are sometimes not represented. Their tastes may differ
from current generations.
Should public policy be based on individual preferences (consumer sovereignty) at all, or
on what is morally right? Utilitarianism is a branch of teleological ethics. An alternative ethical
system is deontological ethics (deontology= science of duty; moral obligations). Immanuel Kant
(Metaphysics of Morals (1785) considered that actions can be judged by their intrinsic "rightness"
and not by the extent to which they serve to further one's goals or aspirations. Two types of
imperatives direct our proper behavior:
1) hypothetical imperatives (present the practical necessity of a possible action as a
means of achieving something else which one desires (or which one may possibly
desire; e.g. IF you want that, you must do this!)
2) categorical imperatives: present an action as of itself objectively necessary,
without regard to any other end. e.g. You must do this!
If you subscribe to a deontological ethics, you would not require a benefit cost analysis to
justify environmental regulation, you could justify it solely on the basis of an argument that
"humans have no business messing up the environment with HAPs." The major problem in using
deontological ethics as a basis for policy is that reasonable people can differ in terms of what they
judge to be "intrinsically right." These systems can work pretty well in a homogeneous society,
but the more heterogeneous the society, the more difficult it is to agree on what constitutes an
intrinsically right course of action with respect to policy. We rely on utilitarianism because it
makes policy evaluations easier to effectuate. (See Hackett (1998).)
(Some, and perhaps much, of the animosity towards economic welfare analysis stems from
misunderstandings about what it IS. There is a vitally important distinction between using
Benefit-Cost Analysis to MAKE environmental decisions, versus using Benefit-Cost Analysis to
INFORM environmental decisions.)
j) Summary. As an economist who has struggled for quite some time with a number of
different problems in valuation of non-market benefits of environmental goods, I really appreciate
having had an opportunity to get confirmation from toxicology experts that there are fundamental
(and probably unresolvable) gaps in our knowledge about the measurable health consequences of
hazardous air pollutants.
This insight means that an unambiguous, objectively calculated., bottom-up measure of the
social benefits from HAP regulation is unlikely to be forthcoming. But this certainly does not
mean that these benefits are zero. People may be willing to allocate society's resources (including
their own) to control of hazardous air pollutants simply because there is a possibility that they
could have health effects, even if these have not yet been detected (or are unlikely to be detected
G-19
-------
any any feasible cost of scientific research). Deciding upon an alternative valuation strategy is
the next step.
It seems likely that stated preference research will be the most fruitful way to proceed.
Experts in economics and cognitive psychology will have to address the issue of elicitation of
subjective probability distributions for health risks (and other risks) that may emanate from
ambient levels of hazardous air pollutants. We will need to study how public perceptions of risk
can be modified by new scientific results, or by "education" (propaganda) campaigns. People's
values for environmental programs depend upon what they think the programs are buying them.
Their beliefs may or may not be consistent with current scientific understanding. But this does
not preclude a strategy of first elicitation, and then simulation, to ascertain what would have been
the public's value of a specified program had they fully accepted the best current (and possibly
incomplete) scientific knowledge.
References
Cameron, T. A. "Credibility of Information Sources and the Formation of Individuals'
Option Prices for Climate Change Mitigation." . Department of Economics, UCLA, June
1998.
Graham, D. A. "Cost-Benefit Analysis under Uncertainty." American Economic Review 71,
(1981): 715-725.
Hackett, S. C. Environmental and Natural Resources Economics. Armonk, New York: M.E.
Sharpe, 1998.
Kolstad, C. D. Environmental Economics. New York: Oxford University Press, 2000.
G.4. Comments of Ms. Laurie Chestnut, Stratus Consulting, Boulder CO
a) Comments relevent to workshop agenda questions 1 and 2:
Question 1) Proposed approaches for hazard assessments for selected HAPs that would
facilitate benefit assessments for those chemicals.
Question 2) Expert discussants' views on whether it is possible to produce a methodology
for developing central tendencies and distributions in hazard assessment for HAPs for use
in benefits analyses and how that might best be done\
It seems pretty clear from the information presented at the workshop that available
toxicology and epidemiology evidence is insufficient to provide dose-response relationships over
relevant ranges of exposures for most of the air toxics for which EPA is required to make
regulatory decisions. Data based on human studies tend to be at much higher occupational
exposures, requiring heroic assumptions when extrapolating to typical community exposures.
Data based on animal studies are suggestive of whether or not, and by what mechanism, a
chemical may be harmful, but are very difficult to interpret when it comes to quantitative dose
G-20
-------
response for humans. Al McGartland gave references to some published papers that look at
drawing dose response from information used to determine reference doses for some chemicals.
These should be reviewed by the toxicologists to see if the methods are sound and how widely
they could be applied. It appears, however, that the information necessary to do a comprehensive
quantitative benefits assessment of air toxics reductions, such as is the goal for the 812
assessment, is not going to be available any time soon.
Some questions arose at the workshop regarding if and how economics would use
information on variations in dose response for different population groups. Heterogeneity in dose
response is a part of the dose response information needed for a benefits assessment. For each
dose response function, we need to know to what population it applies. It does not need to be for
the general population. There may be different dose response for different age groups, for
example. If certain groups are affected significantly differently than others, this should be
included in the presentation of the results of the assessment. Similarly, it is important to know
whether, for example, there are 100 people facing a change in risk of cancer of 1 in 100 or a
million people facing a change in risk of cancer of 1 in a million. The bottom line in both cases
may be one expected cancer case, but the risk situation is quite different.
b) Comments relevent to workshop agenda questions 3 and 4
Question 3) How best to identify limitations and uncertainties in both risk assessment
methods and economic models.
Question 4) Suggestions and prioroities for a research agenda to address identified gaps
in available data and methods needed to conduct HAPs-related benefit analyses
In the regulatory decision making context, there is strong motivation to make reasonable
use of the available information, even if all the important questions cannot be answered. In this
context, ranges of estimates of risk changes, upper or lower bounds on risks, and other
information short of a best estimate of dose response, can be utilized to help inform regulatory
decisions. Highly variable, uncertain, and inconsistent information, however, is difficult to use in
a quantitative benefits assessment unless there is some way to assess how likely it is that each
result is accurate. For example, if some studies have found that a given chemical is a carcinogen,
and others obtain negative results for the same chemical, we can say that the expected reduction
in cancer cases ranges from zero to the amount suggested by the studies that have found an
association. Such inconsistencies in results can result in such a large range in benefits estimates
that the assessment is not very useful (e.g., saying that the benefits are somewhere between zero
and $10 billion is not usually very useful information for decision makers). However, if some
assessment of the likelihood that each of the available results is accurate can be made, then the
assessment can be made more informative. Continuing with the same example, perhaps there may
be a basis for determining that studies finding a carcinogenic effect are more likely to be correct
than studies that have not found carcinogenic effect. Thus, it may be possible to say that there is a
75% chance that the benefits are $10 billion and a 25% chance that the benefits are zero. This
G-21
-------
does not mean that the only benefits number that should be presented is the $7.5 billion expected
value. The range and the probabilities should be presented.
Probability distributions (or some other form of assessment of the likelihood that various
results are correct) are also very useful when combining the many steps involved in a benefits
assessment. When many ranges of values are all multiplied together, the high and low values can
end up being very far apart. A very wide range in the result is not very useful information,
especially when the chances that all the low values or all the high values are correct is small. For
example, if we take the highest value at each step and multiply them all together, we get a very
high value result that is very unlikely to be correct. The ideal information needed to determine
probability distributions on results used as inputs for benefits assessment is seldom available, and
some professional judgment is inevitable. Sometimes it is better to say we don't know than to rely
on pure guesswork. Where that line is is also a matter of judgment. Sometimes the best we can do
is some simple sensitivity analyses on key assumptions in the assessment. For example, if this
chemical is not a carcinogen, the answer is X; if it is a carcinogen the studies showing an effect
suggest the answer is Y.
G.5. Comments of Dr. A. Myrick Freeman, Department of Economics, Bowdoin College,
Brunswick, ME
The question posed for us was "whether it is possible to produce a methodology for
developing central tendencies and distributions in hazard assessments for HAPs for use in
benefits analyses..." I think that in short the answer is "No," because of both the variety of
endpoints of concern (cancer and a variety of non-cancer endpoints) and the variety of sources of
data (human epidemiology, long term animal feeding studies for cancer, and other human and
animal data for those non-cancer endpoints that have been studied). Rather, different approaches
will be necessary for different endpoints and types of data.
For the 20 or so known human carcinogens, best estimates of dose-response (D-R)
functions and uncertainty bounds can be obtained using meta-analyses or applying Monte Carlo
methods to the available human data. This leaves us with the questions of extrapolation to low
doses and the possible existence of thresholds; but these are well known questions.
For the other possible carcinogens, one possible answer is to try to obtain better human
data on D-R relationships; but there are the well known problems of estimating human exposures
and the typical low power of epidemiology studies. Lacking additional human data, the only
recourse is to the animal data. This would mean obtaining maximum likelihood estimates of
cancer slope factors and their confidence intervals from the original animal test data. Again there
is the question of high to low dose extrapolation as well as the animal to human concordance and
extrapolation questions. But I don't see any other way to proceed. I second Lester Lave's
suggest concerning the comparison of animal and human data for those substances for which both
kinds of data exist.
G-22
-------
Resorting to the animal data for all of the possible HAPs carcinogens that lack human data
is probably impractical. It will be important to perform some kind of screening and prioritizing
exercise based on, e.g.,, indications of large volumes of emissions or where human exposures are
thought to exceed RfCs. Screening the noncancer HAPs can proceed in a similar fashion. For
example, where present human exposure is less than the RfC, the benefits of further reductions in
emissions are likely to be zero.
The preceding advice has been based on the assumption that benefits are being defined in
the standard way as willingness to pay (WTP) for reductions in the numbers of cases of various
types of disease, eg., cancer, obstructive lung disease, etc. An alternative approach to benefit
estimation (which might be pursued in parallel with the standard approach) is to investigate
different ways to define the commodity to be valued that reflect the ways in which individuals
actually think about reducing the risks of environmental disease. For example, economists have
used the averting behavior model to analyze data on bottled water purchases as a way of valuing
reductions in the risk of waterborne disease. My conjecture is that many individuals view their
purchases of bottled water, not as a means of reducing the risk of specific diseases, but as a means
of increasing safety more broadly conceived. If there is anything to this conjecture, then
individuals might view a broad or comprehensive policy of controlling emissions of HAPs as
producing safety or "peace of mind" rather than as yielding reductions in the risks of specific
diseases. And if that is the case, then best estimates of reductions in specific risks are not
required for benefit estimation. What would be required, however, is a better understanding of
the relationship between controls on emissions of HAPs and individuals' perceptions of safety or
"peace of mind."
G.6. Comments of Dr. Dennis Paustenbach, Exponent, Menlo Park, CA
Response to workshop agenda Question :2: expert discussants' views on whether it is
possible to produce a methodology for developing central tendencies and distributions in hazard
assessments for HAPs for use in benefits analyses and how that might best be done.
The answer to this question depends on the level of certaintythat one needs to satsify those
who will perform and use benefit/cost analyses. Certainly, we can produce cancer risk estimates
for animal and human carcinogens using various models. We have used these models in the past
in an attempt to rank (in a relative way) the carcinogenic potency of chemicals. However, like
most risk assessors, I don't believe that they can accurately predict the actual human response
following low exposure to these substances.
I acknowledge that a couple published analyses have suggested that for the genotoxic
chemicals, where we have exposure and epidemiology data, there has been a reasonable level of
agreement between the number of cases predicted vs the number observed. However, the doses
in these studies were those observed in workplaces of nearly 30-40 years ago and these are not
even within 100 fold of the doses to which the public is exposed today. In addition, for the cancer
estimates from the basic models to be accurate, they should be attached to the "internal" doses
estimated by physiologically-based pharmacokinetic (PB-PK) models. This has been best
G-23
-------
exemplied in a paper by Reitz et al (1996) which studied vinyl chloride. In short, for virtually all
of the animal carcinogens, we are not in a position to use low-dose models to predict the actual
cancer risk in humans. We can, for purposes of benefit/cost analyses, use them as a relative index
of hazard which could then be "modified" if one wanted to consider other
biological factors (like genotoxicity).
Thus, the answer to the question "is it possible to produce a methodology for developing
central tendencies and distributions" is yes for the carcinogens. This could be done and, given the
abovementioned caveats, it might serve the purposes of the exercise.
With respect to the non-carcinogens, a different approach would be needed as there is
assumed to be no risk associated with doses below certain values. One possible approach is to
calculate "margins of safety" (MOS) for the non-carcinogens. In this approach, one would take
the EPA Reference Dose (RfD) or the Reference Concentration (RfC) and determine the cost
associated with achieving doses below these "safe" concentrations. As was mentioned by
Dr. Lave, often the public simply wants "to feel safe'....no matter that scientists may give
assurances that current conditions pose no significant risk. Assume, for example, background
concentrations of formaldehyde in some cities can reach, under certain conditions, about 50 ppt.
An airborne concentration generally thought to pose no risk of even transient eye irritation is
about 250 ppt. Perhaps, the public wants then never to have concentrations get above 25 ppt.
The economists could then provide information to Congress which indicates that the cost of
providing a MOS of 10 for formaldehyde is $100,000,000. A decisionmaker can then compare
this cost to achieve an MOS of 10 for formaldehyde to the cost of achieving an MOS of 10 for
another non-carcinogenic chemical, for example manganese, and could then weigh the relative
importance of the potential adverse effects. In the case of formaldehye, the threat is transient eye
irritation, while for manganese, it could be premature aging of the nervous system. If the costs
were similar to achieve an equivalent MOS, then it is likely that the decisionmaker would choose
to regulate manganese to rather than formaldehyde given the major differences in toxicity. This
approach doesn't equate to a "life save" but should be a perfectly useful metric for both the
economist and regulator.
As shown in the above example, it is necessary that the process of benefit-cost analyses
for the HAPs be tackled according to the adverse effect of concern. For example, probably about
33% of the chemicals are listed due to their carcinogenicity, 33% are listed because they are
systemic (non-carcinogenic) toxicants, while about 33% are irritants. Each might require a
slightly different approach. Nonetheless, each has a dose-response curve (or one could be built)
and a distribution around the various points on the dose-response curve could be built (for both
the carcinogens and non-carcinogens). Over time, the process would almost certainly be
modified as more is learned about its strengths and weaknesses.
G.7. Identifying Limitations and Uncertainties in Risk Assessment and Benefit
Measurement Methods. Comments of Dr. V. Kerry Smith, Center for Environmental and
Resource Economics Policy ^Department of Agricultural and Resource Economics, North
Carolina State University, Raleigh, NC
G-24
-------
This paper is intended as a summary of some of the issues raised in the SAB/EPA
Workshop on Benefit Analysis for Policies that reduce exposures to Hazardous Air Pollution
(HAP). My specific focus is on the research needed to address limitations and uncertainties in
risk assessment (RA) and benefit measurement (BM) methods. The paper is developed in five
short sections after this introduction. The first describes a few features of EPA's conventional
practices linking RA and BM. The second explains how the issues posed by HAP are different.
In the third I define consequential uncertainty and how this concept may offer an approach to help
conceptualizing some of the research needed in this area. Section four discusses some activities
in risk assessment and benefit measurement where there appear to be opportunities to co-ordinate
research. The last discusses the merits of a parallel research strategy.
a) Conventional Practice. The logic linking RA and BM in the evaluation of criteria air
pollutants has been unidirectional. For each air pollutant RA estimates a set of health outcomes
such as changes in the probability of premature deaths for the general population, or for specific
sub-groups (e.g., elderly, asthmatics, etc.), with a well-defined change in each group's exposure to
a specific air pollutant (see Chapter 5 and Appendix D of U.S. Environmental from reductions in
the ambient concentrations of a pollutant rely on developing benefit Protection Agency [1997] as
examples). Monetary values for the health benefits arising measures for unit changes in each
health outcome or risk change. For example, if a reduction in the ambient concentration of
particulate matter reduces the expected number of days with respiratory illness, then the
monetization of the value of this change generally seeks a measure of the benefits of avoiding a
day of respiratory illness. These unit benefit measures may not be associated with the specific
source of the health effect. As a result, this approach assumes that a day of respiratory illness is
essentially the same regardless of what caused it and therefore would have the same economic
consequences for affected individuals.
The unidirectional logic is important. It establishes the equivalent of a chain of functions
linking emissions to the ultimate health outcome, and, from the economists' perspective, the
change in well-being experienced by each individual (see Trudy Cameron's paper for a more
detailed elaboration of this logic).
Any description of the limitations and uncertainties in existing practice usually begins by
distinguishing uncertainties that arise because many of the components of the chain of
relationships linking emissions of pollutants to changes in well-being are stochastic processes.
Reductions in the mortality effects of a pollutant, for example, represent changes in the
probability of premature death for specific groups of individuals. Because the outcome is a
probability change, the framework used in this case to measure the health effect acknowledges
that there are many external and internal influences on an individual's probability of dying in a
given year. Exposures to specific air pollutants are only one class of these influences.
The specific models used to fill in the logical chain may incorporate the inherent stochastic nature
of the process or they may deal with expected values, implicitly assuming the relationship can be
treated as exact except for measurement errors. This distinction determines whether the analysis
acknowledges what one might term "structural uncertainty".
G-25
-------
A second important source of uncertainty will be labeled "estimation uncertainty". This
source arises because one must estimate each of the functions comprising the logical chain of
linkages from the emission to the change in people's well-being. Measurement error, omitted
variables, incorrect functional form, etc. can introduce error and therefore uncertainty in the
actual policy description used to implement the chain of linkages from emission to benefit. While
in practice we cannot separate these two sources of uncertainty, at a conceptual level it is
generally important to distinguish them.
An important lesson that has been derived from two decades of research developing RA
and BM methods for air quality regulations is that health outcomes identified in the RA process
must be capable of being associated with "things" people can, in other contexts, choose. For
example, small changes in mortality risk can be selected by people in a number of types of
behavior. The most common source of behavioral information used in benefit measurement is in
job choices. Increased risks on the job generally lead to compensating differentials in wage rates
(see Viscusi 1993]). Changes in a person's physical condition, e.g., difficulty in breathing,
increase in hypertension, do not generally associate with a choice we can observe people making
in another context. In these circumstances a new link in the chain must be introduced. That is,
the analyst must associate a change in lung functioning with greater incidence of sick days or an
increased change of more serious respiratory illness and then evaluate whether there are actual or
stated choices that can be related to these outcomes.
This need to augment the chain of linkages is important because to the extent there is
discretion in either the intermediate measure of the health characteristic sought or the observable
health outcome that results, risk assessment decisions should be made in ways that facilitate
making connection to an economic choice. Otherwise, one simply adds to the uncertainty in the
measurement process.
Co-ordination between the design of risk assessment and benefit measurement methods is
important for another reason as well. Benefit measurement methods rely on observing (or
offering potential opportunities) for people to make choices that reveal how they would tradeoff
some outcome that can be related to the pollutant of interest and another commodity that can be
expressed in monetary terms. The unidirectional logic of most benefit information for air
pollution policy assumes that these choices can be observed in another context and transferred to
any evaluation of policies involving air pollutants. Expressed in terms of the chain of linkages
discussed by Cameron, and illustrated in simple terms in Figure 1, this process allows
monetization of the changes in outcomes at step 5 in the process as approximate measures for the
value of their consequences for well being identified as step 6 in this logic.
Figure IrChain of Linkages for Environmental Policy Evaluations
for Air Pollutants
Characterization of: Step Number:
G-26
-------
Emission Rates for Pollutants
by Source
• •
Spatial Diffusion, Atmospheric
Interactions, and Other Influences
to Ambient Concentration of
Pollutants
• •
Exposure Patterns for Receptors
at Each Location
• •
Physical Response to Exposure
to Pollutants
Health Outcome Resulting from 5
Physical Response
• •
Effect on Individual Well-being 6
Unfortunately the unidirectional flow is itself an approximation and the very behavioral
choice relied upon to make benefit measures for health outcomes can affect the reliability of the
one way flow of causation. That is, revealed preference arguments assume to the extent people
recognize the effects from steps 2 to 6, they will react to them and try to adapt.1 The hedonic
model assumes they consider site specific amenities in residential location choices. Models of
averting and mitigation behavior suggest that other, less costly, responses will also be made.
These can include spending less time outside during high pollution times or purchasing central air
conditioning, etc. These choices imply the equivalent of feedback loops between steps 6 and 3.
The physical responses to some types of pollution may be more easily recognized by people. As
a result, in this case the mitigating responses may be more likely and the feedback important.
This dimension also will be important to the RA/BM connections with hazardous air pollutants.
b) Three case studies at the SAB/EPA Workshop illustrated the range of possibilities in
information likely to be available about hazardous air pollutants. Benzene offered a case with
substantial information (compared to other HAP's) on the relationship between exposure (at high
doses) and human health outcomes. While there was little basis for evaluating the extrapolation
to low dose levels, there also was no basis for arguing modifications to conventional practices.
averting or mitigating behaviors imply continuous adjustment is feasible. To the
extent the set of available choice alternatives is finite, then, as Bockstael and McConnell (1999)
suggest, people will select the best alternative in the set. This does not imply the choice will be
at a point where the marginal value of the amenity underlying the choice equals the marginal
cost of adjusting to obtain it.
G-27
-------
By contrast, perclorethylene illustrated a situation where despite substantial data for humans and
animals, the strategies for measuring impacts were incomplete due to data limitations. Thus the
evidence available could only be regarded as "weak signals" of potentially more important
effects. Nonetheless, over the short term there seemed to be little basis for improving the
information available and some perception that judgments connecting exposures to health
outcomes would need to rely on encoding experts' judgments rather than more formal empirical
tests.
The last case, manganese, had most of its information concentrated in describing aspects
of step 5 in the linkage chain given in Figure 1. Completing the linkage required judgments to
connect physical responses to more conventional health outcomes and to evaluate how
economically meaningful choices could be connected to non-traditional outcomes.
An overall implication of the background presentations by both Farland and Lave was that
policymakers are unlikely to have the type or the level of detail in information available for the
HAPs to be evaluated. Moreover, decisions about regulating them will be made before the
research required to implement the conventional RA/BM logic would be available.
Three issues emerged implicitly or explicitly in the resulting discussion: screening rules across
HAPs were needed to identify the most likely candidates for regulation; the methods used to
conduct RA and BM need to consider the treatment of uncertainties in the component elements
of policy analyses evaluating practices based on they become "consequential" for decisions; and
future research programs should be structured to include parallel research activities, creating
pathways for cross-checking findings. The first of these is discussed in comments prepared by
Chestnut and Locke; the second is discussed in the next section; and the last is considered in
Cameron's comments and in the last section of this paper.
c) Consequential Uncertainty. Regulatory policy based on risk assessment by definition
recognizes that a policy is intended to change the stochastic environment in which lay people
must make their decisions.2 Thus, a policy evaluation of a regulation in this context describes
uncertainties people face with and without a regulation. Estimation and structural uncertainties
are both reflected in such descriptions.
The treatment of uncertainty becomes consequential when it alters one or more aspects of
the criteria influencing policy decisions in a way that would alter a choice. In short, analysis
decisions about how to reflect the sources of uncertainty for HAP don't matter if the policy
choices would not be affected. Thus, improvements in Monte Carlo simulation or other analytical
details that would improve variance estimates for health effects are not consequential, if policy
choices are always based on central tendencies and these conclusions would not change with the
refinement.
2I will not attempt to discuss in this short paper subtle distinctions sometimes considered
in economics (and psychology) between the terms risk and uncertainty.
G-28
-------
Figure 2 illustrates how the decisions made about coordinating risk assessment and benefit
measurement influence whether the treatment of uncertainty is consequential to policy decisions.
Five issues contribute to this judgment: the baseline distribution of exposures people receive in
the absence of action; the ambient concentration judged to be associated with health outcomes
that "count" for regulatory purposes; the estimation uncertainty in describing that ambient
concentration; and the risk factor applied based on EPA's propensity to include a margin of safety
(due presumably to a composite of concern about structural and estimation uncertainty) complete
the elements usually associated with risk assessment. Measures of the economic importance (e.g.,
unit benefit measures) complete the interacting factors that, together with cost, determine whether
regulatory decisions will be consequential to policy choices.
Using Figure 2, on the horizontal axis is plotted the proportion of people experiencing at
least the amount of a pollutant measured on the vertical axis.3 The point AO designates
background, so 100% of the population experiences at least this level. For each HAP we would
expect a different relationship to describe these exposure profiles. Different population groups
could also be represented as having different patterns of exposure. Curves AT and AJ illustrate
two alternatives. They are presented as straight lines for graphical convenience only.
3This distribution could refer to the whole population or a specialized group such as
children or the elderly.
G-29
-------
Figure 2: Illustration of Effect of Treatment of Components of RA and BM for
AitThiVtrf "Consequential" Uncertainty
f'ancenlratioii
of Pollutant
0
E
•B
°l
_>*.
-r
0
TT
in
1.0
People
Policy choices to recognize a health outcome (whether traditional or non-traditional) and
link it to an ambient concentration implicitly solve "backwards" or invert the functions one
associates with the connections implied by the ambient concentration • "exposure level; the
exposure level • •physical response; and physical response • •health outcome linkages (steps 2,
G-30
-------
3, 4 and 5 in Figure 1). This defines point E and the estimation uncertainty around it in Figure 2.
Estimation uncertainty is represented in the figure as the range for E + • *to E - • r Policy
analyses requiring a margin of safety also then impose further risk factors such as a reduction of
•fin addition to the reduction of* in the concentration regarded as "safe". This specification for
a desired (after regulation) ambient concentration determines whether the earlier decisions on
how to measure the central tendency and uncertainties in health outcomes will be consequential.
That is, consider the cases • i, • 2 and • 3 with the two different exposure profiles. My argument
says that decisions about whether health outcome is serious enough to count (for regulatory
purposes) and about the functions linking ambient concentrations to that outcome implicitly
determine the starting point on the vertical axis — E. Change anyone of them and we move E up
or down the vertical axis. Likewise the treatment of estimation uncertainty (q) and judgments
about risk factors (• ) and judgments about risk factors (i.e., whether • ?, • ? or • ?) will for a given
distribution of people (AT versus AJ) yield different proportions of the population that are
affected -- 0 I, 0 II, 0 III or 0 IV.
Recognizing these choices as potentially consequential implies we should be evaluating
them by asking whether a count of affected people, or a cost per person, or a net aggregate benefit
would be important to the regulatory decision. Each is affected in principle by decisions about
what counts (and thus the E) and how uncertainty is incorporated (i.e., the • "and •$. An
important element in the discussion at the workshop was that consideration should be given to the
implications of making these decisions differently depending on whether the outcomes lead to
dramatically different net benefits.
d) Research Complementarities Between Risk Assessment and Benefits Measurement.
Table 1 summarizes a few general research tasks that were discussed at the workshop for
hazardous air pollutants and the potential for complementarity or interactions that stem from how
the research is designed. Three of the four tasks generally defined as involving toxicology or
epidemology would benefit from complementary economic research related to specific
modifications in the elements of policy analyses that could be consequential to decisions.
G-31
-------
Table 1 Complementarities in RA4 and BM5 Research
For Policy Associated with
Hazardous Air Pollutants
TASK
Evaluating the nature of the
physical effect on people
Evaluating the nature of the health
outcome affecting people
Measuring distribution of ambient
concentrations and number of
people experiencing them
Estimating the overall
consequences of Baseline (no
regulation) and Regulated
Alternatives
Evaluating importance of
layperson's "worry quotient" or
policy as insurance
RA
X
X
X
X
BM
-
Eliciting lay person's preferences
for different health endpoints
Evaluating the prospects for private
action to mitigate or reduce
exposure received
Defining and measuring economic
choices for identified health
outcomes so tradeoffs could be
used to estimate benefits from
policy
Measure economic benefits as
value of a regulatory program or
policy
e) Parallel Research. Given the limited information, the number of hazardous air
pollutants to be evaluated, and the time and resources available to develop such evaluations
Lave's paper and presentation at the workshop suggested a different strategy for measuring the
benefits due to HAP regulations.. He proposed that we consider measuring the economic value of
the "policy" as an object of choice rather than the reductions in specified health conditions
attributed to reduced ambient concentrations of individual hazardous air pollutants.6 As noted in
Table 1, this approach was discussed as a method for assessing the importance of HAP policy as
providing a type of insurance for lay persons' "worries" about serious health outcomes that arise
as surprises from exposures to these pollutants.
4RA refers to research in field related to risk assessment. The primary areas considered
in the workshop discussion were toxicology and epidemiology. An "X" means research is
clearly needed.
5BM refers to benefit measurement. The elements in the table illustrate research tasks
that would be complementary to the task to be addressed in risk assessment
6This approach parallels innovations in the use of contingent valuation for the damage assessments
associated with natural resource damage cases. To my knowledge it was developed for the Exxon Valdez case by
Carson etal. (1992).
G-32
-------
The workshop discussions suggested to me an opportunity for economic research to
proceed along three inter-related lines that would complement each other and provide
opportunities for cross validation of benefit estimates for policy. Unfortunately, there was not
sufficient time to discuss this strategy at the workshop so it is not reflected in the summary of this
discussion. As a result, this paper ends with a discussion of each of the three lines of activity.
The first entails a proposal implicit in Chestnut and Freeman's comments and in the
summary given in Table 1. This encompasses evaluation of whether we can measure people's
preferences over non-traditional health outcomes. This task involves not only rating outcomes
likely to be associated with hazardous air pollutants, but also investigating the feasibility of using
existing revealed preference information and stated preference surveys to recover benefit
measures for these types of non-traditional choices. Agee and Crocker's (1994) study of parents'
willingness to pay for reducing blood lead levels for their children is an example of the type of
revealed preference analysis envisioned in this proposal. Recent applications of conjoint methods
(Johnson and Desvousges [1997]) suggest it may be feasible to offer non-traditional health
outcomes within this format.
The second line of research involves the focus group, survey development and pilot
studies required to evaluate whether Lave's proposals to evaluate the control policy as an object of
choice can actually be presented as a plausible choice alternative. It is not clear that it can, but
following the protocols used in developing modern CV surveys (especially those for large scale
damage assessments) it should be possible to resolve this issue without conducting a full
contingent valuation study.7
The last line of research was presented briefly in my comments, but time did not permit it
to be discussed in specific terms at the workshop. It argues for conducting the first two together
because, in principle, we should be able to establish a relationship between measures of the
economic value of the policy and the benefits for reducing specific health outcomes. The former
is a type of ex ante option price with some private mitigation (as a type of private insurance), and
the latter is a set of ex post values for the outcomes being avoided. The early logic developed by
Anderson [1979] should, with modification using Graham's [1991] extension to the definition of
option price, allow one to relate the two measures under specific conditions. This implies the
economic value of the policy (as an object of choice) could be compared to the sum of the
economic values of the avoided health outcomes. This would serve to unify the analysis, provide
a check for both the CV (i.e. contingent valuation) estimate and a gauge of the potential for
omission in cases where only a subset of the physical effects can be measured.
7 A contingent valuation study rather than conjoint is proposed here because the object of choice is a policy
and not a specific set of health outcomes with varying attributes. The full attributes of the results of the policy could
not be described. If they could, then more conventional methods would be used. Indeed an identification of the
uncertainty in the nature of the avoided health effects would likely be included as part of the description of the
policy.
G-33
-------
References
Agee, MarkD. and Thomas D. Crocker, 1994, "Parental and Social Valuations of Child
Health Information", Journal of Public Economics, Vol 55 (September): 89-106.
Anderson, James E., 1979, "On the Measurement of Welfare Cost Under Uncertainty", Southern
Economic Journal, Vol 45 (December): 1160-1171.
Bockstael, Nancy E. and Kenneth E. McConnell, 1999, "The Behavioral Basis of Non-Market
Valuation" in Joseph A. Herriges and Catherine L. Kling, editors, Valuing Recreation and the
Environment (Cheltenham, U.K.: Edward Elgar).
Carson, Richard T., Robert C. Mitchell, W. Michael Hanemann, Raymond J. Kopp, Stanley
Presser and Paul A. Rudd, 1992, "A Contingent Valuation Study of Lost Passive Use Values
Resulting from the Exxon Valdez Oil Spill", Report to Attorney General of State of Alaska,
NRDA, Inc., November.
Graham, Daniel A., 1992, "Public Expenditures Under Uncertainty: The Net Benefits Criteria",
American Economic Review, Vol 82 (December): 882-946.
Johnson, F. Reed and William H. Desvousges, 1997, "Estimating Stated Preferences with
Rated-Pair Data: Environmental, Health, and Employment Effects of Energy Policies", Journal of
Environmental Economics and Management, Vol 34 (September): 79-99.
Office of Air and Radiation, 1997, "The Benefits and Costs of the Clean Air Act: 1970 to 1990",
U.S. Environmental Protection Agency, October.
Viscusi, W. Kip, 1997, "The Value of Risks to Life and Health", Journal of Economic Literature,
Vol 31 (December): 1912-1946.
G-34
------- |